pair_id string | source string | ai_paper_id string | human_paper_id string | ai_title string | human_title string | human_arxiv string | human_venue string | human_v1_year int64 | human_iclr_rating float64 | contribution_type string | human_contribution_type string | same_contribution_type bool | pool string | sim_tier int64 | relation string | tier_reason string | H1_human string | H1_by string | H2_top_tier string | H3_type_match string | H4_source string | specter2_cosine float64 | tfidf_similarity float64 | match_rank int64 | human_pangram_ai_fraction float64 | ai_body_words int64 | human_body_words int64 | body_length_ratio float64 | topic string | secondary_topic string | a4s_submission_id string | a4s_openreview_url string | a4s_decision string | a4s_reviewer_scores string | a4s_ai_involvement string | ai_tex_source string |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
FA0001 | fars | AI_FA0001 | HU_2405.16833 | Canary-Controlled Safe-Data Interleaving for Reducing Emergent Misalignment | Safe LoRA: the Silver Lining of Reducing Safety Risks when Fine-tuning Large Language Models | 2405.16833 | NeurIPS 2024 | 2,024 | null | method | method | true | cited | 3 | cited | null | pass | year<=2024 | pass | pass | pass | 0.9546 | null | 1 | null | 2,109 | 4,249 | 2.01 | Safety, security, and privacy | null | null | null | null | null | null | shipped_tex |
FA0002 | fars | AI_FA0002 | HU_2306.14048 | Adaptive SRE-Mass Cache Sizing for Hybrid Linear Attention | H$_2$O: Heavy-Hitter Oracle for Efficient Generative Inference of Large Language Models | 2306.14048 | NeurIPS 2023 | 2,023 | null | method | method | true | cited | 3 | parent | Dynamic KV-cache token retention policy; same which-tokens-to-keep problem, same contribution kind. | null | year<=2024 | null | pass | null | null | null | null | null | 2,294 | 4,530 | 1.97 | Efficient inference, architecture, and training | null | null | null | null | null | null | shipped_tex |
FA0004 | fars | AI_FA0004 | HU_2401.10480 | Anytime-CBU: Adaptive Rollout Allocation for Consequence-Based Utility Scoring | Escape Sky-high Cost: Early-stopping Self-Consistency for Multi-step Reasoning | 2401.10480 | ICLR 2024 | 2,024 | null | method | method | true | cited | 3 | cited | Early-stopping scheme to cut multi-sample reasoning cost; same adaptive-sampling-budget problem, same kind. | null | year<=2024 | null | pass | null | null | null | null | null | 1,867 | 3,064 | 1.64 | Post-training and reasoning | null | null | null | null | null | null | shipped_tex |
FA0005 | fars | AI_FA0005 | HU_2312.02119 | Selective Delexicalization to Defend Structured-Output LLM APIs from Control-Plane Jailbreaks | Tree of Attacks: Jailbreaking Black-Box LLMs Automatically | 2312.02119 | NeurIPS 2023 | 2,023 | null | method | method | true | cited | 1 | cited | Automated black-box jailbreak attack; same jailbreak area, attack generation rather than structured-output defense. | null | year<=2024 | null | pass | null | null | null | null | null | 2,379 | 4,645 | 1.95 | Safety, security, and privacy | null | null | null | null | null | null | shipped_tex |
FA0006 | fars | AI_FA0006 | HU_2407.18370 | View-Disagreement Escalation for Robust Web-Agent Trajectory Judges | Trust or Escalate: LLM Judges with Provable Guarantees for Human Agreement | 2407.18370 | ICLR 2025 oral | 2,024 | 8 | method | method | true | cited | 3 | cited | null | pass | year<=2024 | pass | pass | pass | 0.9473 | null | 1 | null | 1,642 | 3,535 | 2.15 | Evaluation, judging, and verification | null | null | null | null | null | null | shipped_tex |
FA0008 | fars | AI_FA0008 | HU_2207.01780 | Confidence-Bounded Unit-Test Rewards for Reinforcement Learning from Verifiable Rewards | CodeRL: Mastering Code Generation through Pretrained Models and Deep Reinforcement Learning | 2207.01780 | NeurIPS 2022 | 2,022 | null | method | method | true | cited | 3 | cited | RL for code generation with unit-test-derived reward signals; same problem, same kind. | null | year<=2024 | null | pass | null | null | null | null | null | 1,836 | 8,295 | 4.52 | Post-training and reasoning | null | null | null | null | null | null | shipped_tex |
