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
- Cheng WuSuggested from co-authorship
- Gary J. ChengSuggested from co-authorship
Is this you? Claim this profile to confirm or dismiss it.
Works280 from public data
- Densely Connected Convolutional Networks45,608
The Dense Convolutional Network (DenseNet), which connects each layer to every other layer in a feed-forward fashion, and has several compelling advantages: they alleviate the vanishing-gradient problem, strengthen feature propagation, encourage feature reuse, and substantially reduce the number of parameters.
- Learning Efficient Convolutional Networks through Network Slimming2,909
The approach is called network slimming, which takes wide and large networks as input models, but during training insignificant channels are automatically identified and pruned afterwards, yielding thin and compact models with comparable accuracy.
- Deep Networks with Stochastic Depth2,786
Stochastic depth is proposed, a training procedure that enables the seemingly contradictory setup to train short networks and use deep networks at test time and reduces training time substantially and improves the test error significantly on almost all data sets that were used for evaluation.
- Trends in extreme learning machines: A review1,732
This review gives an overview of ELM from the theoretical perspective, including the interpolation theory, universal approximation capability, and generalization ability, and focuses on the various improvements made to ELM which further improve its stability, sparsity and accuracy under general or specific conditions.
- Dynamic Neural Networks: A Survey1,012
This survey comprehensively review this rapidly developing area of dynamic networks by dividing dynamic networks into three main categories: sample-wise dynamic models that process each sample with data-dependent architectures or parameters; spatial-wiseynamic networks that conduct adaptive computation with respect to different spatial locations of image data; and temporal-wise Dynamic networks that perform adaptive inference along the temporal dimension for sequential data.
- Vision Transformer with Deformable Attention972
A novel deformable selfatt attention module is proposed, where the positions of key and value pairs in selfattention are selected in a data-dependent way, which enables the self-attention module to focus on relevant re-gions and capture more informative features.
- CondenseNet: An Efficient DenseNet Using Learned Group Convolutions881
CondenseNet is developed, a novel network architecture with unprecedented efficiency that combines dense connectivity with a novel module called learned group convolution, allowing for efficient computation in practice.
- 872
- 797
- Convolutional Networks with Dense Connectivity590
The Dense Convolutional Network (DenseNet), which connects each layer to every other layer in a feed-forward fashion, is introduced and achieves significant improvements over the state-of-the-art on most of them, whilst requiring less parameters and computation to achieve high performance.
- On the Integration of Self-Attention and Convolution551
A mixed model that enjoys the benefit of both self-Attention and Convolution (ACmix), while having minimum compu-tational overhead compared to the pure convolution or self-attention counterpart.
- Horizontal Pyramid Matching for Person Re-Identification512
A simple yet effective Horizontal Pyramid Matching (HPM) approach to fully exploit various partial information of a given person, so that correct person candidates can be still identified even even some key parts are missing.
- BEVFormer v2: Adapting Modern Image Backbones to Bird's-Eye-View Recognition via Perspective Supervision490
A novel bird's-eye-view (BEV) detector with perspective supervision, which converges faster and bet-suits modern image backbones and achieves new SoTA results on the large-scale nuScenes dataset.
- 3D Object Detection with Pointformer472
This paper proposes Pointformer, a Transformer backbone designed for 3D point clouds to learn features effectively, and introduces an efficient coordinate refinement module to shift down-sampled points closer to object centroids, which improves object proposal generation.
- FLatten Transformer: Vision Transformer using Focused Linear Attention423
This paper proposes a novel Focused Linear Attention module, which introduces a simple yet effective mapping function and an efficient rank restoration module to enhance the expressiveness of self-attention while maintaining low computation complexity.
- Deep learning-based artificial intelligence model to assist thyroid nodule diagnosis and management: a multicentre diagnostic study377
The ThyNet-assisted strategy can significantly improve the diagnostic performance of radiologists and help reduce unnecessary fine needle aspirations for thyroid nodules.
- Ghost in the Minecraft: Generally Capable Agents for Open-World Environments via Large Language Models with Text-based Knowledge and Memory368
Ghost in the Minecraft (GITM), a novel framework that integrates Large Language Models (LLMs) with text-based knowledge and memory, aiming to create Generally Capable Agents (GCAs) in Minecraft, shows the potential of LLMs in developing capable agents for handling long-horizon, complex tasks and adapting to uncertainties in open-world environments.
- Frequency-Aware Feature Fusion for Dense Image Prediction325
Comprehensive visualization and quantitative analysis demonstrate that FreqFusion effectively improves feature consistency and sharpens object boundaries, and extensive experiments across various dense prediction tasks confirm its effectiveness.
- Domain Adaptation via Prompt Learning289
This article introduces a novel prompt learning paradigm for UDA, named domain adaptation via prompt learning (DAPrompt), which outperforms previous methods on several cross-domain benchmarks but also is very efficient to train and easy to implement.
- 277
- Resource Aware Person Re-identification Across Multiple Resolutions266
This work presents a new person re-ID model that combines effective embeddings built on multiple convolutional network layers, trained with deep-supervision, and proposes two new formulations of the person reID problem under resource-constraints.
- Resolution Adaptive Networks for Efficient Inference256
This paper focuses on spatial redundancy of input samples and proposes a novel Resolution Adaptive Network (RANet), inspired by the intuition that low-resolution representations are sufficient for classifying “easy” inputs containing large objects with prototypical features, while only some “hard” samples need spatially detailed information.
- How Far is Video Generation from World Model: A Physical Law Perspective244
This study suggests that scaling alone is insufficient for video generation models to uncover fundamental physical laws, despite its role in Sora's broader success.
- SePiCo: Semantic-Guided Pixel Contrast for Domain Adaptive Semantic Segmentation224
Semantic-Guided Pixel Contrast (SePiCo), a novel one-stage adaptation framework that highlights the semantic concepts of individual pixels to promote learning of class-discriminative and class-balanced pixel representations across domains, eventually boosting the performance of self-training methods.
