Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileAcademic lineage
Possible advisorsa guess from early papers, not confirmed
- Thomas S. HuangSuggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works466 from public data
- Graph Contrastive Learning with Augmentations3,051
The results show that, even without tuning augmentation extents nor using sophisticated GNN architectures, the GraphCL framework can produce graph representations of similar or better generalizability, transferrability, and robustness compared to state-of-the-art methods.
- EnlightenGAN: Deep Light Enhancement Without Paired Supervision2,636
This paper proposes a highly effective unsupervised generative adversarial network, dubbed EnlightenGAN, that can be trained without low/normal-light image pairs, yet proves to generalize very well on various real-world test images.
- 2,374
- Benchmarking Single-Image Dehazing and Beyond2,349
This work presents a new large-scale benchmark consisting of both synthetic and real-world hazy images, called REalistic Single-Image DEhazing (RESIDE), which highlights diverse data sources and image contents, and is divided into five subsets, each serving different training or evaluation purposes.
- DeblurGAN-v2: Deblurring (Orders-of-Magnitude) Faster and Better1,173
It is demonstrated that DeblurGAN-V2 has very competitive performance on several popular benchmarks, in terms of deblurring quality (both objective and subjective), as well as efficiency, and is effective for general image restoration tasks too.
- H$_2$O: Heavy-Hitter Oracle for Efficient Generative Inference of Large Language Models1,025
A novel approach for implementing the KV cache eviction policy that dynamically retains a balance of recent and H$_2$ tokens is introduced and a theoretical guarantee for the novel eviction algorithm is proved.
- Text2Video-Zero: Text-to-Image Diffusion Models are Zero-Shot Video Generators896
This paper proposes a low-cost approach (without any training or optimization) by leveraging the power of existing text-to-image synthesis methods, making them suitable for the video domain, and introduces a new task, zero-shot text-to-video generation.
- GaLore: Memory-Efficient LLM Training by Gradient Low-Rank Projection563
This work proposes GaLore, a training strategy that allows full-parameter learning but is more memory-efficient than common low-rank adaptation methods such as LoRA, and demonstrates the feasibility of pre-training a 7B model on consumer GPUs with 24GB memory without model parallel, checkpointing, or offloading strategies.
- 540
- Single Image Deraining: A Comprehensive Benchmark Analysis371
A comprehensive study and evaluation of existing single image deraining algorithms, using a new large-scale benchmark consisting of both synthetic and real-world rainy images, and provides a rich variety of criteria for dehazing algorithm evaluation.
- Forget-Me-Not: Learning to Forget in Text-to-Image Diffusion Models369
The Forget-Me-Not method is proposed, a model-centric and efficient solution designed to remove identities, objects, or styles from a well-configured text-to-image model in as little as 30 seconds, without significantly impairing its ability to generate other content.
- Advancing Image Understanding in Poor Visibility Environments: A Collective Benchmark Study351
The UG2+ challenge Track 2 competition in IEEE CVPR 2019 is launched, aiming to evoke a comprehensive discussion and exploration about whether and how low-level vision techniques can benefit the high-level automatic visual recognition in various scenarios.
- AutoGAN: Neural Architecture Search for Generative Adversarial Networks308
This paper presents the first preliminary study on introducing the NAS algorithm to generative adversarial networks (GANs), dubbed AutoGAN, and discovers architectures that achieve highly competitive performance compared to current state-of-the-art hand-crafted GANs.
- Adversarial Robustness: From Self-Supervised Pre-Training to Fine-Tuning301
This work introduces adversarial training into self-supervision, to provide general-purpose robust Pretrained models for the first time, and finds these robust pretrained models can benefit the subsequent fine-tuning in two ways: i) boosting final model robustness; ii) saving the computation cost, if proceeding towards adversarial fine- Tuning.
- Versatile Diffusion: Text, Images and Variations All in One Diffusion Model295
This work expands the existing single-flow diffusion pipeline into a multi-task multimodal network, dubbed Versatile Diffusion (VD), that handles multiple flows of text-to-image, image- to-text, and variations in one unified model.
- Drawing Early-Bird Tickets: Toward More Efficient Training of Deep Networks289
This paper discovers for the first time that the winning tickets can be identified at the very early training stage, which it is term as early-bird (EB) tickets, via low-cost training schemes at large learning rates, consistent with recently reported observations that the key connectivity patterns of neural networks emerge early.
- StreamingT2V: Consistent, Dynamic, and Extendable Long Video Generation from Text273
Experiments show that StreamingT2V produces more motion, while competing methods suffer from video stagnation when applied naively in an autoregressive fashion, surpassing competitors in both consistency and motion.
- Anti-Oversmoothing in Deep Vision Transformers via the Fourier Domain Analysis: From Theory to Practice249
This paper establishes a rigorous theory framework to analyze ViT features from the Fourier spectrum domain, and shows that the self-attention mechanism inherently amounts to a low-pass filter, which indicates when ViT scales up its depth, excessive low- pass filtering will cause feature maps to only preserve their Direct-Current component.
- When Image Denoising Meets High-Level Vision Tasks: A Deep Learning Approach248
This paper proposes a convolutional neural network for image denoising which achieves the state-of-the-art performance and proposes a deep neural network solution that cascades two modules for image Denoising and various high-level tasks, respectively, and uses the joint loss for updating only the denoise network via back-propagation.
- Self-Supervised Learning for Generalizable Out-of-Distribution Detection234
This work proposes a new technique relying on self-supervision for generalizable out-of-distribution (OOD) feature learning and rejecting those samples at the inference time, which does not need to pre-know the distribution of targeted OOD samples and incur no extra overheads compared to other methods.
- SinNeRF: Training Neural Radiance Fields on Complex Scenes from a Single Image233
It is shown that even without pre-training on multi-view datasets, SinNeRF can yield photo-realistic novel-view synthesis results and significantly outperforms the current state-of-the-art NeRF baselines in all cases.
- Patch Diffusion: Faster and More Data-Efficient Training of Diffusion Models207
Patch Diffusion is proposed, a generic patch-wise training framework, to significantly reduce the training time costs while improving data efficiency, which thus helps democratize diffusion model training to broader users.
- 202
- Connecting Image Denoising and High-Level Vision Tasks via Deep Learning200
A convolutional neural network in which convolutions are conducted in various spatial resolutions via downsampling and upsampling operations in order to fuse and exploit contextual information on different scales and the benefit of exploiting image semantics simultaneously for image denoising and high-level vision tasks via deep learning is demonstrated.
- Collaborative Global-Local Networks for Memory-Efficient Segmentation of Ultra-High Resolution Images189
This work proposes collaborative Global-Local Networks (GLNet) to effectively preserve both global and local information in a highly memory-efficient manner, and achieves much more competitive accuracy-memory usage trade-offs compared to state-of-the-arts.
- 182
- NeuralLift-360: Lifting an in-the-Wild 2D Photo to A 3D Object with 360° Views176
This work proposes a novel framework, dubbed NeuralLift-360, that utilizes a depth-aware neural radiance representation (NeRF) and learns to craft the scene guided by denoising diffusion models and can be guided with rough depth estimation in the wild.
- M$^3$ViT: Mixture-of-Experts Vision Transformer for Efficient Multi-task Learning with Model-Accelerator Co-design176
A model-accelerator co-design framework to enable efficient on-device MTL that customizes mixture-of-experts layers into a vision transformer (ViT) backbone for MTL, and sparsely activates task-specific experts during training, and is enhanced by hardware-level innovations.
- Towards Privacy-Preserving Visual Recognition via Adversarial Training: A Pilot Study176
This paper aims to improve privacy-preserving visual recognition, an increasingly demanded feature in smart camera applications, by formulating a unique adversarial training framework that explicitly learns a degradation transform for the original video inputs, in order to optimize the trade-off between target task performance and the associated privacy budgets on the degraded video.
- CamCo: Camera-Controllable 3D-Consistent Image-to-Video Generation175
This work introduces CamCo, which allows fine-grained Camera pose Control for image-to-video generation and significantly improves 3D consistency and camera control capabilities compared to previous models while effectively generating plausible object motion.
- Revisiting Zeroth-Order Optimization for Memory-Efficient LLM Fine-Tuning: A Benchmark163
This study unveils previously overlooked optimization principles, highlighting the importance of task alignment, the role of the forward gradient method, and the balance between algorithm complexity and fine-tuning performance, and introduces novel enhancements to ZO optimization, including block-wise descent, hybrid training, and gradient sparsity.
- 155
- Severe aortic stenosis detection by deep learning applied to echocardiography150
A novel deep learning model that relies on two-dimensional parasternal long axis videos from transthoracic echocardiography without Doppler imaging to identify severe AS, suitable for point-of-care ultrasonography is developed and externally validated.
- Federated Dynamic Sparse Training: Computing Less, Communicating Less, Yet Learning Better146
A novel FL framework by which complex neural networks can be deployed and trained with substantially improved efficiency in both on-device computation and in-network communication, and dynamic sparsity naturally introduces an "in-time self-ensembling effect'' into the training dynamics, and improves the FL performance even over dense training.
- 4DGen: Grounded 4D Content Generation with Spatial-temporal Consistency145
Compared to previous image-to-4D and text-to-4D works, 4DGen supports grounded generation, offering users enhanced control and improved motion generation capabilities, a feature difficult to achieve with previous methods.
- 145
- 144
- The Lottery Tickets Hypothesis for Supervised and Self-supervised Pre-training in Computer Vision Models142
This paper examines supervised and self-supervised pre-trained models through the lens of the lottery ticket hypothesis (LTH) and concludes that the core LTH observations remain generally relevant in the pre-training paradigm of computer vision, but more delicate discussions are needed in some cases.
- VisDrone-DET2018: The Vision Meets Drone Object Detection in Image Challenge Results139
A large-scale drone-based dataset, including 8, 599 images with rich annotations, including object bounding boxes, object categories, occlusion, truncation ratios, etc, is released, to narrow the gap between current object detection performance and the real-world requirements.
- DeepFont138
This work builds up the first available large-scale VFR dataset, named AdobeVFR, consisting of both labeled synthetic data and partially labeled real-world data, and introduces a Convolutional Neural Network decomposition approach and a novel learning-based model compression approach in order to reduce the DeepFont model size without sacrificing its performance.
- 137
- End-to-End United Video Dehazing and Detection130
An End-to-End Video Dehazing Network (EVD-Net) is proposed, which concatenates and jointly trains EVD-Net with a video object detection model and demonstrates much more stable and accurate detection results in hazy video.
- In-Context Learning Unlocked for Diffusion Models128
The resulting Prompt Diffusion model is the first diffusion-based vision-language foundation model capable of in- context learning and demonstrates high-quality in-context generation on the trained tasks and generalizes effectively to new, unseen vision tasks with their respective prompts.