FA0012 | fars | AI_FA0012 | HU_2401.12168 | Delta-Map Belief Updates for Stable Spatial Revision in Vision-Language Models | SpatialVLM: Endowing Vision-Language Models with Spatial Reasoning Capabilities | 2401.12168 | CVPR 2024 | 2,024 | null | method | method | true | cited | 1 | cited | Trains VLMs on large spatial VQA data; same area, different problem. | null | year<=2024 | null | pass | null | null | null | null | null | 1,754 | 4,010 | 2.29 | Vision, multimodal, and applied domains | null | null | null | null | null | null | shipped_tex |
FA0013 | fars | AI_FA0013 | HU_2601.03309 | Contractive Recurrent Cores for Depth-Extrapolatable Vision-Language-Action Policies: An Empirical Investigation on LIBERO | VLM4VLA: Revisiting Vision-Language-Models in Vision-Language-Action Models | 2601.03309 | ICLR 2026 poster | null | 7 | analysis | analysis | true | widened_iclr26 | 2 | widened (not cited; topic neighbour) | systematic empirical study of VLM backbone choices in VLA; same area and same study kind | pass (pangram) | null | pass | pass | pass | null | 0.101 | null | 0 | 2,000 | 5,470 | null | Vision, multimodal, and applied domains | null | null | null | null | null | null | shipped_tex |
FA0015 | fars | AI_FA0015 | HU_2402.04333 | Orthogonal Junk: Gradient-Orthogonality Data Selection for Continual Pre-Training on Low-Quality Data | LESS: Selecting Influential Data for Targeted Instruction Tuning | 2402.04333 | ICML 2024 | 2,024 | null | method | method | true | cited | 3 | cited | Gradient-based data selection for targeted LLM training; same problem, same kind. | null | year<=2024 | null | pass | null | null | null | null | null | 2,164 | 5,054 | 2.34 | Post-training and reasoning | null | null | null | null | null | null | shipped_tex |
FA0016 | fars | AI_FA0016 | HU_2504.07109 | Query-Conditioned Marginals for OT-Based Context Compression: An Empirical Investigation | OSCAR: Online Soft Compression And Reranking | 2504.07109 | ICLR 2026 poster | 2,025 | 6.5 | method | method | true | cited | 3 | cited | Query-dependent online soft context compression method, same problem and same kind | null | pangram | null | pass | null | null | null | null | 0 | 1,848 | 3,451 | 1.87 | Retrieval, grounding, and embeddings | null | null | null | null | null | null | shipped_tex |
FA0017 | fars | AI_FA0017 | HU_2504.13936 | Copy-Then-Inpaint: Improving Temporal Consistency in Multi-Step GUI Generation via Selective Region Editing | ViMo: A Generative Visual GUI World Model for App Agents | 2504.13936 | ICLR 2026 poster | 2,025 | 6 | method | method | true | cited | 3 | cited | Visual GUI world model generating future app screens as images, same problem and kind | null | pangram | null | pass | null | null | null | null | 0 | 2,028 | 4,251 | 2.1 | Vision, multimodal, and applied domains | null | null | null | null | null | null | shipped_tex |
FA0018 | fars | AI_FA0018 | HU_2505.22257 | Compute-Matched Evaluation of Transform-Augmented GRPO for Mathematical Reasoning | Revisiting Group Relative Policy Optimization: Insights into On-Policy and Off-Policy Training | 2505.22257 | ICLR 2026 poster | null | 5 | analysis | analysis | true | widened_iclr26 | 2 | widened (not cited; topic neighbour) | controlled revisit of GRPO on/off-policy regimes; same problem family and same study kind | pass (pangram) | null | pass | pass | pass | null | 0.088 | null | 0.038009 | 1,377 | 3,328 | null | Post-training and reasoning | null | null | null | null | null | null | shipped_tex |
FA0019 | fars | AI_FA0019 | HU_2509.16131 | Step-Down Bridge Guidance Scheduling for Dual-CFG in Video-Audio Diffusion | Dynamic Classifier-Free Diffusion Guidance via Online Feedback | 2509.16131 | ICLR 2026 poster | 2,025 | 5.5 | method | method | true | cited | 3 | cited | Dynamic per-timestep CFG scale scheduling, same guidance-scheduling problem and kind | null | pangram | null | pass | null | null | null | null | 0 | 2,110 | 3,839 | 1.82 | Vision, multimodal, and applied domains | null | null | null | null | null | null | shipped_tex |