- 223
- Anytime Stereo Image Depth Estimation on Mobile Devices223
This work proposes a novel approach for disparity prediction in the anytime setting that can trade off computation and accuracy at inference time, and can process $1242 \times 375$ resolution images within a range of 10-35 FPS on an NVIDIA Jetson TX2 module with only marginal increases in error.
- GSVA: Generalized Segmentation via Multimodal Large Language Models214
Generalized Segmentation Vision Assistant (GSVA) is proposed, which reuses the [SEG] token to prompt the segmentation model towards supporting multiple mask ref-erences simultaneously and innovatively learns to generate a [REJ] token to reject the null targets explicitly.
- 214
- Gradient boosted feature selection209
This work proposes a novel feature selection algorithm, Gradient Boosted Feature Selection (GBFS), which satisfies all four requirements and is flexible, scalable, and surprisingly straight-forward to implement.
- 208
- Large-scale nanoshaping of ultrasmooth 3D crystalline metallic structures205
A low-cost, high-throughput benchtop method that enables thin layers of metal to be shaped with nanoscale precision by generating ultrahigh-strain-rate deformations that overcome the metal's normal strength and, thus, resistance to patterning.
- Regularizing Deep Networks with Semantic Data Augmentation197
The proposed implicit semantic data augmentation (ISDA) algorithm amounts to minimizing a novel robust CE loss, which adds minimal extra computational cost to a normal training procedure, and can be applied to semi-supervised learning tasks under the consistency regularization framework.
- Adaptive Rotated Convolution for Rotated Object Detection196
In this paper, the convolution kernels rotate adaptively to extract object features with varying orientations in different images, and an efficient conditional computation mechanism is introduced to accommodate the large orientation variations of objects within an image.
- Deep Residual Correction Network for Partial Domain Adaptation160
An efficiently-implemented Deep Residual Correction Network is proposed by plugging one residual block into the source network along with the task-specific feature layer, which effectively enhances the adaptation from source to target and explicitly weakens the influence from the irrelevant source classes.
- Improved Techniques for Training Adaptive Deep Networks159
This paper considers a typical adaptive deep network with multiple intermediate classifiers and presents three techniques to improve its training efficacy from two aspects: a Gradient Equilibrium algorithm to resolve the conflict of learning of different classifiers; an Inline Subnetwork Collaboration approach and a One-for-all Knowledge Distillation algorithm to enhance the collaboration among classifiers.
- Multi-Scale Dense Convolutional Networks for Efficient Prediction.150
A new convolutional neural network architecture with the ability to adapt dynamically to computational resource limits at test time and substantially improves the state-of-the-art in both settings.
- 148
- Frequency Domain Image Translation: More Photo-realistic, Better Identity-preserving131
A novel frequency domain image translation (FDIT) framework, exploiting frequency information for enhancing the image generation process, and effectively preserves the identity of the source image, and produces photo-realistic images.
- Joint Adversarial Domain Adaptation128
A Joint Adversarial Domain Adaptation approach to simultaneously align domain-wise and class-wise distributions across source and target in a unified adversarial learning process, and can gain remarkable improvements over other state-of-the-art deep domain adaptation approaches.
- Domain Conditioned Adaptation Network125
This paper relaxation a shared-convnets assumption made by previous DA methods and proposes a Domain Conditioned Adaptation Network (DCAN), which aims to excite distinct convolutional channels with a domain conditioned channel attention mechanism, which outperforms existing methods by a large margin.
- Slide-Transformer: Hierarchical Vision Transformer with Local Self-Attention123
A novel local attention module, Slide Attention, which leverages common convolution operations to achieve high efficiency, flexibility and generalizability and is applicable to a variety of advanced Vision Transformer models and compatible with various hardware devices, and achieves consistently improved performances on comprehensive benchmarks.
- Siamese Image Modeling for Self-Supervised Vision Representation Learning121
SiameseIM can surpass both ID and MIM on a wide range of downstream tasks, including ImageNet finetuning and linear probing, COCO and LVIS detection, and ADE20k semantic segmentation.
- Adaptive Focus for Efficient Video Recognition119
This paper model the patch localization problem as a sequential decision task, and proposes a reinforcement learning based approach for efficient spatially adaptive video recognition (AdaFocus), whose features are used by a recurrent policy network to localize the most task-relevant regions.
- Probabilistic Contrastive Learning for Long-Tailed Visual Recognition118
This work proposes a novel probabilistic contrastive (ProCo) learning algorithm that estimates the data distribution of the samples from each class in the feature space, and samples contrastive pairs accordingly, and analyzes the error bound of ProCo.
- Cross-Iteration Batch Normalization111
Cross-Iteration Batch Normalization (CBN) is presented, in which examples from multiple recent iterations are jointly utilized to enhance estimation quality and outperform the original batch normalization and a direct calculation of statistics over previous iterations without the proposed compensation technique.
- Avalon's Game of Thoughts: Battle Against Deception through Recursive Contemplation110
A novel framework, Recursive Contemplation (ReCon), is introduced to enhance LLMs' ability to identify and counteract deceptive information and indicates its efficacy in aiding LLMs to discern and maneuver around deceptive information without extra fine-tuning and data.
- 106
- Spatially Adaptive Inference with Stochastic Feature Sampling and Interpolation106
This work proposes to compute features only at sparsely sampled locations, which are probabilistically chosen according to activation responses, and then densely reconstruct the feature map with an efficient interpolation procedure, using a reparameterization trick based on the Gumbel-Softmax distribution.
- 100
- Interpretable Spatio-Temporal Attention for Video Action Recognition97
This model not only improves video action recognition accuracy, but also localizes discriminative regions both spatially and temporally, despite being trained in a weakly-supervised manner with only classification labels (no bounding box labels or time frame temporal labels).
- 95
- 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.
- 85
- CondenseNet V2: Sparse Feature Reactivation for Deep Networks76
This paper proposes an alternative approach named sparse feature reactivation (SFR), aiming at actively increasing the utility of features for reusing, and achieves promising performance on image classification and object detection in terms of both theoretical efficiency and practical speed.