- Unified Visual Transformer Compression126
This paper proposes a unified ViT compression framework that seamlessly assembles three effective techniques: pruning, layer skipping, and knowledge distillation, and formulate a budget-constrained, end-to-end optimization framework.
- EarlyBERT: Efficient BERT Training via Early-bird Lottery Tickets113
By slimming the self-attention and fully-connected sub-layers inside a transformer, this work is the first to identify structured winning tickets in the early stage of BERT training.
- Controllable Artistic Text Style Transfer via Shape-Matching GAN112
This paper presents the first text style transfer network that allows for real-time control of the crucial stylistic degree of the glyph through an adjustable parameter and proposes a novel bidirectional shape matching framework to establish an effective glyph-style mapping at various deformation levels without paired ground truth.
- RESIDE: A Benchmark for Single Image Dehazing.110
This paper presents a new large-scale benchmark consisting of both synthetic and real-world hazy images, called REalistic Single Image DEhazing (RESIDE), and provides a rich variety of criteria for dehazing algorithm evaluation, ranging from full-reference metrics, to no-reference metric, to subjective evaluation and the novel task-driven evaluation.
- SSH: A Self-Supervised Framework for Image Harmonization108
This work reformulate the image harmonization problem from a representation fusion perspective, which separately processes the foreground and background examples, to address the background occlusion issue and builds a real-world harmonization dataset as carefully created by expert users, for evaluation and benchmarking purposes.
- 106
- Graph Mixture of Experts: Learning on Large-Scale Graphs with Explicit Diversity Modeling105
The proposed Graph Mixture of Experts (GMoE) model empowers individual nodes in the graph to dynamically and adaptively select more general information aggregation experts, which are trained to capture distinct subgroups of graph structures and to incorporate information with varying hop sizes.
- DADA: Deep Adversarial Data Augmentation for Extremely Low Data Regime Classification102
A new discriminator loss is proposed to fit the goal of data augmentation, through which both real and augmented samples are enforced to contribute to and be consistent in finding the decision boundaries.
- Get More with LESS: Synthesizing Recurrence with KV Cache Compression for Efficient LLM Inference100
LESS, a simple integration of a (nearly free) constant sized cache with eviction-based cache methods, such that all tokens can be queried at later decoding steps, shows merit on a variety of tasks where it can help reduce the performance gap from caching everything.
- Found in the Middle: How Language Models Use Long Contexts Better via Plug-and-Play Positional Encoding100
Ms-PoE leverages the position indice rescaling to relieve the long-term decay effect introduced by RoPE, while meticulously assigning distinct scaling ratios to different attention heads to preserve essential knowledge learned during the pre-training step, forming a multi-scale context fusion from short to long distance.
- 98
- Augmentations in Hypergraph Contrastive Learning: Fabricated and Generative96
Among fabricated augmentations in HyperGCL, augmenting hyperedges provides the most numerical gains, implying that higher-order information in structures is usually more downstream-relevant and boosts robustness and fairness in hypergraph representation learning.
- Dual Dynamic Inference: Enabling More Efficient, Adaptive, and Controllable Deep Inference95
A dual dynamic inference (DDI) framework that integrates both input-dependent and resource-dependent dynamic inference mechanisms under a unified framework in order to fit the varying IoT resource requirements in practice and demonstrates the superior inference accuracy-resource trade-off achieved by DDI.
- 93
- Prompt-Free Diffusion: Taking “Text” Out of Text-to-Image Diffusion Models92
The proposed frame-work, Prompt-Free Diffusion, relies on only visual inputs to generate new images and outperform prior exemplar-based image synthesis approaches, and is naturally extensible to other downstream applications such as anime figure generation and virtual try-on, with promising quality.
- Auto-ViT-Acc: An FPGA-Aware Automatic Acceleration Framework for Vision Transformer with Mixed-Scheme Quantization92
This work proposes an FPGA-aware automatic ViT acceleration framework based on the proposed mixed-scheme quantization, to the best of the knowledge, this is the first FPGa-based ViTs acceleration framework exploring model quantization.
- 92
- Delving Into Robust Object Detection From Unmanned Aerial Vehicles: A Deep Nuisance Disentanglement Approach92
This work proposes to utilize free meta-data in conjunction with associated UAV images to learn domain-robust features via an adversarial training framework dubbed Nuisance Disentangled Feature Transform (NDFT), for the specific challenging problem of object detection in Uav images, achieving a substantial gain in robustness to those nuisances.
- Complete AI-Enabled Echocardiography Interpretation With Multitask Deep Learning90
An AI system that automatically interprets echocardiogram interpretation with multitask deep learning maintained high accuracy across geography and time from complete and limited studies.
- 90
- 89
- HD-Painter: High-Resolution and Prompt-Faithful Text-Guided Image Inpainting with Diffusion Models87
The Prompt-Aware Introverted Attention layer enhancing self-attention scores by prompt information resulting in better text aligned generations is designed and the Reweighting Attention Score Guidance mechanism seamlessly integrating a post-hoc sampling strategy into the general form of DDIM is introduced.
- GAN Slimming: All-in-One GAN Compression by a Unified Optimization Framework85
This work proposes the first unified optimization framework combining multiple compression means for GAN compression, dubbed GAN Slimming (GS), which seamlessly integrates three mainstream compression techniques: model distillation, channel pruning and quantization, together with the GAN minimax objective, into one unified optimization form that can be efficiently optimized from end to end.
- DP-OPT: Make Large Language Model Your Privacy-Preserving Prompt Engineer84
This paper introduces the first private prompt generation mechanism, by a differentially-private (DP) ensemble of in-context learning with private demonstrations, and demonstrates that prompts suggested by LLMs themselves can be transferred without compromising performance significantly.
- NeRF-SOS: Any-View Self-supervised Object Segmentation on Complex Scenes84
This paper carries out the exploration of self-supervised learning for object segmentation using NeRF for complex real-world scenes with extensive results on the LLFF, Tank&Temple, and BlendedMVS datasets validate the effectiveness of NeRF-SOS.
- Bag of Tricks for Training Deeper Graph Neural Networks: A Comprehensive Benchmark Study83
This work presents the first fair and reproducible benchmark dedicated to assessing the “tricks” of training deep GNNs, and demonstrates that an organic combo of initial connection, identity mapping, group and batch normalization attains the new state-of-the-art results for deep Gnns on large datasets.
- 83
- LLM-PBE: Assessing Data Privacy in Large Language Models82
LLM-PBE is a toolkit crafted specifically for the systematic evaluation of data privacy risks in LLMs, designed to analyze privacy across the entire lifecycle of LLMs, incorporating diverse attack and defense strategies, and handling various data types and metrics.
- Sparse MoE as the New Dropout: Scaling Dense and Self-Slimmable Transformers82
This work proposes a new plug-and-play training framework, SMoE-Dropout, to enable scaling transformers to better accuracy in their full capacity without collapse and explores the overlooked scalability bottleneck of SMoEs and leveraging it to effectively scale dense transformers.
- 82
- Decoding Compressed Trust: Scrutinizing the Trustworthiness of Efficient LLMs Under Compression81
This study conducts the first, thorough evaluation of three leading LLMs using five SoTA compression techniques across eight trustworthiness dimensions, finding that quantization is currently a more effective approach than pruning in achieving efficiency and trustworthiness simultaneously.
- Crowdsourced mapping of unexplored target space of kinase inhibitors80
A crowdsourced benchmarking of predictive algorithms for kinase inhibitor potencies across multiple kinase families tested on unpublished bioactivity data finds the top-performing predictions are based on various models, including kernel learning, gradient boosting and deep learning, and their ensemble leads to a predictive accuracy exceeding that of single-dose kinase activity assays.
- Compressing LLMs: The Truth is Rarely Pure and Never Simple79
Knowledge-Intensive Compressed LLM BenchmarK (LLM-KICK), a collection of carefully curated tasks to redefine the evaluation protocol for compressed LLMs, which have significant alignment with their dense counterparts and perplexity fail to capture subtle change in their true capabilities.
- Efficient Split-Mix Federated Learning for On-Demand and In-Situ Customization78
This paper proposes a novel Split-Mix FL strategy for heterogeneous participants that, once training is done, provides in-situ customization of model sizes and robustness and achieves customization with high efficiency in communication, storage, and inference.
- Bringing Your Own View78
The principled and automated approach has proven to be competitive against the state-of-the-art graph self-supervision methods, including GraphCL, on benchmarks of small graphs; and shown even better generalizability on large-scale graphs, without resorting to human expertise or downstream validation.
- Deep Plastic Surgery: Robust and Controllable Image Editing with Human-Drawn Sketches78
This paper proposes Deep Plastic Surgery, a novel, robust and controllable image editing framework that allows users to interactively edit images using hand-drawn sketch inputs and presents a sketch refinement strategy, as inspired by the coarse-to-fine drawing process of the artists, which can help the model well adapt to casual and varied sketches.
- Long-Tailed Classification of Thorax Diseases on Chest X-Ray: A New Benchmark Study73
A comprehensive benchmark study of the long-tailed learning problem in the specific domain of thorax diseases on chest X-rays is presented, focusing on learning from naturally distributed chest X-ray data, optimizing classification accuracy over not only the common “head” classes, but also the rare yet critical “tail” classes.
- Federated Adversarial Debiasing for Fair and Transferable Representations72
This work proposes a novel approach Federated Adversarial DEbiasing (FADE), which does not require users' sensitive group information for debiasing and offers users the freedom to opt-out from the adversarial component when privacy or computational costs become a concern.
- Aug-NeRF: Training Stronger Neural Radiance Fields with Triple-Level Physically-Grounded Augmentations71
Augmented NeRF (Aug-NeRF), which for the first time brings the power of robust data augmentations into regular-izing the NeRF training, and effectively boosts NeRF performance in both novel view synthesis and underlying geometry reconstruction.
- A Comprehensive Benchmark Analysis of Single Image Deraining: Current Challenges and Future Perspectives71
A comprehensive study and evaluation of existing single image deraining algorithms, using a new large-scale benchmark consisting of both synthetic and real-world rainy images of various rain types, which indicates the gap between the achievable performance on synthetic rainy images and the practical demand on real-world images.
- 71
- 71
- Edge-MoE: Memory-Efficient Multi-Task Vision Transformer Architecture with Task-Level Sparsity via Mixture-of-Experts69
Edge-MoE is proposed, the first end-to-end FPGA accelerator for multi-task ViT with a rich collection of architectural innovations and a novel patch reordering method to completely eliminate any memory access overhead.
- StegaNeRF: Embedding Invisible Information within Neural Radiance Fields69
StegaNeRF is introduced, an innovative approach for steganographic information embedding within NeRF renderings that enables precise retrieval of hidden information from images generated by NeRF, while ensuring the original visual quality of the rendered images to remain intact.
- 68
- Towards long-tailed, multi-label disease classification from chest X-ray: Overview of the CXR-LT challenge67
A large-scale benchmark dataset of over 350,000 CXRs, each labeled with at least one of 26 clinical findings following a long-tailed distribution is released, providing practical recommendations for long-tailed, multi-label medical image classification.