FA0020 | fars | AI_FA0020 | HU_2310.01691 | AlignDefTok: Training-Free Transfer of DefensiveTokens via Embedding-Space Alignment | Zero-Shot Continuous Prompt Transfer: Generalizing Task Semantics Across Language Models | 2310.01691 | ICLR 2023 | 2,023 | null | method | method | true | cited | 3 | cited | Zero-shot cross-model continuous prompt transfer, same problem and same kind | null | year<=2024 | null | pass | null | null | null | null | null | 1,787 | 4,087 | 2.29 | Safety, security, and privacy | null | null | null | null | null | null | shipped_tex |
FA0021 | fars | AI_FA0021 | HU_2305.20050 | Does iGRPO Need a Good Draft? Best-vs-Worst Self-Conditioning Ablation for RLVR Math | Let's Verify Step by Step | 2305.20050 | ICLR 2023 | 2,023 | null | analysis | analysis | true | cited | 1 | cited | Process vs outcome supervision study for math, same math-training area, different question | null | year<=2024 | null | pass | null | null | null | null | null | 1,656 | 4,135 | 2.5 | Post-training and reasoning | null | null | null | null | null | null | shipped_tex |
FA0022 | fars | AI_FA0022 | HU_2305.11206 | The Repetition Advantage in Long-CoT SFT is a Termination Effect | LIMA: Less Is More for Alignment | 2305.11206 | NeurIPS 2023 | 2,023 | null | analysis | analysis | true | cited | 2 | cited | SFT data-quantity/quality study (less is more), same SFT data-scaling problem, alignment not long-CoT termination | null | year<=2024 | null | pass | null | null | null | null | null | 2,323 | 3,516 | 1.51 | Post-training and reasoning | null | null | null | null | null | null | shipped_tex |
FA0023 | fars | AI_FA0023 | HU_2410.02712 | Answer-Free Self-Referential Critics: Training Solve-Then-Judge VLM Judges with Preference Labels but Without Ground-Truth Answers | LLaVA-Critic: Learning to Evaluate Multimodal Models | 2410.02712 | CVPR 2024 | 2,024 | null | method | method | true | cited | 3 | cited | Trains VLM critics/evaluators for multimodal judging, same problem and same kind | null | year<=2024 | null | pass | null | null | null | null | null | 2,067 | 3,970 | 1.92 | Evaluation, judging, and verification | null | null | null | null | null | null | shipped_tex |
FA0025 | fars | AI_FA0025 | HU_2405.20915 | Risk-Controlled Early Exit for Diffusion Language Models | Fast yet Safe: Early-Exiting with Risk Control | 2405.20915 | NeurIPS 2024 | 2,024 | null | method | method | true | cited | 3 | parent | Risk control for early-exit networks, same risk-controlled early-exit problem and kind | null | year<=2024 | null | pass | null | null | null | null | null | 1,953 | 4,979 | 2.55 | Efficient inference, architecture, and training | null | null | null | null | null | null | shipped_tex |
FA0027 | fars | AI_FA0027 | HU_2402.14016 | RefSwap: Counterfactual Reference-Swap Verification for Robust LLM Verifiers | Is LLM-as-a-Judge Robust? Investigating Universal Adversarial Attacks on Zero-shot LLM Assessment | 2402.14016 | EMNLP 2024 | 2,024 | null | method | method | true | cited | 3 | cited | Universal adversarial phrases fooling LLM judges, same judge-robustness problem, method (attack side) | null | year<=2024 | null | pass | null | null | null | null | null | 2,049 | 3,683 | 1.8 | Evaluation, judging, and verification | null | null | null | null | null | null | shipped_tex |
FA0028 | fars | AI_FA0028 | HU_2510.01161 | Acceptance-Controlled MIS-PO: Adaptive Trajectory Filtering for Stable Off-Policy RLVR Training | Prosperity before Collapse: How Far Can Off-Policy RL Reach with Stale Data on LLMs? | 2510.01161 | ICLR 2026 poster | 2,025 | 5 | method | method | true | cited | 3 | parent | Stabilizing off-policy LLM RL under stale data via importance-weight control, same problem and kind | null | pangram | null | pass | null | null | null | null | 0 | 2,080 | 3,643 | 1.75 | Post-training and reasoning | null | null | null | null | null | null | shipped_tex |