- Twin extreme learning machines for pattern classification76
A novel twin extreme learning machine to simultaneously train two ELMs with two nonparallel classification hyperplanes, which incorporates the idea of twin support vector machine (TSVM) into the basic framework of ELM, thus TELM could have the advantages of the both algorithms.
- Fine-Grained Recognition With Learnable Semantic Data Augmentation75
This paper proposes diversifying the training data at the feature-level to alleviate the discriminative region loss problem, and produces diversified augmented samples by translating image features along semantically meaningful directions.
- Pseudo-Q: Generating Pseudo Language Queries for Visual Grounding75
A novel method, named Pseudo-Q, to automatically generate pseudo language queries for supervised training, which can reduce human annotation costs and achieves supe-rior or comparable performance compared to state-of-the-art weakly-supervised visual grounding methods on all the five datasets the authors have experimented.
- Glance and Focus Networks for Dynamic Visual Recognition75
The proposed Glance and Focus Network (GFNet) first extracts a quick global representation of the input image at a low resolution scale, and then strategically attends to a series of salient regions to learn finer features, mimicking the human visual system.
- Exploring the Equivalence of Siamese Self-Supervised Learning via A Unified Gradient Framework74
UniGrad is proposed, a simple but effective gradient form for self-supervised learning that does not require a memory bank or a predictor network, but can still achieve state-of-the-art performance and easily adopt other training strategies.
- Synthesis of 2,5-Disubstituted 3-Iodofurans via Palladium-Catalyzed Coupling and Iodocyclization of Terminal Alkynes74
2,5-Disubstituted 3-iodofurans are readily prepared under very mild reaction conditions by the palladium/copper-catalyzed cross-coupling of (Z)-β-bromoenol acetates and terminal alkynes, followed by iodocyclization.
- 73
- Mask Grounding for Referring Image Segmentation72
A novel Mask Grounding auxiliary task that significantly improves visual grounding within language features, by explicitly teaching the model to learn fine-grained correspondence between masked textual tokens and their matching visual objects is introduced, demonstrating the method's effectiveness in addressing current limitations of RIS algorithms.
- AdaFocus V2: End-to-End Training of Spatial Dynamic Networks for Video Recognition72
This work reformulates the training of AdaFocus as a simple one-stage algorithm by introducing a differentiable interpolation-based patch selection operation, enabling efficient end-to-end optimization and presenting an improved training scheme to address the issues introduced by the one- stage formulation.
- Generalized Domain Conditioned Adaptation Network71
This paper relaxes the completely-shared convnets assumption adopted by previous DA methods and proposes Domain Conditioned Adaptation Network (DCAN), which introduces domain conditioned channel attention module with a multi-path structure to separately excite channel activation for each domain.
- 70
- 68
- 66
- Smooth Diffusion: Crafting Smooth Latent Spaces in Diffusion Models65
This work introduces Step-wise Variation Regularization to enforce the proportion between the variations of an arbitrary input latent and that of the output image is a constant at any diffusion training step, and devise an interpolation standard deviation (ISTD) metric to effectively assess the latent space smoothness of a diffusion model.
- Deep Siamese Networks Based Change Detection with Remote Sensing Images65
A supervised change detection method based on a deep siamese semantic segmentation network to handle the proposed data effectively and demonstrate the promising performance of the proposed network compared to existing approaches.
- Cross-modal adapter for vision–language retrieval64
A novel Cross-Modal Adapter for parameter-efficient transfer learning that reduces the vast majority of fine-tuned parameters, saves training time, and allows all the pre-trained parameters to be fixed, enabling the pre-trained model to be shared across datasets.
- 63
- Towards All-in-One Pre-Training via Maximizing Multi-Modal Mutual Information58
This paper proposes a general multimodal mutual information formula as a unified optimization target and demonstrates that all mainstream approaches are special cases of the authors' framework, named Maximizing Multi-modal Mutual Information Pre-Training (M3I Pre-training).
- Unsupervised neighborhood component analysis for clustering56
The proposed algorithm aims to maximize a stochastic variant of the leave-one-out K-nearest neighbor (KNN) score on unlabeled data, which performs distance metric learning and clustering simultaneously, and can also learn a low dimensional projection of high dimensional data.
- PsychoGAT: A Novel Psychological Measurement Paradigm through Interactive Fiction Games with LLM Agents54
This work proposes PsychoGAT (Psychological Game AgenTs) to achieve a generic gamification of psychological assessment, and demonstrates that powerful LLMs can function both as adept psychologists and innovative game designers.
- 54
- 54
- 53
- 53
- Discriminative clustering via extreme learning machine52
This paper proposes three discriminative clustering approaches based on Extreme Learning Machine (ELM), and shows that the proposed algorithms can be easily implemented, and yield competitive clustering accuracy on real world data sets compared to state-of-the-art clustering methods.
- 49
- DAT++: Spatially Dynamic Vision Transformer with Deformable Attention48
A novel deformable multi-head attention module, where the positions of key and value pairs in self-attention are adaptively allocated in a data-dependent way, which enables the proposed deformable attention to dynamically focus on relevant regions while maintains the representation power of global attention.
- 47
- ART: Anonymous Region Transformer for Variable Multi-Layer Transparent Image Generation45
The Anonymous Region Transformer (ART) is introduced, which facilitates the direct generation of variable multi-layer transparent images based on a global text prompt and an anonymous region layout that allows the generative model to autonomously determine which set of visual tokens should align with which text tokens.
- 4D LangSplat: 4D Language Gaussian Splatting via Multimodal Large Language Models45
4D LangSplat is proposed, which learns 4D language fields to handle time-agnostic or time-sensitive open-vocabulary queries in dynamic scenes efficiently and attains precise and efficient results for both timesensitive and time-agnostic open-vocabulary queries.
- Dynamic Perceiver for Efficient Visual Recognition45
This paper proposes Dynamic Perceiver (Dyn-Perceiver) to decouple the feature extraction procedure and the early classification task with a novel dual-branch architecture, which significantly improves the inference efficiency of different backbones, outperforming numerous competitive approaches across a broad range of computational budgets.