- 67
- 4K4DGen: Panoramic 4D Generation at 4K Resolution65
This work proposes a novel Panoramic Denoiser that adapts generic 2D diffusion priors to animate consistently in 360 images, transforming them into panoramic videos with dynamic scenes at targeted regions, and proposes Dynamic Panoramic Lifting to elevate the panoramic video into a 4D immersive environment while preserving spatial and temporal consistency.
- 65
- Learning Deep ℓ0 Encoders65
The proposed deep encoders enjoy faster inference, larger learning capacity, and better scalability compared to conventional sparse coding solutions, and under task-driven losses, the models can be conveniently optimized from end to end.
- A Multimodal Video-Based AI Biomarker for Aortic Stenosis Development and Progression64
A new AI-based video biomarker was independently associated with AS development and progression, enabling opportunistic risk stratification across cardiovascular imaging modalities as well as potential application on handheld devices.
- Zero-Shot Neural Architecture Search: Challenges, Solutions, and Opportunities64
This paper aims to comprehensively review and compare the state-of-the-art (SOTA) zero-shot NAS approaches, with an emphasis on their hardware awareness.
- AGG: Amortized Generative 3D Gaussians for Single Image to 3D64
An Amortized Generative 3D Gaussian framework (AGG) is introduced that instantly produces 3D Gaussians from a single image, eliminating the need for per-instance optimization.
- APOLLO: SGD-like Memory, AdamW-level Performance64
This work identifies that AdamW's learning rate adaptation rule can be effectively coarsened as a structured learning rate update, and proposes Approximated Gradient Scaling for Memory-Efficient LLM Optimization (APOLLO), which approximates learning rate scaling using an auxiliary low-rank optimizer state based on pure random projection.
- PAIR-Diffusion: A Comprehensive Multimodal Object-Level Image Editor64
This work proposes Structure-and-Appearance Paired Diffusion model ( PAIR-Diffusion ), which is trained using structure and appearance information explicitly extracted from the images, which enables users to inject a reference image’s appearance into the input input.
- Taxonomy of Machine Learning Safety: A Survey and Primer64
The taxonomy of ML safety presents a safety-oriented categorization of ML techniques to provide guidance for improving dependability of the ML design and development and can serve as a safety checklist to aid designers in improving coverage and diversity of safety strategies employed in any given ML system.
- 64
- 63
- Bridging the Gap Between Computational Photography and Visual Recognition63
Six new algorithms for image restoration or enhancement, created as part of the IARPA sponsored UG, are introduced, including a novel psychophysics-based evaluation regime for human assessment and a realistic set of quantitative measures for object recognition performance.
- 61
- The Emergence of Essential Sparsity in Large Pre-trained Models: The Weights that Matter58
This paper comprehensively studies induced sparse patterns across multiple large pre-trained vision and language transformers and proposes the existence of essential sparsity defined with a sharp dropping point beyond which the performance declines much faster w.r.t the rise of sparsity level.
- 58
- 58
- The Principle of Diversity: Training Stronger Vision Transformers Calls for Reducing All Levels of Redundancy56
This paper systematically studies the ubiquitous existence of re-dundancy in ViTs at all three levels: patch embedding, attention map, and weight space, and advocates a principle of diversity for training ViTs, by presenting cor-responding regularizers that encourage the representation diversity and coverage at each of those levels, that enabling capturing more discriminative information.
- 56
- Comp4D: LLM-Guided Compositional 4D Scene Generation54
Comp4D, a novel framework for Compositional 4D Generation that employs a compositional score distillation technique guided by the pre-defined trajectories, utilizing pre-trained diffusion models across text-to-image, text-to-video, and text-to-3D domains.
- AligNeRF: High-Fidelity Neural Radiance Fields via Alignment-Aware Training53
This work proposes the corresponding solutions to marry the multilayer perceptron (MLP) with convolutional layers which can encode more neighborhood information while reducing the total number of parameters, and proposes a novel training strategy to address misalignment caused by moving objects or small camera calibration errors.
- Removing Batch Normalization Boosts Adversarial Training53
This work proposes to completely eliminate this bottleneck by removing all BN layers in AT and achieves adversarial robustness with only a minor sacrifice on clean sample accuracy, and observes better model smoothness and larger decision margins from NoFrost, which make the models less sensitive to input perturbations and thus more robust.
- 53
- Taming Mode Collapse in Score Distillation for Text-to-3D Generation52
To tame mode collapse, the existing score distillation-based text-to-3D generation frameworks degenerate to maximal likelihood seeking on each view independently and thus suffer from the mode collapse problem, manifesting as the Janus artifact in practice.
- GDP: Stabilized Neural Network Pruning via Gates with Differentiable Polarization52
Experiments show that the proposed GDP algorithm achieves the state-of-the-art performance on various benchmark DNNs at a broad range of pruning ratios, and applies to DeepLabV3Plus-ResNet50 on the challenging Pascal VOC segmentation task, whose test performance sees no drop.
- 52
- 51
- MM3DGS SLAM: Multi-modal 3D Gaussian Splatting for SLAM Using Vision, Depth, and Inertial Measurements50
This work shows for the first time that using 3D Gaussians for map representation with unposed camera images and inertial measurements can enable accurate SLAM, and addresses the limitations of prior neural radiance field-based representations by enabling faster rendering, scale awareness, and improved trajectory tracking.
- Flextron: Many-in-One Flexible Large Language Model50
This paper presents a sample-efficient training method and associated routing algorithms for systematically transforming an existing trained LLM into a Flextron model, and evaluates Flextron on the GPT-3 and LLama-2 family of LLMs.
- Robust Mixture-of-Expert Training for Convolutional Neural Networks50
A new router-expert alternating Adversarial training framework for MoE is proposed, termed AdvMoE, which achieves 1% ~ 4% adversarial robustness improvement over the original dense CNN, and enjoys the efficiency merit of sparsity-gated MoE, leading to more than 50% inference cost reduction.
- 50
- Fractional Skipping: Towards Finer-Grained Dynamic CNN Inference49
More visualizations indicate a smooth and consistent transition in the DFS behaviors, especially the learned choices between layer skipping and different quantizations when the total computational budgets vary, validating the hypothesis that layer quantization could be viewed as intermediate variants of layer skipping.
- A Stitch in Time Saves Nine: Small VLM is a Precise Guidance for Accelerating Large VLMs47
This work employs the attention map aggregated from a small VLM to guide visual token pruning in a large VLM, and introduces a training-free method, Small VLM Guidance for accelerating Large VLMs (SGL), demonstrating the effectiveness and generalizability of SGL.
- 47
- Q-GaLore: Quantized GaLore with INT4 Projection and Layer-Adaptive Low-Rank Gradients46
Q-Galore is a novel approach that substantially reduces memory usage by combining quantization and low-rank projection, surpassing the benefits of GaLore, and achieves highly competitive performance with exceptional memory efficiency.
- “BNN - BN = ?”: Training Binary Neural Networks without Batch Normalization46
This work demonstrates for the first time that BNs can be completely removed from BNN training and inference regimes, and by plugging in and customizing techniques including adaptive gradient clipping, scale weight standardization, and specialized bottleneck block, a BN-free BNN is capable of maintaining competitive accuracy compared to its Bn-based counterpart.
- SmartExchange: Trading Higher-cost Memory Storage/Access for Lower-cost Computation46
A novel algorithm to enforce a specially favorable DNN weight structure, where each layerwise weight matrix can be stored as the product of a small basis matrix and a large sparse coefficient matrix whose non-zero elements are all power-of-2.
- Artificial intelligence-guided detection of under-recognised cardiomyopathies on point-of-care cardiac ultrasonography: a multicentre study44
An AI framework that enables scalable, opportunistic screening of under-recognised cardiomyopathies through simple POCUS acquisitions is developed and validated.
- Physics-Driven Turbulence Image Restoration with Stochastic Refinement44
The Physics-integrated Restoration Network (PiRN) is proposed to bring the physics-based simulator directly into the training process to help the network to disentangle the stochasticity from the degradation and the underlying image.
- 44
- Training Your Sparse Neural Network Better with Any Mask43
This paper demonstrates an alternative opportunity: one can carefully customize the sparse training techniques to deviate from the default dense network training protocols, consisting of introducing ``ghost"neurons and skip connections at the early stage of training, and strategically modifying the initialization as well as labels.
- Enhancing NeRF akin to Enhancing LLMs: Generalizable NeRF Transformer with Mixture-of-View-Experts41
This proposed model, dubbed GNT with Mixture-of-View-Experts (GNT-MOVE), has experimentally shown state-of-the-art results when transferring to unseen scenes, indicating remarkably better cross-scene generalization in both zero-shot and few-shot settings.
- 41
- AutoShot: A Short Video Dataset and State-of-the-Art Shot Boundary Detection40
This work releases a new public Short video sHot bOundary deTection dataset, named SHOT, consisting of 853 complete short videos and 11,606 shot annotations, with 2,716 high quality shot boundary annotations in 200 test videos, and proposes to optimize the model design for video SBD by conducting neural architecture search in a search space encapsulating various advanced 3D ConvNets and Transformers.
- Outline, Then Details: Syntactically Guided Coarse-To-Fine Code Generation40
This work proposes ChainCoder, a program synthesis language model that generates Python code progressively, i.e. from coarse to fine in multiple passes, demonstrating that the progressive generation eases the reasoning procedure and guides the language model to generate higher-quality solutions.
- MMG-Ego4D: Multi-Modal Generalization in Egocentric Action Recognition39
This paper evaluates a diverse array of models on MMG-Ego4D and proposes new methods with improved generalization ability, including a new fusion module with modality dropout training, contrastive-based alignment training, and a novel cross-modal prototypical loss for better few-shot performance.
- Improving model fairness in image-based computer-aided diagnosis38
This work proposes an algorithm that leverages the marginal pairwise equal opportunity to reduce bias in medical image classification and demonstrates that the proposed algorithm not only improves fairness in individual and intersectional subgroups but also maintains overall performance.
- 38
- 38
- 38
- Harnessing the power of longitudinal medical imaging for eye disease prognosis using Transformer-based sequence modeling37
Using longitudinal imaging data from the Age-Related Eye Disease Study (AREDS) and Ocular Hypertension Treatment Study (OHTS), LTSA significantly outperformed a single-image baseline in 19/20 head-to-head comparisons on late AMD prognosis and 18/20 comparisons on POAG prognosis.
- Ten Lessons We Have Learned in the New "Sparseland": A Short Handbook for Sparse Neural Network Researchers37
This article attempts to summarize some most common confusions in SNNs, that one may come across in various scenarios such as paper review/rebuttal and talks - many drawn from the authors' own bittersweet experiences!
