{"total":74,"rows":[{"id":"2505.06708","title":"Gated Attention for Large Language Models: Non-linearity, Sparsity, and Attention-Sink-Free","created":"2025-05-10","cited_by_count":283,"area":"moe"},{"id":"2410.10781","title":"When Attention Sink Emerges in Language Models: An Empirical View","created":"2024-10-14","cited_by_count":208,"area":"inference"},{"id":"2503.03321","title":"See What You Are Told: Visual Attention Sink in Large Multimodal Models","created":"2025-03-05","cited_by_count":132,"area":"mllm"},{"id":"2407.08454","title":"Model Tells You Where to Merge: Adaptive KV Cache Merging for LLMs on Long-Context Tasks","created":"2024-07-11","cited_by_count":96,"area":"inference"},{"id":"2503.01773","title":"Why Is Spatial Reasoning Hard for VLMs? An Attention Mechanism Perspective on Focus Areas","created":"2025-03-03","cited_by_count":95,"area":"lr"},{"id":"2410.13835","title":"Active-Dormant Attention Heads: Mechanistically Demystifying Extreme-Token Phenomena in LLMs","created":"2024-10-17","cited_by_count":56,"area":null},{"id":"2411.09968","title":"Seeing Clearly by Layer Two: Enhancing Attention Heads to Alleviate Hallucination in LVLMs","created":"2024-11-15","cited_by_count":52,"area":"mllm"},{"id":"2506.03524","title":"Seed-Coder: Let the Code Model Curate Data for Itself","created":"2025-06-04","cited_by_count":49,"area":null},{"id":"2501.16383","title":"RotateKV: Accurate and Robust 2-Bit KV Cache Quantization for LLMs via Outlier-Aware Adaptive Rotations","created":"2025-01-25","cited_by_count":46,"area":"inference"},{"id":"2508.02546","title":"What are you sinking? A geometric approach on attention sink","created":"2025-08-04","cited_by_count":28,"area":null},{"id":"2603.05498","title":"The Spike, the Sparse and the Sink: Anatomy of Massive Activations and Attention Sinks","created":"2026-03-05","cited_by_count":27,"area":null},{"id":"2601.22966","title":"A Unified View of Attention and Residual Sinks: Outlier-Driven Rescaling is Essential for Transformer Training","created":"2026-01-30","cited_by_count":23,"area":null},{"id":"2510.08510","title":"To Sink or Not to Sink: Visual Information Pathways in Large Vision-Language Models","created":"2025-10-09","cited_by_count":20,"area":"mllm"},{"id":"2502.00919","title":"Attention Sinks: A 'Catch, Tag, Release' Mechanism for Embeddings","created":"2025-02-02","cited_by_count":16,"area":null},{"id":"2503.08908","title":"Interpreting the Repeated Token Phenomenon in Large Language Models","created":"2025-03-11","cited_by_count":16,"area":null},{"id":"2501.15269","title":"Mirage in the Eyes: Hallucination Attack on Multi-modal Large Language Models with Only Attention Sink","created":"2025-01-25","cited_by_count":15,"area":"mllm"},{"id":"2502.14317","title":"ParallelComp: Parallel Long-Context Compressor for Length Extrapolation","created":"2025-02-20","cited_by_count":14,"area":"inference"},{"id":"2505.11739","title":"ZeroTuning: Unlocking the Initial Token's Power to Enhance Large Language Models Without Training","created":"2025-05-16","cited_by_count":12,"area":null},{"id":"2510.22603","title":"Mitigating Attention Sinks and Massive Activations in Audio-Visual Speech Recognition with LLMs","created":"2025-10-26","cited_by_count":11,"area":null},{"id":"2603.11487","title":"Attention Sinks Are Provably Necessary in Softmax Transformers: Evidence from Trigger-Conditional Tasks","created":"2026-03-12","cited_by_count":11,"area":null},{"id":"2512.08417","title":"Attention is All You Need to Defend Against Indirect Prompt Injection Attacks in LLMs","created":"2025-12-09","cited_by_count":10,"area":"agents"},{"id":"2410.01866","title":"House of Cards: Massive Weights in LLMs","created":"2024-10-02","cited_by_count":9,"area":null},{"id":"2503.13208","title":"Improving Complex Reasoning with Dynamic Prompt Corruption: A soft prompt Optimization Approach","created":"2025-03-17","cited_by_count":9,"area":"lr"},{"id":"2505.15865","title":"How Do Large Vision-Language Models See Text in Image? Unveiling the Distinctive Role of OCR Heads","created":"2025-05-21","cited_by_count":9,"area":"mllm"},{"id":"2510.10276","title":"Lost in the Middle: An Emergent Property from