- Learning Specialized Activation Functions for Physics-Informed Neural Networks43
This work reveals the connection between the optimization difficulty of PINNs and activation functions and proposes to tailor the idea of learning combinations of candidate activation functions to the PINNs optimization, which has a higher requirement for the smoothness and diversity on learned functions.
- 43
- 43
- EfficientTrain: Exploring Generalized Curriculum Learning for Training Visual Backbones42
A new curriculum learning approach for the efficient training of visual backbones, inspired by the inherent learning dynamics of deep networks, that reduces the wall-time training cost of a wide variety of popular models by > 1.5× on ImageNet-1K/22K without sacrificing accuracy.
- Artificial intelligence for stepwise diagnosis and monitoring of COVID-1942
CoviDet can be used as an adjunct to aid clinicians with the CT diagnosis of COVID-19 and can potentially be used for disease monitoring; federated learning can potentially open opportunities for global collaboration.
- 37
- Towards expert-level autonomous carotid ultrasonography with large-scale learning-based robotic system37
It is shown that large-scale deep learning offers a promising pathway toward autonomous, high-precision ultrasonography in clinical practice, and an embodied foundation model using deep learning for intelligent, high-precision ultrasound is built.
- Latency-Aware Unified Dynamic Networks for Efficient Image Recognition37
Latency-Aware Unified Dynamic Networks (LAUDNet) is introduced, a general framework that integrates three fundamental dynamic paradigms–spatially-adaptive computation, layer skipping, and channel skipping–into a single unified formulation and significantly bridges the gap between theoretical and practical efficiency.
- Zero-Shot Generative Model Adaptation via Image-Specific Prompt Learning37
Qualitative and quantitative evaluations on various domains demonstrate that IPL effectively improves the quality and diversity of synthesized images and alleviates the mode collapse.
- AutoLoss-Zero: Searching Loss Functions from Scratch for Generic Tasks36
This paper proposes AutoLoss-Zero, which is a general framework for searching loss functions from scratch for generic tasks, and designs an elementary search space composed only of primitive mathematical operators to accommodate the heterogeneous tasks and evaluation metrics.
- Revisiting Non-Autoregressive Transformers for Efficient Image Synthesis35
This paper identifies the complexities in properly configuring non-autoregressive Transformers strategies and indicates the possible sub-optimality in existing heuristic-driven designs and proposes to go beyond existing methods by directly solving the optimal strategies in an automatic framework.
- EfficientTrain++: Generalized Curriculum Learning for Efficient Visual Backbone Training34
This work reformulate the training curriculum as a soft-selection function, which uncovers progressively more difficult patterns within each example during training, instead of performing easier-to-harder sample selection, and proposes a curriculum where the model always leverages all the training data at every learning stage, with harder patterns gradually introduced as training progresses.
- Learning to Estimate 3-D States of Deformable Linear Objects from Single-Frame Occluded Point Clouds34
Simulation and real-world experimental results demonstrate that the proposed two-branch network architecture can generate globally smooth and locally precise DLO state estimation results even with heavily occluded point clouds, which can be directly applied to real- world robotic manipulation of DLOs in 3-D space.
- Fine-grained few shot learning with foreground object transformation34
A novel method named foreground object transformation (FOT), which is composed of a foreground object extractor and a posture transformation generator, which aims to remove image background and can be competitive with the state-of-the-art methods both in inductive setting and transductive setting.
- Causal Intervention for Human Trajectory Prediction with Cross Attention Mechanism33
This work proposes a Social Environment ADjustment (SEAD) method, to remove the confounding effect of the social environment, which has consistent improvements on ETH-UCY datasets with three baseline models and achieves competitive performances with existing methods.
- 33
- ConvLLaVA: Hierarchical Backbones as Visual Encoder for Large Multimodal Models32
This work proposes ConvLLaVA, which employs ConvNeXt, a hierarchical backbone, as the visual encoder of LMM to replace Vision Transformer, and proposes two critical optimizations that enable ConvLLaVA to support inputs of 1536x1536 resolution generating only 576 visual tokens, capable of handling images of arbitrary aspect ratios.
- Meta-Semi: A Meta-Learning Approach for Semi-Supervised Learning32
A novel meta-learning based SSL algorithm (Meta-Semi) that requires tuning only one additional hyper-parameter, compared with a standard supervised deep learning algorithm, to achieve competitive performance under various conditions of SSL.
- Boosting Offline Reinforcement Learning via Data Rebalancing31
This work proposes a simple yet effective method to boost offline RL algorithms based on the observation that resampling a dataset keeps the distribution support unchanged, which can be implemented with less than 10 lines of code change and adds negligible running time.
- FSD-10: A fine-grained classification dataset for figure skating31
A keyframe based temporal segment network (KTSN) for classification and achieve remarkable performance is proposed by the idea that domain knowledge is of great concern in sports field and is designed to have a large collection of finegrained actions.
- 30
- 30
- 30
- 30
- Detecting Generated Images by Real Images Only29
This paper approaches the generated image detection problem from a new perspective: Start from real images and mapping them to a dense subspace in feature space, so that generated images, regardless of their generative model, are then projected outside the subspace.
- 29
- 29
- Emulating human-like adaptive vision for efficient and flexible machine visual perception28
A deep learning approach that shifts machine vision models from passive to active to mimic human-like perception, AdaptiveNN achieves inference costs that are up to 28-times lower without accuracy loss, while showcasing online-adaptable and interpretable behaviours.
- 28
- 28
- 28
- 27
- DiSparse: Disentangled Sparsification for Multitask Model Compression27
This paper proposes DiSparse, a simple, effective, and first-of-its-kind multitask pruning and sparse training scheme that even observed better per-formance than some dedicated multitask learning methods in several cases despite the high model sparsity enforced by DiS parse.
- 27
- 26
- CheXWorld: Exploring Image World Modeling for Radiograph Representation Learning25
This work develops a unified framework that simultaneously models three aspects of medical knowledge essential for qualified radiologists, including 1) local anatomical structures describing the fine-grained characteristics of local tissues; 2) global anatomical layouts describing the global organization of the human body; and 3) domain variations that encourage CheXWorld to model the transitions across different appearance domains of radiographs.