- AdaDeep: A Usage-Driven, Automated Deep Model Compression Framework for Enabling Ubiquitous Intelligent Mobiles37
AdaDeep is proposed, a usage-driven, automated DNN compression framework for systematically exploring the desired trade-off between performance and resource constraints, from a holistic system level.
- A Simple Single-Scale Vision Transformer for Object Detection and Instance Segmentation36
A simple and compact ViT architecture called Universal Vision Transformer (UViT) is proposed that achieves strong performance on COCO object detection and instance segmentation tasks and completes a scaling rule to optimize the model's trade-off on accuracy and computation cost / model size.
- 36
- FarSight: A Physics-Driven Whole-Body Biometric System at Large Distance and Altitude35
The end-to-end design, development and evaluation of FarSight, an innovative software system designed for whole-body biometric recognition that combines the physics of imaging and deep learning models to enhance image restoration and biometric feature encoding are presented.
- 35
- Sparsity May Cry: Let Us Fail (Current) Sparse Neural Networks Together!35
A systemic evaluation of the most representative sparse algorithms reveals an important obscured observation: the state-of-the-art magnitude- and/or gradient-based sparse algorithms seemingly fail to perform on SMC-Bench when applied out of the box, sometimes at significantly trivial sparsity as low as 5%.
- Point Cloud Domain Adaptation via Masked Local 3D Structure Prediction35
This paper proposes a Masked Local Structure Prediction method to encode target data and enables models to embed source and target data in a shared feature space through estimating point cardinality, position and normal.
- Learning Simple Thresholded Features With Sparse Support Recovery35
TheDLTF model, which learns an optimized dictionary for applying the thresholded feature, is formulated and evaluated on a vast range of synthetic and real-data tasks, where DLTF demonstrates remarkable efficiency, effectiveness, and robustness in all experiments.
- Efficient deep learning-based automated diagnosis from echocardiography with contrastive self-supervised learning34
EchoCLR is unique in its ability to learn representations of echocardiogram videos and demonstrates that SSL can enable label-efficient disease classification from small amounts of labeled data.
- 34
- Learning to Estimate 6DoF Pose from Limited Data: A Few-Shot, Generalizable Approach using RGB Images33
A new cascade framework named Cas6D for few-shot 6DoF pose estimation that is generalizable and uses only RGB images, which can overcome the large gap between pose candidates and ground truth poses, which is a common failure mode in sparse-view scenarios.
- Safeguarded Learned Convex Optimization33
This work presents a Safe-L2O framework, which includes a safeguard to guarantee convergence for convex problems with proximal and/or gradient oracles, and yields the numerical benefits of employing machine learning to create rapid L2O algorithms while still guaranteeing convergence.
- 33
- Sparse Coding and its Applications in Computer Vision33
This book provides a broader introduction to the theories and applications of sparse coding techniques in computer vision research, and introduces sparse coding in the context of representation learning, illustrates the fundamental concepts, and summarizes the most active research directions.
- On The Planning Abilities of OpenAI's o1 Models: Feasibility, Optimality, and Generalizability32
This pilot study provides foundational insights into the planning limitations of LLMs, offering key directions for future research on improving memory management, decision-making, and generalization in LLM-based planning.
- Biometric contrastive learning for data-efficient deep learning from electrocardiographic images31
A pretraining strategy that leverages biometric signatures of different ECGs from the same patient enhances the efficiency of developing AI models for ECG images, representing a major advance in detecting disorders from ECG images with limited labeled data.
- Federated Robustness Propagation: Sharing Adversarial Robustness in Heterogeneous Federated Learning31
This paper studies a novel FL strategy: propagating adversarial robustness from rich-resource users that can afford AT, to those with poor resources that cannot afford it, during federated learning and proposes an efficient propagation approach by the proper use of batch-normalization.
- 31
- Quarantine: Sparsity Can Uncover the Trojan Attack Trigger for Free30
This work proposes a novel Trojan network detection regime: first locating a “winning Trojan lottery ticket” which preserves nearly full Trojan information yet only chance-level performance on clean inputs; then recovering the trigger embedded in this already isolated sub-network.
- Troubleshooting Blind Image Quality Models in the Wild30
Inspired by recent findings that difficult samples of deep models may be exposed through network pruning, a set of "self-competitors" is constructed, as random ensembles of pruned versions of the target model to be improved.
- 30
- Robust emotion recognition from low quality and low bit rate video: A deep learning approach30
A novel framework to achieve robust emotion recognition from low bit rate video with significantly improved stand-alone recognition performance, as well as rate-distortion (R-D) performance, than either directly recognizing from LR frames, or separating SR and recognition.
- Dynamic Sparsity Is Channel-Level Sparsity Learner29
Channel-aware dynamic sparse (Chase) is proposed, which for the first time seamlessly translates the promise of unstructured dynamic sparsity to GPU-friendly channel-level sparsity during one end-to-end training process, without any ad-hoc operations.
- 28
- R-Sparse: Rank-Aware Activation Sparsity for Efficient LLM Inference27
R-Sparse is introduced, a training-free activation sparsity approach capable of achieving high sparsity levels in advanced LLMs that replaces the linear layers in LLMs with a rank-aware sparse inference method that leverages the sparsity of input channels and singular value components, eliminating the need for active channel prediction like the output sparsity based approaches.
- Shake to Leak: Fine-tuning Diffusion Models Can Amplify the Generative Privacy Risk27
A new risk, Shake-to-Leak (S2L), is revealed that fine-tuning the pre-trained models with manipulated data can amplify the existing privacy risks, and underscores that the privacy risk with diffusion models is even more severe than previously recognized.
- 27
- 27
- HALO: Hardware-Aware Learning to Optimize27
This paper proposes hardware-aware learning to optimize (HALO) , a practical meta optimizer dedicated to resource-efficient on-device adaptation and features the following highlights: faster adaptation speed, lower per-iteration complexity, and a stochastic structural sparsity regularizer being enforced.
- 27
- 26
- Cavia: Camera-controllable Multi-view Video Diffusion with View-Integrated Attention26
Cavia is a novel framework for camera-controllable, multi-view video generation, capable of converting an input image into multiple spatiotemporally consistent videos, and the first of its kind that allows the user to precisely specify camera motion while obtaining object motion.
- Instant Soup: Cheap Pruning Ensembles in A Single Pass Can Draw Lottery Tickets from Large Models26
The authors' extensive experiments and ablation on two popular large-scale pre-trained models: CLIP and BERT across multiple benchmark vision and language datasets validate the effectiveness of ISP compared to several state-of-the-art pruning methods.
- Radiomics-Guided Global-Local Transformer for Weakly Supervised Pathology Localization in Chest X-Rays26
A Radiomics-Guided Transformer (RGT) that fuses global image information with local radiomics-guided auxiliary information to provide accurate cardiopulmonary pathology localization and classification without any bounding box annotations is proposed.
- 26
- Efficient Image Generation with Variadic Attention Heads25
This work proposes a simple, yet powerful method to reduce trade-offs: allow the attention heads of a single transformer to attend to multiple receptive fields in a Neighborhood Attention architecture for image generation.
- U-Finger: Multi-Scale Dilated Convolutional Network for Fingerprint Image Denoising and Inpainting25
It is shown that the usage of dilated convolutions as well as the removal of padding have important positive impacts on the final restoration performance, in addition to multi-scale cascaded feature modules.
- 25
- Understanding and Accelerating Neural Architecture Search With Training-Free and Theory-Grounded Metrics24
This work designs a principled and unified training-free framework for Neural Architecture Search (NAS), with high performance, low cost, and in-depth interpretation, and presents a unified framework to understand and accelerate NAS by disentangling “TEG” characteristics of searched networks – Trainability, Expressivity, Generalization – all assessed in a training-free manner.
- SteinDreamer: Variance Reduction for Text-to-3D Score Distillation via Stein Identity24
The results suggest that SSD can effectively reduce the distillation variance and consistently improve visual quality for both object- and scene-level generation, and it is demonstrated that SteinDreamer achieves faster convergence than existing methods due to more stable gradient updates.
- 24
- LLM-AutoDiff: Auto-Differentiate Any LLM Workflow23
LLM-AutoDiff is introduced: a novel framework for Automatic Prompt Engineering (APE) that extends textual gradient-based methods (such as Text-Grad) to multi-component, potentially cyclic LLM architectures and unifying prompt optimization through a graph-centric lens.
- Exposing Semantic Segmentation Failures via Maximum Discrepancy Competition23
This paper exposes failures of existing semantic segmentation methods in the open visual world under the constraint of very limited human labeling effort, and conducts a thorough MAD diagnosis of ten PASCAL VOC semantic segmentsation algorithms.
- Finding Fantastic Experts in MoEs: A Unified Study for Expert Dropping Strategies and Observations22
This work proposes MoE Experts Compression Suite (MC-Suite), which is a collection of some previously explored and multiple novel recipes to provide a comprehensive benchmark for estimating expert importance from diverse perspectives, as well as unveil numerous valuable insights for SMoE experts.
- Data Distillation Can Be Like Vodka: Distilling More Times For Better Quality22
Progressive Dataset Distillation synthesizes multiple small sets of synthetic images, each conditioned on the previous sets, and trains the model on the cumulative union of these subsets without requiring additional training time, which can effectively improve the performance of existing dataset distillation methods.
- You Can Have Better Graph Neural Networks by Not Training Weights at All: Finding Untrained GNNs Tickets22
The first-of-its-kind exploration of discovering matching untrained GNNs with sparsity as the core tool is carried out and it is shown that the found untrained subnetworks can substantially mitigate the GNN over-smoothing problem, hence becoming a powerful tool to enable deeper GNN's without bells and whistles.
- Symbolic Learning to Optimize: Towards Interpretability and Scalability22
This paper proves the concept that "kill two birds by one stone", by introducing the powerful tool of symbolic regression to L2O, and establishes a holistic symbolic representation and analysis framework for L1O, which yields a series of insights for learnable optimizers.
- 22
- A Roadmap for Automatic Surgical Site Infection Detection and Evaluation Using User-Generated Incision Images22
Interactive image acquisition as well as customized image analysis and machine learning methods for SSI monitoring will play critical roles in developing sustainable mHealth apps to achieve the expected outcomes of patient-taken incision images for effective out-of-clinic patient-centered healthcare with substantially reduced cost.
- Predicting Depression Severity by Multi-Modal Feature Engineering and Fusion22
A multi-modal fusion model that combines three different modalities: audio, video, and text features is proposed that outperforms each single-modality prediction model, and surpasses the dataset baseline with a nice margin.
- VASE: Object-Centric Appearance and Shape Manipulation of Real Videos21
This work introduces a framework that is object-centric and is designed to control both the object's appearance and, notably, to execute precise and explicit structural modifications on the object.
- Are Large Kernels Better Teachers than Transformers for ConvNets?21
A first-of-its-kind study unveiling that modern large-kernel ConvNets, a compelling competitor to Vision Transformers, are remarkably more effective teachers for small-kernel convolutional Neural Networks, due to more similar architectures.