FA0029 | fars | AI_FA0029 | HU_2504.02010 | Output-Space Allocation Costs for Calibration-Guided LLM Compression: An Empirical Study | When Reasoning Meets Compression: Understanding the Effects of LLMs Compression on Large Reasoning Models | 2504.02010 | ICLR 2026 poster | null | 4.4 | analysis | analysis | true | widened_iclr26 | 2 | widened (not cited; topic neighbour) | in-depth study of compression effects on reasoning LLMs; same problem area and kind | pass (pangram) | null | pass | pass | pass | null | 0.094 | null | 0 | 1,478 | 3,673 | null | Efficient inference, architecture, and training | null | null | null | null | null | null | shipped_tex |
FA0031 | fars | AI_FA0031 | HU_2310.01798 | Evidence-Grounded Constraint Schemas Do Not Improve Medical LLM Guardrails on LiveMedBench | Large Language Models Cannot Self-Correct Reasoning Yet | 2310.01798 | ICLR 2023 | 2,023 | null | analysis | analysis | true | cited | 1 | cited | Negative analysis of self-correction; same kind but different problem (reasoning self-correction). | null | year<=2024 | null | pass | null | null | null | null | null | 1,564 | 3,556 | 2.27 | Safety, security, and privacy | null | null | null | null | null | null | shipped_tex |
FA0032 | fars | AI_FA0032 | HU_2507.02259 | RC-MemStop: Risk-Controlled Early Stopping for Long-Context Memory Agents | MemAgent: Reshaping Long-Context LLM with Multi-Conv RL-based Memory Agent | 2507.02259 | ICLR 2026 oral | 2,025 | 6.5 | method | method | true | cited | 2 | parent | MemAgent is the base memory agent; same problem area, different contribution. | null | pangram | null | pass | null | null | null | null | 0 | 1,851 | 4,170 | 2.25 | Agents, memory, and code | null | null | null | null | null | null | shipped_tex |
FA0033 | fars | AI_FA0033 | HU_2405.15793 | Interface-Aware Smoke Tests and Deterministic Import Autofix for Feature-Level Coding Agents: A Negative Result | SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering | 2405.15793 | NeurIPS 2024 | 2,024 | null | method | method | true | cited | 3 | cited | SWE-agent designs agent-computer interface tooling for coding agents; same problem and kind. | null | year<=2024 | null | pass | null | null | null | null | null | 2,142 | 4,237 | 1.98 | Agents, memory, and code | null | null | null | null | null | null | shipped_tex |
FA0034 | fars | AI_FA0034 | HU_2402.13217 | Adaptive Rerank Budgeting for Video-Text Retrieval via Layer-Disagreement Routing | VideoPrism: A Foundational Visual Encoder for Video Understanding | 2402.13217 | ICML 2024 | 2,024 | null | method | method | true | cited | 1 | cited | VideoPrism is the video encoder substrate. | null | year<=2024 | null | pass | null | null | null | null | null | 1,546 | 3,621 | 2.34 | Retrieval, grounding, and embeddings | null | null | null | null | null | null | shipped_tex |
FA0035 | fars | AI_FA0035 | HU_2510.00615 | LASCon: Loop-Aware Scratchpad Condensation for Terminal Agents | ACON: Optimizing Context Compression for Long-horizon LLM Agents | 2510.00615 | ICML 2026 (ICLR 2026 reject) | 2,025 | 4 | method | method | true | cited | 3 | cited | null | pass | pangram | pass | pass | pass | 0.9587 | null | 1 | 0.044512 | 2,183 | 3,435 | 1.57 | Agents, memory, and code | null | null | null | null | null | null | shipped_tex |
FA0036 | fars | AI_FA0036 | HU_2507.20673 | EMA-KPO: Simplifying Kalman Policy Optimization with Fixed-Gain Exponential Smoothing | Geometric-Mean Policy Optimization | 2507.20673 | ICLR 2026 poster | 2,025 | 5 | method | method | true | cited | 3 | cited | GMPO stabilizes GRPO against extreme importance ratios; same problem and same kind. | null | pangram | null | pass | null | null | null | null | 0.045098 | 1,726 | 2,987 | 1.73 | Post-training and reasoning | null | null | null | null | null | null | shipped_tex |
