Is this you? Claim this profile to correct it, add a bio and choose the work people see first.
Claim this profileWorks62 from public data
- LLaMA-Adapter: Efficient Fine-tuning of Language Models with Zero-init Attention1,068
A zero-initialized attention mechanism with zero gating is proposed, which adaptively injects the new instructional cues into LLaMA, while effectively preserves its pre-trained knowledge on traditional vision and language tasks, demonstrating the superior generalization capacity of the approach.
- LLaMA-Adapter V2: Parameter-Efficient Visual Instruction Model774
This work augments LLaMA-Adapter by unlocking more learnable parameters and proposes an early fusion strategy to feed visual tokens only into the early LLM layers, contributing to better visual knowledge incorporation and achieves strong multi-modal reasoning with only a small-scale image-text and instruction dataset.
- Incorporating Convolution Designs into Visual Transformers613
A new Convolution-enhanced image Transformer (CeiT) is proposed which combines the advantages of CNNs in extracting low-level features, strengthening locality, and the disadvantages of Transformers in establishing long-range dependencies.
- Omni-Dimensional Dynamic Convolution479
Intriguingly, thanks to its improved feature learning ability, ODConv with even one single kernel can compete with or outperform existing dynamic convolution counterparts with multiple kernels, substantially reducing extra parameters.
- 439
- MathCoder: Seamless Code Integration in LLMs for Enhanced Mathematical Reasoning220
This paper proposes a method to fine-tune open-source language models, enabling them to use code for modeling and deriving math equations and, consequently, enhancing their mathematical reasoning abilities, and yields the MathCoder models, a family of models capable of generating code-based solutions for solving challenging math problems.
- Solving Challenging Math Word Problems Using GPT-4 Code Interpreter with Code-based Self-Verification219
The effect of code on enhancing LLMs' reasoning capability by introducing different constraints on the Code Usage Frequency of GPT-4 Code Interpreter is explored, and a novel and effective prompting method, explicit \uline{c}ode-based \ULine{s}elf-\uline {v}erification~(CSV), is proposed to further boost the mathematical reasoning potential of GPN.
- Adversarial Robustness vs. Model Compression, or Both?186
It is found that weight pruning is essential for reducing the network model size in the adversarial setting; training a small model from scratch even with inherited initialization from the large model cannot achieve neither adversarial robustness nor high standard accuracy.
- Not All Features Matter: Enhancing Few-shot CLIP with Adaptive Prior Refinement143
APE, an Adaptive Prior rEfinement method for CLIP’s pre-trained knowledge, which achieves superior accuracy with high computational efficiency and introduces two model variants, a training-free APE and a training-required APE-T.
- Deep Neural Network Compression With Single and Multiple Level Quantization125
This paper is the first to consider the network quantization from both width and depth level, and introduces incremental layer compensation to quantize layers iteratively which decreases the quantization loss in each iteration.
- An Algorithm–Hardware Co-Optimized Framework for Accelerating N:M Sparse Transformers102
A sparsity inheritance mechanism along with inherited dynamic pruning (IDP) to obtain a series of N:M sparse candidate Transformers rapidly and a model compression scheme is further proposed to significantly reduce the storage requirement for deployment.
- RecRanker: Instruction Tuning Large Language Model as Ranker for Top-k Recommendation101
This article introduces Ranker for top-k Recommendations (RecRanker), tailored for instruction tuning LLMs to serve as the Ranker for top-k Recommendations, and proposes a hybrid ranking method to enhance the model performance by ensembling these ranking tasks.
- 94
- ThinK: Thinner Key Cache by Query-Driven Pruning84
ThinK is proposed, a novel query-dependent KV cache pruning method designed to minimize attention weight loss while selectively pruning the least significant channels, which achieves a reduction in KV cache memory costs by over 20% compared with vanilla KV cache eviction and quantization methods.
- Deeply-Supervised Knowledge Synergy84
Deeply-supervised Knowledge Synergy is proposed, a new method aiming to train CNNs with improved generalization ability for image classification tasks without introducing extra computational cost during inference, and a novel synergy loss, which considers pairwise knowledge matching among all supervision branches.
- 73
- 71
- Integrating Large Language Models Into Recommendation via Mutual Augmentation and Adaptive Aggregation62
A general and model-agnostic framework known as Large language models with mutual augmentation and adaptive aggregation for Recommendation (Llama4Rec) is introduced, aiming to bridge the gap via explicitly ensemble LLM and conventional recommendation model for more effective recommendation.
- Group R-CNN for Weakly Semi-supervised Object Detection with Points56
Group R-CNN significantly outperforms the prior method Point DETR by 3.9 mAP with 5% well-labeled images, which is the most challenging scenario, and is also the most effective method on the MS-COCO benchmark.
- HBONet: Harmonious Bottleneck on Two Orthogonal Dimensions43
Harmonious Bottleneck on two Orthogonal dimensions (HBO), a novel architecture unit, specially tailored to boost the accuracy of extremely lightweight MobileNets at the level of less than 40 MFLOPs, is presented.
- Step-Controlled DPO: Leveraging Stepwise Error for Enhanced Mathematical Reasoning42
This work proposes Step-Controlled DPO (SCDPO), a method for automatically providing stepwise error supervision by creating negative samples of mathematical reasoning rationales that start making errors at a specified step, which can better align the model to understand reasoning errors and output accurate reasoning steps.