- CERL: A Unified Optimization Framework for Light Enhancement With Realistic Noise21
This work presents Coordinated Enhancement for Real-world Low-light Noisy Images (CERL), that seamlessly integrates light enhancement and noise suppression parts into a unified and physics-grounded optimization framework.
- Scalable Perception-Action-Communication Loops With Convolutional and Graph Neural Networks21
Through a multi-agent flocking application, it is demonstrated that VGAI yields performance comparable to or better than other decentralized controllers, using only the visual input modality and without accessing precise location or motion state information.
- SAFIN: Arbitrary Style Transfer with Self-Attentive Factorized Instance Normalization21
SAIN is essentially a spatially-adaptive normalization module whose parameters are inferred through attention on the content and style image and it is demonstrated that plugging SAFIN into the base network of another state-of-the-art method results in enhanced stylization.
- EnergyNet: Energy-Efficient Dynamic Inference21
A CNN for energy-aware dynamic routing, called EnergyNet, is proposed that achieves adaptive-complexity inference based on the inputs, leading to an overall reduction of run time energy cost while actually improving accuracy.
- 20
- PanEcho: Complete AI-enabled echocardiography interpretation with multi-task deep learning20
PanEcho is a uniquely view-agnostic, multi-task, open-source model that enables state-of-the-art echocardiographic interpretation across complete and limited studies, serving as an efficient echocardiographic foundation model.
- Saccharide mapping as an extraordinary method on characterization and identification of plant and fungi polysaccharides: A review20
This review filled the blank of basic information about saccharide mapping and was helpful for the establishment of a professional workflow for the saccharide mapping application to promote the deep study of polysaccharide structure.
- A Multi-Purpose Realistic Haze Benchmark With Quantifiable Haze Levels and Ground Truth20
This paper introduces the first realistic haze image benchmark, from both aerial and ground view, with paired haze-free images, and in-situ haze density measurements, and evaluates a set of representative state-of-the-art dehazing approaches as well as object detectors on the dataset.
- Knowledge-Augmented Contrastive Learning for Abnormality Classification and Localization in Chest X-rays with Radiomics using a Feedback Loop20
An end-to-end semi-supervised knowledge-augmented contrastive learning framework, that simultaneously performs disease classification and localization tasks, and constitutes a feedback loop for image and radiomic features to mutually reinforce each other.
- Linearity Grafting: Relaxed Neuron Pruning Helps Certifiable Robustness20
This work proposes a novel solution to strategically manipulate neurons, by "grafting" appropriate levels of linearity, to eliminate the non-linear components that are both redundant for DNN performance and harmful to its certification.
- Fast and High Quality Image Denoising via Malleable Convolution20
This work presents Malleable Convolution (MalleConv), which performs spatial-varying processing with minimal computational overhead, and uses it to build an efficient denoising network the authors call MalleNet, which achieves high-quality results without very deep architectures.
- Scaling Multimodal Pre-Training via Cross-Modality Gradient Harmonization20
This paper proposes to harmonize cross-modality gradient realignment, via two techniques: modifying different CMA loss gradients for each sample triplet, so that their gradient directions are more aligned, and gradient-based curriculum learning: leveraging the gradient conflict information on an indicator of sample noisiness, to develop a curriculum learning strategy to prioritize training on less noisy sample triplets.
- Steepest Descent Density Control for Compact 3D Gaussian Splatting19
SteepGS is introduced, incorporating steepest density control, a principled strategy that minimizes loss while maintaining a compact point cloud and achieves a ~ 50% reduction in Gaussian points without compromising rendering quality, significantly enhancing both efficiency and scalability.
- 19
- 19
- Enhancing Item Tokenization for Generative Recommendation through Self-Improvement18
This work proposes a self-improving item tokenization method that allows the LLM to refine its own item tokenizations during training process and is simple to implement and can be integrated as a plug-and-play enhancement into existing generative recommendation systems.
- Lift3D: Zero-Shot Lifting of Any 2D Vision Model to 3D18
The question of whether any 2D vision model can be lifted to make 3D consistent predictions is asked and answered in the affirmative; the new Lift3D method trains to predict unseen views onfeature spaces generated by afew visual models (i.e. DINO and CLIP), but then generalizes to novel vision operators and tasks.
- 18
- 18
- Symbolic Visual Reinforcement Learning: A Scalable Framework With Object-Level Abstraction and Differentiable Expression Search17
This work proposes Differentiable Symbolic Expression Search (DiffSES), a novel symbolic learning approach that discovers discrete symbolic policies using partially differentiable optimization that is able to leverage the simplicity and scalability advantages of symbolic expressions, while also incorporating the strengths of neural networks for feature learning and optimization.
- Dynamic Privacy Budget Allocation Improves Data Efficiency of Differentially Private Gradient Descent17
This talk introduces dynamic techniques for learning rate, batch size, noise magnitude and gradient clipping, and discusses how the dynamic policy could change the convergence bounds which further provides insight of the impact of dynamic methods.
- Can pruning improve certified robustness of neural networks?17
It is demonstrated for the first time that pruning can generally improve certified robustness for ReLU-based NNs under the complete verification setting and a new stability-based pruning method tailored for reducing neuron instability is presented, that outperforms existing pruning methods in enhancing certify robustness.
- Take the Bull by the Horns: Hard Sample-Reweighted Continual Training Improves LLM Generalization16
Through rigorous experimentation with various models and datasets, the findings indicate that the sample-targeted methods significantly improve LLM performance across multiple benchmarks, in both continual pre-training and instruction tuning scenarios.
- Search Behavior Prediction: A Hypergraph Perspective16
A Dual-Channel Attention-Based Hypergraph Neural Network (DCAH), which synergizes information from two potentially noisy sources ( original query-item edges and item-item hyperedges) and enhances their link prediction performance, is developed.
- Robust Weight Signatures: Gaining Robustness as Easy as Patching Weights?16
A minimalistic model robustness"patching" framework that carries a model trained on clean data together with its pre-extracted RWSs that is remarkably lightweight, transferable and cross-dataset transferable.
- 16
- MM-Hand16
This work proposes a 3D-aware multi-modal guided hand generative network (MM-Hand), together with a novel geometry-based curriculum learning strategy that can consistently improve the quantitative performance of the state-of-the-art 3D hand pose estimators on two benchmark datasets.
- Focus Longer to See Better: Recursively Refined Attention for Fine-Grained Image Classification16
This paper shows through interpretability techniques how the network focus changes from coarse to fine details, and shows that a simple attention model can aggregate (weighted) these finer details to focus on the most dominant discriminative part of the image.
- Cone-Beam Computed Tomography (CBCT) Segmentation by Adversarial Learning Domain Adaptation16
The experiments on the bladder images from Radiation Oncology clinics have shown that the CBCT segmentation with adversarial learning domain adaptation significantly improves segmentation accuracy compared to the existing methods without doing domain adaptation from CT to CBCT.
- Artificial Intelligence Methods for Surgical Site Infection: Impacts on Detection, Monitoring, and Decision Making16
Artificial intelligence has the potential to provide insight in detecting and decision-support of SSI and increase the possibility of improving surgical practice with the promise of a future optimized for the highest quality patient care.
- 16
- Know Where You're Uncertain When Planning with Multimodal Foundation Models: A Formal Framework15
This work introduces a framework for uncertainty disentanglement, isolating perception uncertainty arising from limitations in visual understanding and decision uncertainty relating to the robustness of generated plans, and proposes methods tailored to the unique properties of perception and decision-making.
- Grounded-Instruct-Pix2Pix: Improving Instruction Based Image Editing with Automatic Target Grounding15
- Neural Implicit Dictionary via Mixture-of-Expert Training15
A generic INR framework that achieves both data and training efficiency by learning a Neural Implicit Dictionary (NID) from a data collection and representing INR as a functional combination of basis sampled from the dictionary is presented.
- 15
- 15
- DD-Ranking: Rethinking the Evaluation of Dataset Distillation14
DD-Ranking, a unified evaluation framework, along with new general evaluation metrics to uncover the true performance improvements achieved by different methods are proposed, which provide a more comprehensive and fair evaluation standard for future research advancements.
- 14
- Evaluate underdiagnosis and overdiagnosis bias of deep learning model on primary open-angle glaucoma diagnosis in under-served patient populations14
The results show that the widely-used deep learning model can underdiagnose or overdiagnose under-served populations, and the most overdiagnosed group is Black older (≥ 60 yrs) group, which is raising ethical concerns about using deep learning models in ophthalmology clinics.
- How Does Pruning Impact Long-Tailed Multi-label Medical Image Classifiers?14
This work performs the first analysis of pruning’s effect on neural networks trained to diagnose thorax diseases from chest X-rays (CXRs) on two large CXR datasets, and finds that radiologists perceive PIEs as having more label noise, lower image quality, and higher diagnosis difficulty.
- 14
- How Robust is Your Fairness? Evaluating and Sustaining Fairness under Unseen Distribution Shifts14
A novel fairness learning method termed CUrvature MAtching (CUMA), which can achieve robust fairness generalizable to unseen domains with unknown distributional shifts and enforces the model to have similar generalization ability on the majority and minority groups.
- Debiased Subjective Assessment of Real-World Image Enhancement14
This work takes steps towards debiasing conventional subjective assessment by automatically sampling a set of adaptive and diverse images for subsequent testing, and casting sample selection into a joint maximization of the discrepancy between the enhancers and the diversity among the selected input images.
- 14
- 14
- 14
- 14
- An Overview of Low-Rank Structures in the Training and Adaptation of Large Models: From Implicit Low-Dimensionality to Efficient Training and Fine-Tuning [Special Issue on the Mathematics of Deep Learning]13
Key insights are highlighted from a rich line of research focused on theoretically understanding and leveraging low-rank structures in deep learning both at every iteration of training as well as at the global minimum.
- Can Test-Time Scaling Improve World Foundation Model?13
This work introduces SWIFT, a test-time scaling framework tailored for WFMs that integrates the authors' extensible WFM evaluation toolkit with process-level inference strategies, including fast tokenization, probability-based Top-K pruning, and efficient beam search.
- FlexGS: Train Once, Deploy Everywhere with Many-in-One Flexible 3D Gaussian Splatting13
This work introduces a tiny learnable module that controls Gaussian selection based on the input percentage, along with a transformation module that adjusts the selected Gaussians to complement the performance of the reduced model.
- Scalable Learning to Optimize: A Learned Optimizer Can Train Big Models13
This work proposes a new scalable learning to optimize ( SL2O ) framework which first constrains the network updates in a tiny subspace and then explores learning rules on top of it, showing that the scalability roadblock of applying L2O to training large models is now removed.
- RoS-KD: A Robust Stochastic Knowledge Distillation Approach for Noisy Medical Imaging13
A Robust Stochastic Knowledge Distillation (RoS-KD) framework which mimics the notion of learning a topic from multiple sources to ensure deterrence in learning noisy information and its robustness to adversarial attacks is proposed.