- Towards Learning Spatially Discriminative Feature Representations25
A novel loss function, termed as CAM-loss, is proposed to constrain the embedded feature maps with the class activation maps (CAMs) which indicate the spatially discriminative regions of an image for particular categories.
- Learning Large Euclidean Margin for Sketch-based Image Retrieval.24
A novel loss function, named Euclidean Margin Softmax (EMS), is proposed that not only minimizes intra-class distances but also maximizes inter- class distances simultaneously simultaneously, enabling us to learn a feature space with high discriminability, leading to highly accurate retrieval.
- SimPro: A Simple Probabilistic Framework Towards Realistic Long-Tailed Semi-Supervised Learning23
This framework, grounded in a probabilistic model, innovatively refines the expectation-maximization (EM) algorithm by explicitly decoupling the modeling of conditional and marginal class distributions, leading to the formulation of a Bayes classifier.
- 23
- Uni-AdaFocus: Spatial-Temporal Dynamic Computation for Video Recognition23
The resulting algorithm, Uni-AdaFocus, establishes a comprehensive framework that seamlessly integrates spatial, temporal, and sample-wise dynamic computation, while it preserves the merits of AdaFocus in terms of efficient end-to-end training and hardware friendliness.
- 23
- 23
- Large scale air pollution prediction with deep convolutional networks23
A large-scale air quality dataset with 393 deployed air quality monitoring stations in a 120 km × 70 km region is collected, which is more than ten times denser than existing AQP datasets and a novel method to handle the data effectively is presented.
- ADDP: Learning General Representations for Image Recognition and Generation with Alternating Denoising Diffusion Process22
This method represents the first successful development of general representations applicable to both generation and dense recognition tasks and can be used to generate diverse high-fidelity images and also demonstrate excellent transfer performance on recognition tasks.
- 22
- EchoWorld: Learning Motion-Aware World Models for Echocardiography Probe Guidance21
EchoWorld is presented, a motion-aware world modeling framework for probe guidance that encodes anatomical knowledge and motion-induced visual dynamics, while effectively leveraging past visual-motion sequences to enhance guidance precision.
- 21
- Hundreds Guide Millions: Adaptive Offline Reinforcement Learning With Expert Guidance21
It is argued that different samples should be treated with different policy constraint intensities, and a novel plug-in approach named guided offline RL (GORL) is proposed, which employs a guiding network to adaptively determine the relative importance of the policy improvement and policy constraint for every sample.
- A privacy-preserving image retrieval scheme based secure kNN, DNA coding and deep hashing21
This paper presents the CBIR solution and completes the relevant software development, and proposes a novel privacy protection algorithm to encrypt images and an integrated deep hash algorithm to extract the high-level features of images.
- 21
- Temporal-Spatial Causal Interpretations for Vision-Based Reinforcement Learning20
A temporal-spatial causal interpretation (TSCI) model is presented to understand the agent’s long-term behavior, which is essential for sequential decision-making and is applicable to recurrent agents and can be used to discover causal features with high efficiency once trained.
- 19
- 19
- 19
- Self-Supervised Discovering of Causal Features: Towards Interpretable Reinforcement Learning19
A self- supervised interpretable framework, which employs a self-supervised interpretables network (SSINet) to discover and locate fine-grained causal features that constitute most evidence for the agent's decisions.
- 19
- Boosting Offline Reinforcement Learning with Action Preference Query18
The main insight is that, compared to online fine-tuning, querying the preferences between pre-collected and learned actions can be equally or even more helpful to the erroneous estimate problem.
- 18
- 18
- Assessing a Single Image in Reference-Guided Image Synthesis18
This paper proposes a general learning-based framework, Reference-guided Image Synthesis Assessment (RISA), to quantitatively evaluate the quality of a single generated image, and introduces an unsupervised contrastive loss to effectively capture the style similarity between a generated image and its reference image.
- 17
- 17
- Domain-Aware SE Network for Sketch-based Image Retrieval with Multiplicative Euclidean Margin Softmax17
A Domain-Aware Squeeze-and-Excitation (DASE) network is presented, which seamlessly incorporates the prior knowledge of sample sketch or photo into SE module and make the SE module capable of emphasizing appropriate channels according to domain signal.
- ViT$^3$: Unlocking Test-Time Training in Vision16
This paper presents a systematic empirical study of TTT designs for visual sequence modeling and develops the Vision Test-Time Training (ViT$^3$) model, a pure TTT architecture that achieves linear complexity and parallelizable computation.
- Multidimensional Refinement Graph Convolutional Network With Robust Decouple Loss for Fine-Grained Skeleton-Based Action Recognition16
This work proposes a flexible attention block called channel-variable spatial–temporal attention (CVSTA) to enhance the discriminative power of spatial–temporal joints and obtain a more compact intraclass feature distribution in graph convolutional networks.
- OStr-DARTS: Differentiable Neural Architecture Search Based on Operation Strength16
It is shown that the degeneration issue can be effectively addressed by using the proposed criterion without any modification of supernet optimization, indicating that the magnitude-based selection method can be a critical reason for the instability of DARTS.
- 16
- Deep Incubation: Training Large Models by Divide-and-Conquering16
Deep Incubation is presented, a novel approach that enables the efficient and effective training of large models by dividing them into smaller sub-modules which can be trained separately and assembled seamlessly.
- 16
- DiveR-CT: Diversity-enhanced Red Teaming Large Language Model Assistants with Relaxing Constraints15
DiveR-CT is introduced, which relaxes conventional constraints on the objective and semantic reward, granting greater freedom for the policy to enhance diversity, and provides an effective and efficient approach to LLM red teaming, accelerating real-world deployment.
- DenseGrounding: Improving Dense Language-Vision Semantics for Ego-Centric 3D Visual Grounding15
This work proposes DenseGrounding, a novel approach designed to address issues by enhancing both visual and textual semantics in egocentric 3D visual grounding and introduces the Hierarchical Scene Semantic Enhancer, which retains dense semantics by capturing fine-grained global scene features and facilitating cross-modal alignment.