FA0038 | fars | AI_FA0038 | HU_2310.07712 | Citation-Consistent Voting for Permutation-Robust Retrieval-Augmented Generation | Found in the Middle: Permutation Self-Consistency Improves Listwise Ranking in Large Language Models | 2310.07712 | NAACL 2023 | 2,023 | null | method | method | true | cited | 3 | cited | Permutation self-consistency marginalizes list orders against position bias; same problem and kind. | null | year<=2024 | null | pass | null | null | null | null | null | 1,758 | 4,140 | 2.35 | Retrieval, grounding, and embeddings | null | null | null | null | null | null | shipped_tex |
FA0039 | fars | AI_FA0039 | HU_2410.01679 | Prefix-Ratio GRPO: Improving Gradient Quality for Reinforcement Learning with Verifiable Rewards | VinePPO: Refining Credit Assignment in RL Training of LLMs | 2410.01679 | ICML 2024 | 2,024 | null | method | method | true | cited | 2 | cited | VinePPO refines credit assignment in LLM RL; same area, different problem (credit vs staleness). | null | year<=2024 | null | pass | null | null | null | null | null | 2,198 | 4,338 | 1.97 | Post-training and reasoning | null | null | null | null | null | null | shipped_tex |
FA0040 | fars | AI_FA0040 | HU_2306.10763 | Typed-DSL Constrained Data Recipes for Higher Executability in DataChef | Guiding Language Models of Code with Global Context using Monitors | 2306.10763 | NeurIPS 2023 | 2,023 | null | method | method | true | cited | 2 | cited | Monitor-guided decoding constrains code generation for validity; same problem family, different mechanism. | null | year<=2024 | null | pass | null | null | null | null | null | 1,812 | 5,280 | 2.91 | Agents, memory, and code | null | null | null | null | null | null | shipped_tex |
FA0041 | fars | AI_FA0041 | HU_2303.11366 | MEL-Code: Transferring Meta-Experience Learning to Code RLVR with Unit-Test Rewards | Reflexion: Language Agents with Verbal Reinforcement Learning | 2303.11366 | NeurIPS 2023 | 2,023 | null | method | method | true | cited | 2 | cited | Reflexion internalizes experience via verbal feedback; same idea family, different setting. | null | year<=2024 | null | pass | null | null | null | null | null | 1,700 | 4,026 | 2.37 | Post-training and reasoning | null | null | null | null | null | null | shipped_tex |
FA0042 | fars | AI_FA0042 | HU_2402.15449 | Distilling Bidirectional Embedding Teachers into Streaming-Compatible Causal Students | Repetition Improves Language Model Embeddings | 2402.15449 | ICLR 2024 | 2,024 | null | method | method | true | cited | 3 | parent | Echo embeddings get bidirectional-quality embeddings from causal LMs; same problem and kind. | null | year<=2024 | null | pass | null | null | null | null | null | 1,898 | 4,714 | 2.48 | Retrieval, grounding, and embeddings | null | null | null | null | null | null | shipped_tex |
FA0043 | fars | AI_FA0043 | HU_2403.17710 | Isolated Solve-Then-Judge: A Simple Defense Against Candidate-Response Prompt Injection for Multimodal LLM Judges | Optimization-based Prompt Injection Attack to LLM-as-a-Judge | 2403.17710 | CCS 2024 | 2,024 | null | method | method | true | cited | 3 | cited | JudgeDeceiver is an optimization-based prompt-injection attack on LLM-as-a-judge; same problem, attack side of the same method space. | null | year<=2024 | null | pass | null | null | null | null | null | 1,930 | 8,383 | 4.34 | Safety, security, and privacy | null | null | null | null | null | null | shipped_tex |
FA0044 | fars | AI_FA0044 | HU_2404.04475 | Selective Self-Reference for LLM-as-a-Judge: Using Self-Consistency to Reduce Error Propagation | Length-Controlled AlpacaEval: A Simple Way to Debias Automatic Evaluators | 2404.04475 | COLM 2024 | 2,024 | null | method | method | true | cited | 3 | cited | Also proposes a technique to fix a systematic error of LLM auto-evaluators (length bias), same judge-reliability problem. | null | year<=2024 | null | pass | null | null | null | null | null | 1,929 | 3,868 | 2.01 | Evaluation, judging, and verification | null | null | null | null | null | null | shipped_tex |