- MathCoder2: Better Math Reasoning from Continued Pretraining on Model-translated Mathematical Code41
A novel method for generating mathematical code accompanied with corresponding reasoning steps for continued pretraining, leading to a 19.2B-token high-performing mathematical pretraining corpus, which is training several popular base models with this corpus significantly improves their mathematical abilities, leading to the creation of the MathCoder2 family of models.
- Pyramid Fusion Transformer for Semantic Segmentation39
This study finds that the per-mask classification decoder on top of a single-scale feature is not effective enough to extract reliable probabilities or masks, and proposes a transformer-based Pyramid Fusion Transformer (PFT) for per-mask approach semantic segmentation with multi-scale features.
- BlueLM-V-3B: Algorithm and System Co-Design for Multimodal Large Language Models on Mobile Devices31
This paper redesigns the dynamic resolution scheme adopted by mainstream MLLMs and implements system optimization for hardware-aware deployment to optimize model inference on mobile phones.
- 29
- FeatAug-DETR: Enriching One-to-Many Matching for DETRs With Feature Augmentation28
This paper proposes two methods that realize one-to-many matching from a different perspective of augmenting images or image features and significantly accelerates DETR training and boosts detection performance while keeping the inference speed unchanged.
- 26
- Efficient N:M Sparse DNN Training Using Algorithm, Architecture, and Dataflow Co-Design25
A bidirectional weight pruning method, dubbed BDWP, is proposed to leverage the N:M sparsity of weights during both forward and backward passes of DNN training, which can significantly reduce the computational cost while maintaining model accuracy.
- 24
- MAVIS: Mathematical Visual Instruction Tuning with an Automatic Data Engine22
This paper proposes MAVIS, the first MA thematical VIS ual instruction tuning paradigm for MLLMs, involving a series of mathematical visual datasets and specialized MLLMs, and introduces MAVIS-Instruct, which is adopted to instruct-tune the MLLM for robust mathematical reasoning skills.
- 22
- 22
- 18
- SPP: Sparsity-Preserved Parameter-Efficient Fine-Tuning for Large Language Models17
Results show that SPP significantly enhances the performance of models with different sparsity patterns, especially for those with high sparsity ratios, making it a promising solution for the efficient fine-tuning of sparse LLMs.
- 14
- Chimera: Improving Generalist Model with Domain-Specific Experts11
Chimera, a scalable and low-cost multi-modal pipeline designed to boost the ability of existing LMMs with domain-specific experts with a novel Generalist-Specialist Collaboration Masking mechanism, results in a versatile model that excels across the chart, table, math, and document domains.
- 10
- 9
- Adaptive Markup Language Generation for Contextually-Grounded Visual Document Understanding8
This work proposes an innovative pipeline that utilizes adaptive generation of markup languages, such as Markdown, JSON, HTML, and TiKZ, to build highly structured document representations and deliver contextually-grounded responses, facilitating advanced reasoning and comprehension capabilities in complex visual scenarios.
- 8
- TerDiT: Ternary Diffusion Models with Transformers7
TerDiT, the first quantization-aware training (QAT) and efficient deployment scheme for extremely low-bit diffusion transformer models, is proposed, demonstrating the feasibility of training extremely low-bit DiT models from scratch while maintaining competitive image generation capacities compared to full-precision models.
- Second Rethinking of Network Pruning in the Adversarial Setting.7
A framework of concurrent adversarialTraining and weight pruning that enables model compression while still preserving the adversarial robustness and essentially tackles the dilemma of adversarial training is proposed.
- 6
- 6
- From Solver to Tutor: Evaluating the Pedagogical Intelligence of LLMs with KMP-Bench5
KMP-Bench, a comprehensive K-8 Mathematical Pedagogical Benchmark designed to assess LLMs from two complementary perspectives, is introduced, underscoring the value of pedagogically-rich training data for developing more effective AI math tutors.
- GenieBlue: Integrating Both Linguistic and Multimodal Capabilities for Large Language Models on Mobile Devices4
GenieBlue, an efficient MLLM structural design that integrates both linguistic and multimodal capabilities for LLMs on mobile devices, and freezes the original LLM parameters during MLLM training to maintain pure language capabilities.
- NODI: Out-Of-Distribution Detection with Noise from Diffusion4
This work introduces the diffusion process into the OOD task, and outperforms previous OOD methods across all types of image encoders, and exhibits good robustness with all the image encoders.
- 3
- 3
- NOAH: Learning Pairwise Object Category Attentions for Image Classification3
This paper revisits the feature encoding problem, and proposes Non-glObal Attentive Head (NOAH) that relies on a new form of dot-product attention called pairwise object category attention (POCA), efficiently exploiting spatially dense category-specific attentions to augment classification performance.
- Towards Improving Generalization of Deep Networks via Consistent Normalization.3
This work proposes Consistent Normalization (CN) that not only retains the advantages of the existing normalization methods, but also achieves state-of-the-art performance on various tasks including image classification, segmentation, and machine translation.
- 2
- 2
- 1
- 1
- 1
- Reflect to Inform: Boosting Multimodal Reasoning via Information-Gain-Driven Verification–
This work proposes Visual Re-Examination (VRE), a self-evolving training framework that enables MLLMs to autonomously perform visual introspection during reasoning without additional visual inputs, and consistently improves reasoning accuracy and perceptual reliability, while substantially reducing hallucinations, especially in long-chain settings.
- –
- –
- –
- –
- –
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.