- 15
- 15
- Anytime Representation Learning15
Anytime Feature Representations (AFR) is introduced, a novel algorithm that explicitly addresses this trade-off in the data representation rather than in the classifier, allowing conventional classifiers to be turned into test-time cost sensitive anytime classifiers.
- Exploring Text-to-Motion Generation with Human Preference14
It is found that current improvements in text-to-motion generation still rely on datasets requiring expert labelers with motion capture systems, but learning from human preference data does not require motion capture systems; a labeler with no expertise simply compares two generated motions.
- Convolution-Enhanced Evolving Attention Networks14
This work proposes a novel and generic evolving attention mechanism, which directly models the evolution of inter-token relationships through a chain of residual convolutional modules, and is the first work that explicitly models the layer-wise evolution of attention maps.
- Joint optic disc and cup segmentation based on elliptical-like morphological feature and spatial geometry constraint14
Joint optic disc and cup segmentation, which utilizes the elliptical-like morphological features and spatial geometry constraint, could improve the performance of optic disc-and-cup segmentation.
- 14
- 12
- Decoupled Prioritized Resampling for Offline RL12
This article proposes offline decoupled prioritized resampling (ODPR), which designs specialized priority functions for the suboptimal policy constraint issue in offline RL and employs unique decoupled resampling for training stability.
- 12
- Provable General Function Class Representation Learning in Multitask Bandits and MDPs12
This work theoretically validate the benefit of multitask representation learning within general function class for bandits and linear MDP for the first time and conducts experiments to demonstrate the effectiveness of the proposedGFUCB algorithm with neural net representation.
- 12
- Maximin Separation Probability Clustering12
This paper defines a novel metric to evaluate the quality of a clustering labeling, named Minimum Separation Probability (MSP), which is a lower bound of the generalization accuracy of a classifier learnt from the clustering labels.
- Adaptive noise variance identification for probability hypothesis density‐based multi‐target filter by variational Bayesian approximations12
Simulation results show that the proposed filter can accommodate the unknown measurement variances effectively, and improve the estimation accuracy of both the number of targets and their states.
- From ReLU to GeMU: Activation functions in the lens of cone projection11
It is proved that the expressive power of FNNs activated by the proposed GeMU is strictly greater than those activated by ReLU, and that it can outperform various existing activation functions.
- HoVLE: Unleashing the Power of Monolithic Vision-Language Models with Holistic Vision-Language Embedding11
This paper presents a novel high-performance monolithic VLM named HoVLE, which introduces a holistic embedding module that converts visual and textual inputs into a shared space, allowing LLMs to process images in the same way as texts.
- Advancing Generalization in PINNs Through Latent-Space Representations11
This work proposes physics-informed dynamics representation learner (PiDo), a novel physics-informed neural PDE solver designed to generalize effectively across diverse PDE configurations, including varying initial conditions, PDE coefficients, and training-time horizons.
- 11
- 11
- 11
- Fusion layer attention for image-text matching11
This method creates a fusion layer to extract intermediate modes, thus improving the image-text processing results and proposes a concise way to update the loss function that makes it easier for neural networks to handle difficult problems.
- Efficient Lasso training from a geometrical perspective11
This work reformulates the Lasso as a problem of finding the nearest point in a polytope to the origin, which circumvents the L1 norm constraint, and compares with least angle regression (LARS), which is a conventional method to solve Lasso, in both efficiency and numerical stability.
- SiameseNorm: Breaking the Barrier to Reconciling Pre/Post-Norm10
This work proposes SiameseNorm, a simple yet effective two-stream architecture that remains compatible with Pre-Norm training recipes, and shows that SiameseNorm consistently improves performance while maintaining strong training stability across architectures and modalities.
- 10
- 10
- 10
- UniTTA: Unified Benchmark and Versatile Framework Towards Realistic Test-Time Adaptation9
A versatile UniTTA framework is proposed, which includes a Balanced Domain Normalization layer and a COrrelated Feature Adaptation method--designed to mitigate distribution gaps in domain and class, respectively--designed to mitigate distribution gaps in domain and class.
- A Unified Interaction Control Framework for Safe Robotic Ultrasound Scanning with Human-Intention-Aware Compliance9
Experimental results validate the framework’s ability to cope with all common interactions including intended intervention and unintended collision in a collaborative carotid artery ultrasound scanning task.
- Rethinking the Architecture Design for Efficient Generic Event Boundary Detection9
This paper reexamining the architecture of GEBD models and uncovering several surprising findings, including that a concise GEBD base model already achieves promising performance without any sophisticated design and that the widely applied image backbone based GEBDs contain plenty of redundancy.
- Adapting Across Domains via Target-Oriented Transferable Semantic Augmentation Under Prototype Constraint9
A target-oriented transferable semantic augmentation method, which enhances the generalization ability of the classifier by training it with a target-like augmented domain, constructed by semantically augmenting source data towards target at the feature level in an implicit manner.
- A second order cone programming approach for semi-supervised learning9
A novel SSL algorithm based on the multiple clusters per class assumption that is efficient for handling big data sets with a large amount of unlabeled data and guaranteed by the maximal margin classifier is proposed.
- Scaffolded Language Models with Language Supervision for Mixed-Autonomy: A Survey8
A survey of the principles and emerging trends in LLM-based optimization of compound AI systems covers archetypes of compound AI systems, approaches to LLM-based end-to-end optimization, and insights into future directions and broader impacts.
- Multi-Modal Interaction Control of Ultrasound Scanning Robots with Safe Human Guidance and Contact Recovery8
A novel multi-modal control scheme for ultrasound scanning robots, in which three interaction modes are integrated into a single control input, which allows the doctor to intervene safely at any time during the task and also maximizes the robot's autonomous scanning ability.
- Learn From the Past: Experience Ensemble Knowledge Distillation8
The experimental results show that EEKD outperforms the mainstream knowledge distillation methods and achieves the state-of-the-art, and even surpasses the standard ensemble distillation on the premise of saving training cost.