- 13
- 13
- Frank-Wolfe Network: An Interpretable Deep Structure for Non-Sparse Coding13
An interpretable deep structure namely Frank- Wolfe Network (F-W Net), whose architecture is inspired by unrolling and truncating the Frank-Wolfe algorithm for solving the non-sparse coding problem, is proposed and the hyper-parameter can be made learnable instead of pre-chosen in F-WNet.
- 13
- Learning to Generalize Provably in Learning to Optimize12
This paper theoretically establish an implicit connection between the local entropy and the Hessian and hence unify their roles in the handcrafted design of generalizable optimizers as equivalent metrics of the landscape flatness of loss functions and proposes to incorporate these two metrics as flatness-aware regularizers into the L2O framework in order to meta-train optimizers to learn to generalize.
- Efficient-3DiM: Learning a Generalizable Single-image Novel-view Synthesizer in One Day12
Efficient-3DiM, a simple but effective framework to learn a single-image novel-view synthesizer, and several pragmatic strategies to reduce the training overhead to a manageable scale, including a crafted timestep sampling strategy, a superior 3D feature extractor, and an enhanced training scheme.
- 12
- Data Augmentation for Improving Deep Learning Models in Building Inspections or Postdisaster Evaluation12
This work states that current building evaluations, whether for occupant safety or insurance appraisal, are conducted primarily via visual inspections performed by certified individuals, and that these inspections, which...
- 12
- 12
- L p -Norm Constrained Coding With Frank-Wolfe Network.12
An interpretable deep structure namely Frank- Wolfe Network (F-W Net), whose architecture is inspired by unrolling and truncating the Frank-Wolfe algorithm for solving the non-sparse coding problem, is proposed and the hyper-parameter can be made learnable instead of pre-chosen in F-WNet.
- VideoLifter: Lifting Videos to 3D with Fast Hierarchical Stereo Alignment11
VideoLifter is introduced, a novel video-to-3D pipeline that leverages a local-to-global strategy on a fragment basis, achieving both extreme efficiency and SOTA quality and significantly accelerates the reconstruction process.
- CXR-LT 2024: A MICCAI challenge on long-tailed, multi-label, and zero-shot disease classification from chest X-ray11
An overview of CXR-LT 2024 is provided, detailing the data curation process and consolidating state-of-the-art solutions, including the use of multimodal models for rare disease detection, advanced generative approaches to handle noisy labels, and zero-shot learning strategies for unseen diseases.
- 11
- Model-GLUE: Democratized LLM Scaling for A Large Model Zoo in the Wild11
This work starts with a benchmarking of existing LLM scaling techniques, especially selective merging, and variants of mixture, and forms an optimal strategy for the selection and aggregation of a heterogeneous model zoo characterizing different architectures and initialization.
- DSEE: Dually Sparsity-embedded Efficient Tuning of Pre-trained Language Models11
This work proposes a framework for resource- and parameter-efficient fine-tuning by leveraging the sparsity prior in both weight updates and the final model weights, and leverages sparsity in these two directions by exploiting both unstructured and structured sparse patterns in pre-trained language models viaa unified approach.
- 11
- Can We Solve 3D Vision Tasks Starting from A 2D Vision Transformer?11
To build a 3D ViT from its 2D sibling, the patch embedding and token sequence is inflated, accompanied with new positional encoding mechanisms designed to match the 3D data geometry, and the resultant, named Simple3D-Former, performs surprisingly robustly on popular 3D tasks such as object classification, point cloud segmentation and indoor scene detection.
- Sandwich Batch Normalization: A Drop-In Replacement for Feature Distribution Heterogeneity11
This work demonstrates the prevailing effectiveness of SaBN as a drop-in replacement in four tasks: conditional image generation, neural architecture search, adversarial training, and arbitrary style transfer, and provides visualizations and analysis to help understand why SaBN works.
- 11
- Learning Model-Based Privacy Protection under Budget Constraints11
A novel Learning-to-Protect algorithm that automatically learns a model-based protector from a set of non-private learning tasks that can achieve a superior utility with a given privacy constraint and generalize well to new private datasets distributed differently as compared to the hand-designed competitors.
- 11
- 11
- 11
- What Does CNN Shift Invariance Look Like? A Visualization Study11
It is concluded that features extracted from popular networks are not globally invariant, and that biases and artifacts exist within this variance, which means that anti-aliased models significantly improve local invariance but do not impact global invariance.
- 10
- Polynomial Width is Sufficient for Set Representation with High-dimensional Features10
It is demonstrated that $L$ being poly$(N, D)$ is sufficient for set representation using both embedding layers, and a lower bound of$L$ is provided for the LP embedding layer.
- ReCoRo: Re gion- Co ntrollable Ro bust Light Enhancement with User-Specified Imprecise Masks10
Region-Controllable Robust Light Enhancement (ReCoRo), a novel framework that allows users to directly specify "where" and "how much" they want to enhance from an input low-light image; meanwhile, the model will learn to intelligently maintain the overall consistent visual appearance and plausible composition via a discriminator.
- MATE: Plugging in Model Awareness to Task Embedding for Meta Learning10
A novel task representation called model-aware task embedding (MATE) is proposed that incorporates not only the data distributions of different tasks, but also the complexity of the tasks through the models used.
- An Al-based Spatial Knowledge Graph for Enhancing Spatial Data and Knowledge Search and Discovery10
A novel AI-based graph embedding algorithm to build semantic relationships between different spatial data sets to enable efficient and accurate data search and visualize the knowledge graph using the Neo4j database graphical user interface is developed.
- Searching for Two-Stream Models in Multivariate Space for Video Recognition10
This work designs a multivariate search space, including 6 search variables to capture a wide variety of choices in designing two-stream models, and proposes a progressive search procedure, by searching for the architecture of individual streams, fusion blocks and attention blocks one after the other.
- 10
- 10
- 10
- DeepFont10
This proposed technical demonstration will give an audience a tour to the DeepFont system, with the focus on its impacts on real consumer products, including but not limited to: 1) a cloud-based iOS App for font recognition; 2) a web-based tool for font similarity evaluation and discovery.
- 9
- VLN-Zero: Rapid Exploration and Cache-Enabled Neurosymbolic Vision-Language Planning for Zero-Shot Transfer in Robot Navigation9
VLN-Zero is presented, a two-phase vision-language navigation framework that leverages vision-language models to efficiently construct symbolic scene graphs and enable zero-shot neurosymbolic navigation, enabling robust and scalable decision-making in unseen environments.
- HALoS: Hierarchical Asynchronous Local SGD over Slow Networks for Geo-Distributed Large Language Model Training9
HALoS is a hierarchical asynchronous optimization framework that tackles issues by introducing local parameter servers within each region and a global parameter server that merges updates across regions and improves upon existing asynchronous methods by up to 2.1x.
- E3D-Bench: A Benchmark for End-to-End 3D Geometric Foundation Models9
The first comprehensive benchmark for 3D GFMs is presented, covering five core tasks: sparse-view depth estimation, video depth estimation, 3D reconstruction, multi-view pose estimation, novel view synthesis, and spanning both standard and challenging out-of-distribution datasets.
- 9
- Social Reward: Evaluating and Enhancing Generative AI through Million-User Feedback from an Online Creative Community9
This work pioneers a paradigm shift, unveiling Social Reward - an innovative reward modeling framework that leverages implicit feedback from social network users engaged in creative editing of generated images, and utilizes Social Reward to fine-tune text-to-image models, yielding images that are more favored by not only Social Reward, but also other established metrics.
- Attend Who is Weak: Pruning-assisted Medical Image Localization under Sophisticated and Implicit Imbalances9
This paper proposes to use pruning to automatically and adaptively identify hard-to-learn (HTL) training samples, and improve pathology localization by attending them explicitly, during training in supervised, semi- supervised, and weakly-supervised settings.
- Symbolic Distillation for Learned TCP Congestion Control9
A novel symbolic branching algorithm that enables the rule to be aware of the context in terms of various network conditions, eventually converting the NN policy into a symbolic tree and preserving and improving performance over state-of-the-art NN policies while being faster and simpler than a standard neural network.
- Deep Architecture Connectivity Matters for Its Convergence: A Fine-Grained Analysis9
This work theoretically characterize the impact of connectivity patterns on the convergence of DNNs under gradient descent training in fine granularity, and shows that by a simple filtration on "unpromising"connectivity patterns, it can trim down the number of models to evaluate, and significantly accelerate the large-scale neural architecture search without any overhead.
- 9
- 9
- 9
- 9
- 9
- Scaling Textual Gradients via Sampling-Based Momentum8
Textual Stochastic Gradient Descent with Momentum (TSGD-M), which reweights updates through momentum sampling, using bootstrapped minibatch validation accuracy as importance weights over historical prompts to stabilize TSGD and enable effective scaling within a limited context window.
- INR-Arch: A Dataflow Architecture and Compiler for Arbitrary-Order Gradient Computations in Implicit Neural Representation Processing8
The INR-Arch framework is introduced, a framework that transforms the computation graph of an nth-order gradient into a hardware-optimized dataflow architecture, and a compiler that extracts and optimizes computation graphs, automatically configures hardware parameters such as latency and stream depths to optimize throughput, and outputs High-Level Synthesis code for FPGA implementation.
- Graph Ladling: Shockingly Simple Parallel GNN Training without Intermediate Communication8
This work argues to exploit the fundamentals of model soups to mitigate the aforementioned issues of memory bottleneck and trainability during GNNs scaling, and provides a wide variety of model soup preparation techniques by leveraging state-of-the-art graph sampling and graph partitioning approaches that can handle large graphs.
- 8
- Data-Efficient Double-Win Lottery Tickets from Robust Pre-training8
This paper formulate a more rigorous concept, Double-Win Lottery Tickets, in which a located subnetwork from a pre-trained model can be independently transferred on diverse downstream tasks, to reach BOTH the same standard and robust generalization, under BOTH standard and adversarial training regimes, as the full pre- trained model can do.
- 8
- 8
- 8
- 8
- Report on UG 2+ challenge Track 1: Assessing algorithms to improve video object detection and classification from unconstrained mobility platforms8
The protocols and results of Track 1 of the UG^2+ Challenge held in conjunction with IEEE/CVF CVPR 2019 are summarized and a detailed analysis of how they performed on each challenge problem is reported here.
- Adversarially Trained Model Compression: When Robustness Meets Efficiency.8
This paper formally study this new question, by proposing a novel Adversarially Trained Model Compression framework, and demonstrating that ATMC obtains remarkably more favorable trade-off among model size, accuracy and robustness, over currently available alternatives in various settings.
- GradientStabilizer:Fix the Norm, Not the Gradient7
Stable-SPAM is proposed, which incorporates enhanced gradient normalization and clipping techniques and effectively stabilizes gradient norms in 4-bit LLM training, delivering superior performance compared to Adam and SPAM.