FA0046 | fars | AI_FA0046 | HU_2210.08726 | QuoteVerify: Inference-Time Quote-Backed Citation Verification for Deep Research Reports | RARR: Researching and Revising What Language Models Say, Using Language Models | 2210.08726 | ACL 2023 | 2,022 | null | method | method | true | cited | 3 | cited | null | pass | year<=2024 | pass | pass | pass | 0.9055 | null | 1 | null | 2,032 | 4,781 | 2.35 | Retrieval, grounding, and embeddings | null | null | null | null | null | null | shipped_tex |
FA0047 | fars | AI_FA0047 | HU_2302.04761 | Canonical Schema Views for Activation Steering Under Tool-Schema Churn: A Negative Result | Toolformer: Language Models Can Teach Themselves to Use Tools | 2302.04761 | NeurIPS 2023 | 2,023 | null | method | method | true | cited | 1 | cited | Same tool-use area; self-supervised training to call APIs, different problem from churn-invariant steering. | null | year<=2024 | null | pass | null | null | null | null | null | 1,996 | 4,889 | 2.45 | Agents, memory, and code | null | null | null | null | null | null | shipped_tex |
FA0050 | fars | AI_FA0050 | HU_2401.06081 | R-MEL: Recovering Contrastive Signal from All-Negative Groups via Prefix-Primed Revision | Improving Large Language Models via Fine-grained Reinforcement Learning with Minimum Editing Constraint | 2401.06081 | ACL 2024 | 2,024 | null | method | method | true | cited | 3 | cited | Same problem of extracting learning signal from erroneous solutions in RL for reasoning, via rewriting/revision; same kind. | null | year<=2024 | null | pass | null | null | null | null | null | 2,059 | 4,103 | 1.99 | Post-training and reasoning | null | null | null | null | null | null | shipped_tex |
FA0051 | fars | AI_FA0051 | HU_2312.06635 | Toeplitz Block Mixing for Scalable Multi-Head Linear Attention | Gated Linear Attention Transformers with Hardware-Efficient Training | 2312.06635 | ICML 2023 | 2,023 | null | method | method | true | cited | 3 | cited | Same linear-attention performance/efficiency problem; proposes gated linear attention with efficient training. | null | year<=2024 | null | pass | null | null | null | null | null | 2,076 | 4,839 | 2.33 | Efficient inference, architecture, and training | null | null | null | null | null | null | shipped_tex |
FA0052 | fars | AI_FA0052 | HU_2310.03716 | Does MIS-PO Need Ratio-Based Trajectory Selection? A Random-Rejection Mechanism Test | A Long Way to Go: Investigating Length Correlations in RLHF | 2310.03716 | COLM 2024 | 2,023 | null | analysis | analysis | true | cited | 1 | cited | Same kind of mechanism dissection of RL for LLMs, but targets length bias in RLHF, a different problem. | null | year<=2024 | null | pass | null | null | null | null | null | 1,476 | 3,687 | 2.5 | Post-training and reasoning | null | null | null | null | null | null | shipped_tex |
FA0053 | fars | AI_FA0053 | HU_2309.03883 | Draft De-anchoring Decoding Does Not Mitigate Contextual Drag in LLM Reasoning | DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models | 2309.03883 | ICLR 2024 | 2,023 | null | method | method | true | cited | 2 | cited | Closest mechanistic prior: training-free contrastive-logit decoding, but targets factuality/hallucination, not contextual drag. | null | year<=2024 | null | pass | null | null | null | null | null | 2,098 | 3,757 | 1.79 | Efficient inference, architecture, and training | null | null | null | null | null | null | shipped_tex |
FA0055 | fars | AI_FA0055 | HU_2410.18252 | Decoupling Snapshot Publication from Staleness Tolerance in Distributed GRPO via Lossless Sparse Patches | Asynchronous RLHF: Faster and More Efficient Off-Policy RL for Language Models | 2410.18252 | ICLR 2024 | 2,024 | null | method | method | true | cited | 3 | cited | Same problem of asynchronous off-policy LLM RL and how much staleness training tolerates; same systems/method kind. | null | year<=2024 | null | pass | null | null | null | null | null | 2,053 | 4,217 | 2.05 | Post-training and reasoning | null | null | null | null | null | null | shipped_tex |