- 8
- 8
- 8
- 8
- Kernelized LARS–LASSO for constructing radial basis function neural networks8
This paper proposes an alternative selection procedure based on the kernelized least angle regression (LARS)–least absolute shrinkage and selection operator (LASSO) method, formulating the RBF neural network as a linear-in-the-parameters model and derive a l1-constrained objective function for training the network.
- ProxyTransformation: Preshaping Point Cloud Manifold With Proxy Attention For 3D Visual Grounding7
This work proposes Proxy Transformation suitable for multimodal task to efficiently improve the point cloud manifold, and establishes a new SOTA in ego-centric 3D visual grounding, showcasing the effectiveness and robustness of the approach.
- InfoPro: Locally Supervised Deep Learning by Maximizing Information Propagation7
It is shown that the reliance on back-propagating global errors may not be necessary for deep learning, and an information propagation (InfoPro) loss is proposed, which encourages local modules to preserve as much useful information as possible, while progressively discarding task-irrelevant information.
- 7
- Robust Outlier Bound Condition to Phase Retrieval with Adversarial Sparse Outliers7
It is shown that the Robust-PhaseLift model can successfully reconstruct the ground-truth up to global phase for any $s<s^{*}\approx 0.1185$ with $\mathcal{O}(n)$ measurements, even in the case where the sparse outliers may depend on the measurement and the observation.
- UltraSeP: Sequence-aware pre-training for echocardiography probe movement guidance6
A novel probe movement guidance algorithm that has the potential to be applied in guiding robotic systems or novices with probe pose adjustment for high-quality standard plane image acquisition and can effectively reduce probe guidance errors compared to other advanced baseline methods.
- 6
- 6
- Leveraging Reward Consistency for Interpretable Feature Discovery in Reinforcement Learning6
It is argued that the commonly used action matching principle is more like an explanation of deep neural networks (DNNs) than the interpretation of RL agents, and proposed to consider rewards, the essential objective of RL agents, as the essential objective of interpreting RL agents as well.
- Collaborative learning with corrupted labels6
This paper proposes a collaborative learning (co-learning) approach to improve the robustness and generalization performance of DNNs on datasets with corrupted labels by designing a deep network with two separate branches, coupled with a relabeling mechanism.
- Transductive Minimax Probability Machine6
It is shown that the proposed Transductive MPM (TMPM) almost outperforms all the other algorithms in both accuracy and speed.
- AdaGen: Learning Adaptive Policy for Image Synthesis5
AdaGen, a general, learnable, and sample-adaptive framework for scheduling the iterative generation process, is presented and an adversarial reward design is proposed to guide the training of the policy networks effectively.
- UltraDP: Generalizable Carotid Ultrasound Scanning with Force-Aware Diffusion Policy5
UltraDP, a Diffusion-Policy-based method that receives multi-sensory inputs and generates actions that are fit for multi-modal action distributions in autonomous ultrasound scanning of carotid artery achieves a 95% success rate in transverse scanning on previously unseen subjects, demonstrating its effectiveness.
- 5
- Exploring Temporal Feature Correlation for Efficient and Stable Video Semantic Segmentation5
This work addresses the problem of efficient and stable video semantic segmentation using a novel yet simple Temporal Feature Correlation (TFC) module that selectively reuse label-consistent features across frames through linear interpolation and update others through sparse multi-scale deformable attention.
- Training an Open-Vocabulary Monocular 3D Object Detection Model without 3D Data5
A novel open-vocabulary monocular 3D object detection framework, dubbed OVM3D-Det, which trains detectors using only RGB images, making it both cost-effective and scalable to publicly available data.
- A coupled-lines system to determine the anteroposterior position of maxillary central incisors for smiling profile esthetics5
Both upper- and mid-facial soft tissue landmarks were indispensable in assessing anteroposterior positions of U1 for well-balanced smiling profiles, and the most esthetically sensitive parameters were mFA-Glabella and m FA-SSn distances.
- 5
- 5
- 5
- 5
- Multiset Canonical Correlation Analysis Using for Blind Source Separation5
A novel algorithm based on multiset canonical correlation analysis, based on second order statistical characteristic of observation signals, can blind separate super- Gaussian and sub-Gaussian signals successfully at the same time with relatively light computation burden.
- UltraHiT: A Hierarchical Transformer Architecture for Generalizable Internal Carotid Artery Robotic Ultrasonography4
This work proposes a Hierarchical Transformer-based decision architecture, namely UltraHiT, that integrates high-level variation assessment with low-level action decision and achieves a 95% success rate in locating the ICA on unseen individuals, outperforming baselines and demonstrating its effectiveness.
- End-to-End Streaming Video Temporal Action Segmentation With Reinforcement Learning4
Through extensive experiments, the SVTAS-RL model significantly outperforms existing STAS models and achieves competitive performance to the state-of-the-art (SOTA) TAS model on multiple datasets under the same evaluation criteria, demonstrating notable advantages on the ultralong video dataset EGTEA.
- A-SDM: Accelerating Stable Diffusion Through Model Assembly and Feature Inheritance Strategies4
This study designs a model assembly strategy to reconstruct a lightweight model to reconstruct a lightweight model while preserving performance and ensuring semantic stability through distillation and proposes a feature inheritance strategy to accelerate inference by skipping local computations at the block, layer, or unit level within the network structure.
- 4
- Anatomical prior-based vertebral landmark detection for spinal disorder diagnosis4
This paper systematically utilizes anatomical prior knowledge in vertebral landmark detection to formulate anatomical priors of the spine, related to distances among vertebrae and spatial order within the spine, and integrate these geometrical constraints within training loss, inference procedure, and evaluation metrics.
- 4
- 4
- 4
- 4
- Increased frequency of angiotensin converting enzyme D allele in Chinese Han patients with idiopathic pulmonary fibrosis: A systematic review and meta-analysis4
In the Chinese Han population, the D allele of the ACE I/D gene polymorphism is associated with the susceptibility of IPF, and the angiotensin II (Ang Ⅱ) level ofIPF patients was higher than that of the control group.