- Rethinking Addressing in Language Models via Contexualized Equivariant Positional Encoding7
TAPE is proposed, a novel framework that enhances positional embeddings by incorporating sequence content across layers by enforcing permutation and orthogonal equivariance and achieves superior performance in language modeling, arithmetic reasoning, and long-context retrieval tasks compared to existing positional embedding techniques.
- You Only Transfer What You Share: Intersection-Induced Graph Transfer Learning for Link Prediction7
This work develops a framework to effectively leverage the structural prior in this setting, and considers two approaches: a modified label propagation, and a multi-layer perceptron (MLP) model in a teacher-student regime.
- 7
- 7
- Supervised pretraining through contrastive categorical positive samplings to improve COVID-19 mortality prediction7
A supervised pre- training model with a unique embedded k-nearest-neighbor positive sampling strategy that outperforms the alternative pre-training models and traditional machine learning methods in predicting patient mortality in real-world COVID-19 EHR data.
- 7
- Troubleshooting image segmentation models with human-in-the-loop7
The feasibility of the framework using the semantic segmentation task in PASCAL VOC is demonstrated, and it is found that the fine-tuned model exhibits significantly improved generalization when applied to real-world images with greater content diversity.
- 7
- 7
- 7
- 7
- InstantNet: Automated Generation and Deployment of Instantaneously Switchable-Precision Networks7
This work proposes InstantNet to automatically generate and deploy instantaneously switchable-precision networks which operates at variable bit-widths and shows that the proposed InstantNet consistently outperforms state-of-the-art designs.
- 7
- 7
- 7
- 7
- 7
- 7
- 7
- More is Less: The Pitfalls of Multi-Model Synthetic Preference Data in DPO Safety Alignment6
It is demonstrated that multi-model preference data exhibits high linear separability between chosen and rejected responses, which allows models to exploit superficial cues rather than internalizing robust safety constraints.
- Artificial intelligence-guided detection of under-recognized cardiomyopathies on point-of-care cardiac ultrasound: a multi-center study6
An AI framework that enables scalable, opportunistic screening of under-diagnosed cardiomyopathies using POCUS is defined and validated, leveraging a novel acquisition quality-adapted modeling strategy.
- Generalization Error Analysis for Sparse Mixture-of-Experts: A Preliminary Study6
This paper investigates the impact of the number of data samples, the total number of experts, the sparsity in expert selection, the complexity of the routing mechanism, and the complexity of individual experts on Sparse Mixture-of-Experts' generalization error.
- 6
- Data-Model-Circuit Tri-Design for Ultra-Light Video Intelligence on Edge Devices6
This paper proposes a data-model-hardware tri-design framework for high-throughput, low-cost, and high-accuracy multi-object tracking (MOT) on High-Definition (HD) video stream, and makes a solid step towards the synergized software/hardware co-optimization for realistic MOT model implementation.
- Towards Lifelong Learning of Multilingual Text-to-Speech Synthesis6
This work forms the replay process as a supervised learning problem, and proposes a simple yet effective dual-sampler framework to tackle the heavily language-imbalanced training samples, and shows that this supervised learning formulation outperforms other gradient-based and regularization-based lifelong learning methods.
- Is Attention All That NeRF Needs?6
It is shown that depth and occlusion can be inferred from the learned attention maps, which implies that the pure attention mechanism is capable of learning a physically-grounded rendering process.
- Black-Box Diagnosis and Calibration on GAN Intra-Mode Collapse: A Pilot Study6
This study reveals that the intra-mode collapse is still a prevailing problem in state-of-the-art GANs and the mode collapse is diagnosable and calibratable in black-box settings.
- 6
- 6
- Neurosymbolic AI as an antithesis to scaling laws5
Neurosymbolic AI is a growing area that promotes methodological heterogeneity and aims to push the frontiers of AI through affordable data and computing power.
- Demystifying the Visual Quality Paradox in Multimodal Large Language Models5
A visual-quality paradox is uncovered: model, task, and even individual-instance performance can improve when images deviate from human-perceived fidelity, and the need for adaptive, rather than universally ``clean'', imagery, in the new era of AI being the main data customer is highlighted.
- Neon: Negative Extrapolation From Self-Training Improves Image Generation5
This paper introduces Neon (for Negative Extrapolation frOm self-traiNing), a new learning method that turns the degradation from self-training into a powerful signal for self-improvement and demonstrates Neon's universality across a range of architectures and datasets.
- 5
- Oscillation Inversion: Understand the structure of Large Flow Model through the Lens of Inversion Method5
A simple and fast distribution transfer technique that facilitates image enhancement, stroke-based recoloring, as well as visual prompt-guided image editing is introduced, and quantitative results demonstrating the effectiveness of this method for tasks such as image enhancement, makeup transfer, reconstruction quality, and guided sampling quality are provided.
- Reference-based Painterly Inpainting via Diffusion: Crossing the Wild Reference Domain Gap5
A novel diffusion framework to ``inpaint more wildly'' by taking such references with large domain gaps, built with an image-conditioned diffusion model and introduced with a ladder-side branch and a masked fusion mechanism to work with the inpainting mask.
- M-L2O: Towards Generalizable Learning-to-Optimize by Test-Time Fast Self-Adaptation5
This paper investigates a potential solution to this open challenge, by meta-training an L2O optimizer that can perform fast test-time self-adaptation to an out-of-distribution task, in only a few steps, by proposing a proposed framework (termed as M-L2O).
- Meta ControlNet: Enhancing Task Adaptation via Meta Learning5
A novel Meta ControlNet method is introduced, which adopts the task-agnostic meta learning technique and features a new layer freezing design and achieves control within only 100 finetuning steps in more complex non-edge tasks such as Human Pose, outperforming all existing methods.
- 5
- Density-Aware Personalized Training for Risk Prediction in Imbalanced Medical Data5
This work first decouple the feature extraction and classification process, adjusting training batches separately for each component to mitigate bias caused by class density discrepancy, and trains the network with both a density-aware loss and a learnable cost matrix for misclassifications.
- AutoMARS: Searching to Compress Multi-Modality Recommendation Systems5
This paper proposes Auto Multi-modAlity Recommendation System (AutoMARS), a unified modality-aware model compression framework dedicated to multi-modality recommendation systems and demonstrates superior multi- modality compression performance than previous state-of-the-art compression methods.
- TxVAD: Improved Video Action Detection by Transformers5
A conceptually simple paradigm for video action detection using Transformers, which effectively removes the need for specialized components and achieves superior performance, without using pre-trained person/object detectors, RPN, or memory bank.
- 5
- VGAI: A Vision-Based Decentralized Controller Learning Framework for Robot Swarms5
Vision-based Graph Aggregation and Inference (VGAI), a decentralized learning-to-control framework that directly maps raw visual observations to agent actions, is presented, aided by sparse local communication among only neighboring agents.
- 5
- LAD-VF: LLM-Automatic Differentiation Enables Fine-Tuning-Free Robot Planning from Formal Methods Feedback4
This paper proposes LAD-VF, a fine-tuning-free framework that leverages formal verification feedback for automated prompt engineering and presents a scalable and interpretable pathway toward trustworthy, formally-verified LLM-driven control systems.
- On-the-Fly Adaptive Distillation of Transformer to Dual-State Linear Attention4
DSLA is proposed, a novel design that maintains two specialized hidden states-one for preserving historical context and one for tracking recency-thereby mitigating the short-range bias typical of linear-attention architectures and DSLA-Serve, an online adaptive distillation framework that progressively replaces Transformer layers with DSLA layers at inference time, guided by a sensitivity-based layer ordering.
- 4
- Principled Architecture-aware Scaling of Hyperparameters4
This work precisely characterize the dependence of initializations and maximal learning rates on the network architecture, which includes the network depth, width, convolutional kernel size, and connectivity patterns and demonstrates that network rankings can be easily changed by better training networks in benchmarks with architecture-aware learning rates and initialization.
- Safe and Robust Watermark Injection with a Single OoD Image4
This work proposes a safe and robust backdoor-based watermark injection technique that leverages the diverse knowledge from a single out-of-distribution (OoD) image, which serves as a secret key for IP verification and is agnostic to third-party promises of IP security.
- QuantumSEA: In-Time Sparse Exploration for Noise Adaptive Quantum Circuits4
The proposed QuantumSEA is an in-time sparse exploration for noise-adaptive quantum circuits, aiming to achieve two key objectives: implicit circuits capacity during training and noise robustness by jointly optimizing the topology and parameters of quantum circuits under real device noise models.
- 4
- 4
- QuanGCN: Noise-Adaptive Training for Robust Quantum Graph Convolutional Networks4
This work proposes quantum graph convolutional networks (QuanGCN), which learns the local message passing among nodes with the sequence of crossing-gate quantum operations, and applies sparse constraint to sparsify the nodes' connections and use skip connection to augment the quantum outputs with original node features to improve robustness.
- Model elasticity for hardware heterogeneity in federated learning systems4
This work proposes a model-architecture co-design framework for FL optimization based on the new concept of model elasticity, which enables local devices to train different models belonging to the same architecture family, selected to match the resource budgets of various edge devices.
- 4
- 4
- 4
- 4
- Turning A Curse into A Blessing: Data-Aware Memory-Efficient Training of Graph Neural Networks by Dynamic Exiting3
A novel framework, MADE-GNN, which capitalizes on the power-law nature of graph data to enhance training efficiency and turns the power-law distribution from a traditional "curse'' into a strategic "blessing''.
- All Against Some: Efficient Integration of Large Language Models for Message Passing in Graph Neural Networks3
This paper proposes E-LLaGNN (Efficient LLMs augmented GNNs), a framework with an on-demand LLM service that enriches message passing procedure of graph learning by enhancing a limited fraction of nodes from the graph.
- Pose-Free Generalizable Rendering Transformer3
A comprehensive comparison of the performance of DragView with recent scene representation networks operating under pose-free conditions, as well as with generalizable NeRFs subject to noisy test camera poses, demonstrates its superior performance in view synthesis quality, while also being more user-friendly.
- Rethinking PGD Attack: Is Sign Function Necessary?3
A new raw gradient descent (RGD) algorithm is proposed that converts the constrained optimization problem into an unconstrained one, by introducing a new hidden variable of non-clipped perturbation that can move beyond the constraint.
- Abstract 18776: ECG-GPT: Automated Complete Diagnosis Generation From ECG Images Using Novel Vision-Text Transformer Model3
This approach represents an automated and accessible strategy for generating cardiologist-level reporting on ECGs, useful for low-resource settings.
- A Multimodality Video-Based AI Biomarker For Aortic Stenosis Development And Progression3
An AI model built for two-dimensional-echocardiography can stratify the risk of AS progression, with implications for longitudinal monitoring in the community.