FA0056 | fars | AI_FA0056 | HU_2510.03817 | Innovation Saturation Does Not Robustify Kalman-Filtered Importance Ratios in LLM Reinforcement Learning | TROLL: Trust Regions Improve Reinforcement Learning for Large Language Models | 2510.03817 | ICLR 2026 oral | null | 6 | method | method | true | widened_iclr26 | 3 | widened (not cited; topic neighbour) | TROLL replaces PPO clipping of importance ratios with trust regions; same extreme-ratio problem and same kind | pass (pangram) | null | pass | pass | pass | null | 0.075 | null | 0 | 1,739 | 4,211 | null | Post-training and reasoning | null | null | null | null | null | null | shipped_tex |
FA0057 | fars | AI_FA0057 | HU_2504.11373 | LiveMedBench-Ask1: Evaluating Ask-Before-Answer Behavior in Medical LLMs | Cancer-Myth: Evaluating Large Language Models on Patient Questions with False Presuppositions | 2504.11373 | ICLR 2026 poster | null | 5.5 | benchmark | benchmark | true | widened_iclr26 | 3 | widened (not cited; topic neighbour) | benchmark of medical LLM behavior on questions needing more than direct answering; same problem and same kind | pass (pangram) | null | pass | pass | pass | null | 0.094 | null | 0 | 2,298 | 3,458 | null | Evaluation, judging, and verification | null | null | null | null | null | null | shipped_tex |
FA0058 | fars | AI_FA0058 | HU_2410.13232 | Chunked Budget Allocation Prevents Non-Monotonic Regressions in World-Model Verification | Web Agents with World Models: Learning and Leveraging Environment Dynamics in Web Navigation | 2410.13232 | ICLR 2024 | 2,024 | null | method | method | true | cited | 3 | cited | World-model-augmented web agent simulating action outcomes to avoid irreversible mistakes; same problem, same method-type contribution. | null | year<=2024 | null | pass | null | null | null | null | null | 1,889 | 4,638 | 2.46 | Agents, memory, and code | null | null | null | null | null | null | shipped_tex |
FA0059 | fars | AI_FA0059 | HU_2405.14831 | Last-Write-Wins Memory: Isolating Deterministic Overwrite Semantics for Long-Context Conflict Resolution | HippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models | 2405.14831 | NeurIPS 2024 | 2,024 | null | method | method | true | cited | 3 | parent | Long-term memory system for LLMs integrating updated knowledge for multi-hop QA; same problem, same contribution kind. | null | year<=2024 | null | pass | null | null | null | null | null | 2,384 | 4,270 | 1.79 | Agents, memory, and code | null | null | null | null | null | null | shipped_tex |
FA0064 | fars | AI_FA0064 | HU_2512.20569 | NLL-Guided Full-Attention Layer Selection for Training-Free Sliding-Window Adaptation | Distilling to Hybrid Attention Models via KL-Guided Layer Selection | 2512.20569 | ICLR 2026 poster | 2,025 | 6 | method | method | true | cited | 3 | cited | KL-guided layer selection for hybrid attention conversion; same layer-selection problem, same contribution kind. | null | pangram | null | pass | null | null | null | null | 0 | 1,877 | 3,604 | 1.92 | Efficient inference, architecture, and training | null | null | null | null | null | null | shipped_tex |
FA0065 | fars | AI_FA0065 | HU_2112.12777 | Mean-Direction Deflation Reranking for Metric Misuse Repair in Frozen Vector Search | Cross Modal Retrieval with Querybank Normalisation | 2112.12777 | CVPR 2021 | 2,021 | null | method | method | true | cited | 3 | parent | Query-bank normalization fixing hubness in frozen retrieval embeddings without retraining; same problem, same kind. | null | year<=2024 | null | pass | null | null | null | null | null | 1,872 | 4,155 | 2.22 | Retrieval, grounding, and embeddings | null | null | null | null | null | null | shipped_tex |