- 4
- 4
- MOVE: A Simple Motion-Based Data Collection Paradigm for Spatial Generalization in Robotic Manipulation3
This work proposes MOtion-Based Variability Enhancement (MOVE), a simple yet effective data collection paradigm that enables the acquisition of richer spatial information from dynamic demonstrations by injecting motion into any movable objects within the environment for each demonstration.
- 3
- 2023 Low-Power Computer Vision Challenge (LPCVC) Summary3
The setup of the LPCVC is explained, the winners' methods that improve accuracy and shorten execution time are highlighted and the winners' methods that improve accuracy and shorten execution time are highlighted.
- 3
- 3
- 3
- 3
- 3
- Loss Function Optimization Method and Unsupervised Extraction Approach D-DBSCAN for Improving the Moving Target Perception of 3D Imaging Sonar2
3D imaging sonar is utilizes to collect underwater data from three types of targets with 1534 effective frames, including a tire, mannequin, and table, in Liquan Lake, Shanxi Province, China, to focus on three innovative aspects as follows: rapid underwater data annotation, loss function optimization, and unsupervised moving target extraction in water.
- 2
- 2
- 2
- The lower bow-shaped curve as a novel reference frame to determine the lateral limit of the maxillary anterior arch for smile esthetics2
The LBSC could serve as a reference frame to determine the lateral limit of the maxillary anterior arch for the esthetics of 45° angled smiles, and the FA-TL distance, which represented the spatial relation of themaxillary canine with the lower lip, was an esthetically essential parameter.
- 2
- 2
- 2
- 1
- Immune-related long noncoding RNAs in predicting the prognosis and immune landscape of intrahepatic cholangiocarcinoma: a bioinformatics analysis with experimental verification1
The prognostic signature consisting of irlncRNAs could predict the prognosis and immune landscape of patients with ICC and may provide a novel perspective for the individualized intervention in ICC.
- 1
- 1
- 1
- Exploring contextual modeling with linear complexity for point cloud segmentation1
It is shown that: 1) Spatial locality and robust contextual understanding are critical for strong performance, and 2) Mamba features linear computational complexity, offering superior data and inference efficiency compared to Transformers, while still being capable of delivering strong contextual understanding.
- A-SDM: Accelerating Stable Diffusion through Redundancy Removal and Performance Optimization1
This work first explores the computational redundancy part of the network, and then prune the redundancy blocks of the model and maintain the network performance through a progressive incubation strategy, and proposes a global-regional interactive (GRI) attention to speed up the computationally intensive attention part.
- 1
- Augmenting Unsupervised Reinforcement Learning with Self-Reference1
The Self-Reference (SR) approach is proposed, an add-on module explicitly designed to leverage historical information and enhance agent performance within the pretrain-finetune paradigm that achieves state-of-the-art results in terms of Interquartile Mean (IQM) performance and Optimality Gap reduction on the Unsupervised Reinforcement Learning Benchmark for model-free methods.
- 1
- 1
- 1
- 1
- –
- –
- –
- Predictive value of current nodal staging systems and development of machine learning nomogram for resectable pancreatic head cancer: a population-based study and multicenter validation–
The nodal staging system LODDS is the optimal prognostic factor for OS in resectable PHC and could effectively predict OS for resectable PHC patients without considering the numbers of RLN, indicating the machine learning model had good predictive capability and clinical utility.
- BAPE: Learning an Explicit Bayes Classifier for Long-tailed Visual Recognition–
This paper presents a novel approach (BAPE) that provides a more precise theoretical estimation of the data distributions by modeling the parameters of the posterior probabilities and solving them with point estimation, and directly learns the Bayes classifier without gradient descent based on Bayes'theorem.
- CC-FMO: Camera-Conditioned Zero-Shot Single Image to 3D Scene Generation with Foundation Model Orchestration–
CC-FMO is introduced, a zero-shot, camera-conditioned pipeline for single-image to 3D scene generation that jointly conforms to the object layout in input image and preserves instance fidelity, and enables the application of foundational pose estimation models in the scene generation task via a simple yet effective camera-conditioned scale-solving algorithm, to enforce scene-level coherence.
- –
- –
- –
- Generalized Activation via Multivariate Projection–
This work extends ReLU by substituting it with a generalized projection operator onto a convex cone, such as the Second-Order Cone (SOC) projection, thereby naturally extending it to a Multivariate Projection Unit (MPU), an activation function with multiple inputs and multiple outputs.
- Preface–
This special section is launched and aims to seek high quality research work on self-learning with deep neural networks, including a comprehensive review of selfsupervised learning, improving the generalization performance of few-shot intent detection, and conducting musicdriven motion generation.
- –
- –
- –
- –
- Stepwise Diagnosis and Dynamic Change Prediction AI System for COVID-19–
CoviDet is a useful aid for radiologist to diagnose COVID-19 rapidly and accurately based on CT, with or without clinical data, and could be used on a web-based platform with built auto masking of privacy data, whereby clinicians can upload anonymised CT for CO VID-19 analysis without sensitive clinical information from anywhere in the world.
- Learning decomposed subspaces for supervised bidirectional image generation–
A framework to learn a feature space, which breaks into three parts, namely, two personalised semantic subspaces, which encode patterns that are unique for each domain, and a shared semantic subspace that captures the common patterns.
- –
- Semi-Blind Channel Estimation with Orthogonal Superimposed Training for SISO over Doubly Selective Channel–
A semi-blind estimator based on orthogonal superimposed training (OST) for the case of doubly selecting fading channel with the aid of OST can be eliminated completely and data symbols can serve as pseudo-pilots to enhance the estimation performance.
- –
- Spacial Filter of Weighting Method Based on Spectrum Analyse–
A method based on a weighting method to filter the signal in certain directions, namely spacial filter, is proposed, which can be obtained through the analysis of spacial signal and spacial spectrum, and then obtains an appropriate weighting window function.
- Performance Analysis of Radar and ESM Track Association Algorithms–
The simulation results show that both of these two track association algorithms have high probability of correct association, but the algorithm based on statistical theory has faster convergent speed and litter samples, while the other is simple, and has less calculation.
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 2026-10-14. 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.