- 3
- Pruning Before Training May Improve Generalization, Provably3
This work considers a classification task for overparameterized two-layer neural networks, where the network is randomly pruned according to different rates at the initialization, and shows that as long as the pruning fraction is below a certain threshold, gradient descent can drive the training loss toward zero and the network exhibits good generalization performance.
- E^2TAD: An Energy-Efficient Tracking-based Action Detector3
This paper presents a tracking-based solution to accurately and efficiently localize predefined key actions spatially (by predicting the associated target IDs and locations) and temporally (by predicting the time in exact frame indices).
- 3
- 3
- E2VTS: Energy-Efficient Video Text Spotting from Unmanned Aerial Vehicles3
This paper revisits RCNN’s crop & resize training strategy and empirically finds that it outperforms aligned RoI sampling on a real-world video text dataset captured by UAV, and proposes a multi-stage image processor that takes videos’ redundancy, continuity, and mixed degradation into account.
- 3
- 3
- 3
- 3
- 3
- 3
- Lean Refactor: Multi-Objective Controllable Proof Optimization via Agentic Strategy Search2
Lean Refactor steers a frozen agentic LLM with retrievals from a curated database of multi-objective refactoring strategies, each densely annotated with metadata such as supported Lean/Mathlib versions and expected compilation-cost reduction.
- ResearchGPT: Benchmarking and Training LLMs for End-to-End Computer Science Research Workflows2
This work contributes CS-54k, a high-quality corpus of scientific Q&A pairs in computer science, built from 14k CC-licensed papers, and derives two complementary subsets: CS-4k, a carefully curated benchmark for evaluating AI's ability to assist scientific research, and CS-50k, a large-scale training dataset.
- On How Iterative Magnitude Pruning Discovers Local Receptive Fields in Fully Connected Neural Networks2
This work demonstrates first that non-Gaussian input statistics are indeed necessary for IMP to discover localized RFs, and develops a new method for measuring the effect of individual weights on the statistics of the FCN representations ("cavity method"), which allows it to show that IMP systematically increases the non-Gaussianity of pre-activations, leading to the formation of localized RFs.
- 2
- 2
- Doubly Robust Instance-Reweighted Adversarial Training2
A novel doubly-robust instance reweighted AT framework is proposed, which allows to obtain the importance weights via exploring distributionally robust optimization (DRO) techniques, and at the same time boosts the robustness on the most vulnerable examples, and allows to devise new algorithms with a theoretical convergence guarantee.
- Dynimp: Dynamic Imputation for Wearable Sensing Data through Sensory and Temporal Relatedness2
This work proposes a model to handle different time point’s missingness with nearest neighbors along feature axis and then feeding the data into a LSTM-based denoising autoen-coder which can reconstruct missingness along the time axis, and experiments on activity recognition show that the method can exploit the multi-modality features from related sensors.
- SmartDeal: Remodeling Deep Network Weights for Efficient Inference and Training2
SmartDeal is a hardware-friendly algorithm framework to trade higher-cost memory storage/access for lower-cost computation, in order to aggressively boost the storage and energy efficiency, for both DNN inference and training.
- Grasping the Arrow of Time from the Singularity: Decoding Micromotion in Low-dimensional Latent Spaces from StyleGAN2
This paper hypothesize and demonstrate that a series of meaningful, natural, and versatile small, local movements (referred to as"micromotion", such as expression, head movement, and aging effect) can be represented in low-rank spaces extracted from the latent space of a conventionally pre-trained StyleGAN-v2 model for face generation, with the guidance of proper"anchors" in the form of either short text or video clips.
- Cloud2Sketch: Augmenting Clouds with Imaginary Sketches2
This work presents an interesting task that augments clouds in the sky with imagined sketches, and proposes Cloud2Sketch, a novel self-supervised pipeline to tackle the unique challenges of real-world clouds and sketch generation without sketch retrieval.
- 2
- 2
- 2
- Signal Processing2
The experimental results demonstrate that the DUpsampling strategy achieves state-of-the-art performance with competing methods, and can achieve the segmentation speed as 90.84FPS and the segmentation accuracy as 90.770 % mIoU.
- 2
- P1‐118: QUANTIFICATION OF AMYLOID BURDEN FROM FLORBETAPIR PET WITHOUT USING TARGET AND REFERENCE REGIONS: PRELIMINARY FINDINGS BASED ON THE DEEP LEARNING 3D CONVOLUTIONAL NEURAL NETWORK APPROACH2
Structural/biological activity relationship for A b aggregates is provided and the concept of conformational strains to non-PrP prion-like proteins is established to establish the concept of conformational strains to non-PrP prion-like proteins.
- 2
- UNCAP: Uncertainty-Guided Neurosymbolic Planning Using Natural Language Communication for Cooperative Autonomous Vehicles1
UNCAP is a vision–language model-based planning approach that enables CAVs to communicate via lightweight natural language messages while explicitly accounting for perception uncertainty in decision-making, improving both the scalability and reliability of cooperative planning.
- No Hidden Prompts Needed! You Can Game AI Peer Review with Presentation-Only Revisions1
The results show that the deployment risk is not only malicious hidden instructions, but the emergence of paper presentation itself as an optimization surface, and reveals two deeper structural failure modes.
- AI-CONECT: Designing Responsible-AI-based Conversational Chatbots for Dementia Intervention1
The preliminary results demonstrate a pathway for integrating responsible AI into enhancing cognitive functions, offering a practical and scalable alternative to human-facilitated conversational interventions.
- Make Optimization Once and for All with Fine-grained Guidance1
This work analyses general framework for learning optimization, called Diff-L2O, focusing on augmenting sampled solutions from a wider view rather than local updates in real optimization process only, and gives the related generalization bound, showing that the sample diversity of Diff-L2O brings better performance.
- Why Neural Network Can Discover Symbolic Structures with Gradient-based Training: An Algebraic and Geometric Foundation for Neurosymbolic Reasoning1
A theoretical framework that explains how discrete symbolic structures can emerge naturally from continuous neural network training dynamics is developed, and data scaling laws for realizing symbolic tasks are established, linking representational capacity to the group invariance that facilitates symbolic solutions.
- PIPA: Preference Alignment as Prior-Informed Statistical Estimation1
PIPA is introduced, a unified, RL-free probabilistic framework that formulates language model preference alignment as a Maximum Likelihood Estimation (MLE) problem with prior constraints that effectively accommodates both paired and unpaired data, as well as answer and step-level annotations.
- 1
- Chasing Better Deep Image Priors between Over- and Under-parameterization1
The results validate the superiority of LIPs: they can successfully locate the LIP subnetworks from over-parameterized DIPs at substantial sparsity ranges, and those LIP subnetworks significantly outperform deep decoders under comparably compact model sizes, and they also possess high transferability across different images as well as restoration task types.
- 1
- The Counterattack of CNNs in Self-Supervised Learning: Larger Kernel Size might be All You Need1
Self-supervised pre-training is applied to the recently proposed, stronger lager-kernel CNN architecture and an apple-to-apple comparison with Transformers is conducted, to help better understand what is essential (or not) for self-super supervised learning backbones.
- Modeling user choice behavior under data corruption: Robust learning of the latent decision threshold model1
This article shows that, among the existing robust learning frameworks, the L0-norm-based framework can outperform other state-of-the-art methods in terms of prediction accuracy and model estimation, and develops a user screening algorithm to identify potential bad actors.
- 1
- 1
- APP: Anytime Progressive Pruning1
A novel way of progressive pruning is proposed, referred to as Anytime Progressive Pruning (APP), which significantly outperforms the baseline dense and Anytime OSP models across multiple architectures and datasets under short, moderate, and long-sequence training.
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- 1
- From Bi-Level Sparse Clustering to Deep Clustering1
This work proposes to emulate the sparse coding-based clustering pipeline in the context of deep learning, leading to a carefully crafted deep model benefiting from both sparse coding and discriminative clustering, and discovers that connecting deep learning to sparse coding benefits not only the model performance, but also its initialization and interpretation.
- 1
- 1
- 1
- 1
- 1
- –
- When Do Graph Foundation Models Transfer? A Data-Centric Theory–
This work shows that for both set-based and message-passing tokenizations, any Lipschitz backbone admits an explicit decomposition of cross-domain output shift into graph-specific finite-sample approximation terms and an intrinsic, relabeling-invariant domain discrepancy capturing structural mismatch.
- –
- Foundation Models for Logistics: Toward Certifiable, Conversational Planning Interfaces–
A Vision-Language Logistics (VLL) agent is introduced, built on a neurosymbolic framework that pairs the accessibility of natural-language dialogue with verifiable guarantees on user-objective interpretation and surpasses the zero-shot performance of 20x larger models in logistic planning tasks while cutting inference latency by nearly 50%.
- –
- –
- –
- –
- –
- –
- –
- –
- –
- DANCE: DAta-Network Co-optimization for Efficient Segmentation Model Training and Inference–
DANCE is proposed, general automated DAta-Network Co-optimization for Efficient segmentation model training and inference that can achieve “all-win” towards efficient segmentation (reduced training cost, less expensive inference, and better mean Intersection-over-Union (mIoU).
- Abstract 13659: Automated Detection of Aortic Stenosis From Single-View 2-Dimensional Echocardiography Using a Semi-Supervised, Contrastive Learning Approach–
A novel method to detect severe AS using single-view TTE videos without requiring Doppler data is developed, which has significant implications for point-of-care ultrasound screening as part of routine clinic visits and in low-resource settings.
- Editorial for the Special Issue on Advanced Machine Learning Techniques for Sensing and Imaging Applications–
This paper presents a meta-modelling architecture suitable for large-scale optimization of neural networks, and some examples show how this architecture can be applied to image recognition and 3D image analysis.
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- –
- Image Dehazing: Improved Techniques–
Two related but important tasks based on the recently released REalistic Single Image DEhazing (RESIDE) benchmark dataset are explored: (i) single image dehazing as a low-level image restoration problem; and (ii) high-level visual understanding of hazy images.
- –
- Growing Deep Forests Efficiently with Soft Routing and Learned Connectivity–
A probabilistic tree whose nodes make soft routing decisions, a.k.a., "soft routing", rather than hard binary decisions is employed, which enables non-greedy optimization for each tree and proposes an innovative topology learning strategy.
- Task-specific and interpretable feature learning–
This talk will introduce the research on bridging traditional learning models that emphasize problem-specific reasoning, and deep models that allow for larger learning capacity, and reveal how the analytic tools in the classical optimization problems can be translated to guide the architecture design and performance analysis of deep models.
Publication data from OpenAlex, with missing venues and authors filled in from Crossref; citation counts are the higher of OpenAlex and Semantic Scholar, last synced 2026-10-12. One-sentence summaries under some papers are written by Semantic Scholar’s model. Citation counts may be lower than on Google Scholar, which indexes more sources.
Report an error
Wrong papers, two people merged into one, or a profile that should not be here? Tell us and we will fix or hide it.