Information Retrieval Demands in LLMs","created":"2025-10-11","cited_by_count":9,"area":null},{"id":"2601.20279","title":"Hallucination Begins Where Saliency Drops","created":"2026-01-28","cited_by_count":9,"area":"mllm"},{"id":"2504.03889","title":"Identifying and Evaluating Inactive Heads in Pretrained LLMs","created":"2025-04-04","cited_by_count":8,"area":null},{"id":"2505.02130","title":"Attention Mechanisms Perspective: Exploring LLM Processing of Graph-Structured Data","created":"2025-05-04","cited_by_count":8,"area":null},{"id":"2507.03865","title":"OrthoRank: Token Selection via Sink Token Orthogonality for Efficient LLM inference","created":"2025-07-05","cited_by_count":8,"area":"inference"},{"id":"2510.09017","title":"Value-State Gated Attention for Mitigating Extreme-Token Phenomena in Transformers","created":"2025-10-10","cited_by_count":8,"area":null},{"id":"2510.16670","title":"All You Need is One: Capsule Prompt Tuning with a Single Vector","created":"2025-10-19","cited_by_count":8,"area":null},{"id":"2504.09456","title":"Don't Deceive Me: Mitigating Gaslighting through Attention Reallocation in LMMs","created":"2025-04-13","cited_by_count":7,"area":"mllm"},{"id":"2601.20520","title":"Context Tokens are Anchors: Understanding the Repetition Curse in dMLLMs from an Information Flow Perspective","created":"2026-01-28","cited_by_count":6,"area":"mllm"},{"id":"2603.14337","title":"On the Nature of Attention Sink that Shapes Decoding Strategy in Omni-LLMs","created":"2026-03-15","cited_by_count":6,"area":"lr"},{"id":"2601.21709","title":"Why Attention Patterns Exist: A Unifying Temporal Perspective Analysis","created":"2026-01-29","cited_by_count":5,"area":"inference"},{"id":"2602.01203","title":"Attention Sink Forges Native MoE in Attention Layers: Sink-Aware Training to Address Head Collapse","created":"2026-02-01","cited_by_count":5,"area":"moe"},{"id":"2412.08521","title":"EMS: Adaptive Evict-then-Merge Strategy for Head-wise KV Cache Compression Based on Global-Local Importance","created":"2024-12-11","cited_by_count":4,"area":"inference"},{"id":"2508.03668","title":"CTR-Sink: Attention Sink for Language Models in Click-Through Rate Prediction","created":"2025-08-05","cited_by_count":4,"area":null},{"id":"2508.04204","title":"ReasoningGuard: Safeguarding Large Reasoning Models with Inference-time Safety Aha Moments","created":"2025-08-06","cited_by_count":4,"area":"lr"},{"id":"2512.22213","title":"On the Existence and Behavior of Secondary Attention Sinks","created":"2025-12-22","cited_by_count":4,"area":null},{"id":"2602.05228","title":"Surgery: Mitigating Harmful Fine-Tuning for Large Language Models via Attention Sink","created":"2026-02-05","cited_by_count":4,"area":"alignment"},{"id":"2604.03316","title":"When Sinks Help or Hurt: Unified Framework for Attention Sink in Large Vision-Language Models","created":"2026-04-01","cited_by_count":4,"area":"mllm"},{"id":"2410.13060","title":"AERO: Entropy-Guided Framework for Private LLM Inference","created":"2024-10-16","cited_by_count":3,"area":null},{"id":"2503.08963","title":"Gradient-guided Attention Map Editing: Towards Efficient Contextual Hallucination Mitigation","created":"2025-03-11","cited_by_count":3,"area":null},{"id":"2507.10085","title":"Enhancing Chain-of-Thought Reasoning with Critical Representation Fine-tuning","created":"2025-07-14","cited_by_count":3,"area":"lr"},{"id":"2508.18387","title":"Integral Transformer: Denoising Attention, Not Too Much Not Too Little","created":"2025-08-25","cited_by_count":3,"area":null},{"id":"2603.06591","title":"What Makes Position Zero Special? A Mechanistic Study of Position Zero Attention Sinks in LLMs","created":"2026-02-04","cited_by_count":3,"area":null},{"id":"2605.29657","title":"OccamToken: Efficient VLM Inference with Training-Free and Budget-Adaptive Token Pruning","created":"2026-05-28","cited_by_count":3,"area":"mllm"},{"id":"2408.00244","title":"Enhanced Structured State Space Models via Grouped FIR Filtering and Attention Sink Mechanisms","created":"2024-08-01","cited_by_count":2,"area":"ssm"},{"id":"2603.06248","title":"Gradient Flow Polarizes Softmax Outputs towards Low-Entropy Solutions","created":"2026-03-06","cited_by_count":2,"area":null},{"id":"2603.27898","title":"SAGE: Sink-Aware Grounded Decoding for Multimodal Hallucination Mitigation","created":"2026-03-29","cited_by_count":2,"area":"mllm"},{"id":"2604.04500","title":"Saliency-R1: Enforcing Interpretable and Faithful Vision-language Reasoning via Saliency-map Alignment Reward","created":"2026-04-06","cited_by_count":2,"area":"lr"},{"id":"2604.10027","title":"SinkTrack: Attention Sink based Context Anchoring for Large Language Models","created":"2026-04-11","cited_by_count":2,"area":null},{"id":"2606.02288","title":"Massive Spikes in LLMs are Bias Vectors: Mechanistic Uncovering and Spike-Free Quantization","created":"2026-06-01","cited_by_count":2,"area":null},{"id":"2607.00712","title":"Towards Memory-Efficient Autoregressive Video Generation via Instance-Specific Parametric Absorption","created":"2026-07-01","cited_by_count":2,"area":"inference"},{"id":"2505.08739","title":"Probability Consistency in Large Language Models: Theoretical Foundations Meet Empirical Discrepancies","created":"2025-05-13","cited_by_count":1,"area":null},{"id":"2512.13109","title":"Uncovering the Role of Initial Saliency in U-Shaped Attention Bias: Scaling Initial Token Weight for Enhanced Long-Text Processing","created":"2025-12-15","cited_by_count":1,"area":null},{"id":"2604.01280","title":"Look Twice: Training-Free Evidence Highlighting for Knowledge-based Visual Question Answering","created":"2026-04-01","cited_by_count":1,"area":"mllm"},{"id":"2604.06393","title":"ART: Attention Replacement Technique to Improve Factuality in LLMs","created":"2026-04-07","cited_by_count":1,"area":null},{"id":"2605.08453","title":"Sink vs. diagonal patterns as mechanisms for attention switch and oversmoothing prevention","created":"2026-05-08","cited_by_count":1,"area":null},{"id":"2605.10050","title":"EchoPrune: Interpreting Redundancy as Temporal Echoes for Efficient VideoLLMs","created":"2026-05-11","cited_by_count":1,"area":"inference"},{"id":"2605.21954","title":"MLLMs Know When Before Speaking: Revealing and Recovering Temporal Grounding via Attention Cues","created":"2026-05-21","cited_by_count":1,"area":"mllm"},{"id":"2410.12001","title":"Impacts of Continued Legal Pre-Training and IFT on LLMs' Latent Representations of Human-Defined Legal Concepts","created":"2024-10-15","cited_by_count":0,"area":null},{"id":"2507.00537","title":"Not All Attention Heads Are What You Need: Refining CLIP's Image Representation with Attention Ablation","created":"2025-07-01","cited_by_count":0,"area":null},{"id":"2512.20661","title":"Calibrating Transformer Attention via Task-Space Sensitivity Feedback","created":"2025-12-19","cited_by_count":0,"area":null},{"id":"2602.16500","title":"Optimizing Soft Prompt Tuning via Structural Evolution","created":"2026-02-18","cited_by_count":0,"area":null},{"id":"2604.14722","title":"A Mechanistic Account of Attention Sinks in GPT-2: One Circuit, Broader Implications for Mitigation","created":"2026-04-16","cited_by_count":0,"area":null},{"id":"2605.10503","title":"SLASH the Sink: Sharpening Structural Attention Inside LLMs","created":"2026-05-11","cited_by_count":0,"area":null},{"id":"2606.17945","title":"Small Initialization Matters for Large Language Models","created":"2026-06-16","cited_by_count":0,"area":null},{"id":"2607.13395","title":"Self-Improving is Often Sudden: Enlightenment-style Finetuning for Large-Scale Models","created":"2026-07-15","cited_by_count":0,"area":"synth"},{"id":"2607.24017","title":"Disentangling Semantic Attention from Structural Bias in the Attention Manifold","created":"2026-07-27","cited_by_count":0,"area":"mllm"},{"id":"2610.07661","title":"Massive Activation Gating Channel in Large Language Models","created":"2026-10-06","cited_by_count":0,"area":null},{"id":"2610.13337","title":"Exploring Attention Sinks in Text-to-Image Transformers from Diffusion and Autoregressive Models","created":"2026-10-09","cited_by_count":0,"area":null},{"id":"2610.16745","title":"Before Attention Sinks Form: Hidden-State Geometry Reveals and Modulates Their Formation","created":"2026-10-12","cited_by_count":0,"area":null}],"next_cursor":null}