FA0067 | fars | AI_FA0067 | HU_2405.19715 | Delta-Prefill Switching: Adaptive Routing for Speculative Decoding in Multi-Turn LLM Serving | SpecDec++: Boosting Speculative Decoding via Adaptive Candidate Lengths | 2405.19715 | COLM 2025 | 2,024 | null | method | method | true | cited | 3 | cited | Adaptive control of speculative decoding (candidate length policy); same adaptive-speculation problem, same kind. | null | year<=2024 | null | pass | null | null | null | null | null | 2,220 | 3,968 | 1.79 | Efficient inference, architecture, and training | null | null | null | null | null | null | shipped_tex |
FA0069 | fars | AI_FA0069 | HU_1712.00378 | Timeout Bootstrapping for Long-CoT RLVR: Promise and Pitfalls | Time Limits in Reinforcement Learning | 1712.00378 | ICML 2017 | 2,017 | null | method | method | true | cited | 3 | cited | Formalizes time limits in RL and proposes bootstrapping at timeouts; same truncation-handling problem, same kind. | null | year<=2024 | null | pass | null | null | null | null | null | 2,463 | 4,710 | 1.91 | Post-training and reasoning | null | null | null | null | null | null | shipped_tex |
FA0072 | fars | AI_FA0072 | HU_2406.07524 | Execution-Trace Guided Remasking for Diffusion Code Generation | Simple and Effective Masked Diffusion Language Models | 2406.07524 | NeurIPS 2024 | 2,024 | null | method | method | true | cited | 1 | cited | Masked diffusion LM training recipe; supplies the base model class, different problem from repair localization. | null | year<=2024 | null | pass | null | null | null | null | null | 2,098 | 4,236 | 2.02 | Agents, memory, and code | null | null | null | null | null | null | shipped_tex |
FA0073 | fars | AI_FA0073 | HU_2410.10781 | Sink-Free Attention Enables Prefix-Free Streaming KV Caches | When Attention Sink Emerges in Language Models: An Empirical View | 2410.10781 | ICLR 2024 | 2,024 | null | analysis | analysis | true | cited | 3 | cited | Empirical study of when and why attention sinks emerge; same phenomenon studied, same analysis-type contribution. | null | year<=2024 | null | pass | null | null | null | null | null | 1,582 | 3,979 | 2.52 | Efficient inference, architecture, and training | null | null | null | null | null | null | shipped_tex |
FA0074 | fars | AI_FA0074 | HU_2510.09462 | Auditing and Hardening LiveMedBench's Rubric Grader Against Prompt Injection: A Negative Result | Adaptive Attacks on Trusted Monitors Subvert AI Control Protocols | 2510.09462 | ICLR 2026 poster | null | 6.5 | analysis | analysis | true | widened_iclr26 | 3 | widened (not cited; topic neighbour) | adaptive-attack stress test subverting trusted LLM monitors; same judge-integrity problem and same adversarial-audit kind | pass (pangram) | null | pass | pass | pass | null | 0.122 | null | 0.033063 | 2,003 | 5,326 | null | Safety, security, and privacy | null | null | null | null | null | null | shipped_tex |
FA0075 | fars | AI_FA0075 | HU_2506.15076 | Syntax-Diversified Unlearning: Evaluating Data-Side Interventions for Reducing Worst-Case Leakage | Learning-Time Encoding Shapes Unlearning in LLMs | 2506.15076 | ICLR 2026 poster | 2,025 | 4.667 | analysis | analysis | true | cited | 3 | cited | Empirical study of how paraphrase-style encoding choices affect unlearning effectiveness, same problem and same analysis-style contribution | null | pangram | null | pass | null | null | null | null | 0 | 1,747 | 4,419 | 2.53 | Safety, security, and privacy | null | null | null | null | null | null | shipped_tex |
FA0076 | fars | AI_FA0076 | HU_1904.09675 | Entailment-Checklist Scoring: An API-Free Alternative to LLM-Based Dense Video Caption Evaluation | BERTScore: Evaluating Text Generation with BERT | 1904.09675 | ICLR 2019 | 2,019 | null | framework | framework | true | cited | 3 | parent | API-free automatic text-generation evaluation metric, same problem and same kind of contribution | null | year<=2024 | null | pass | null | null | null | null | null | 1,683 | 3,674 | 2.18 | Evaluation, judging, and verification | null | null | null | null | null | null | shipped_tex |
FA0077 | fars | AI_FA0077 | HU_2409.05907 | LogitGate: Probe-Gated Output Logit Bias as a Simplification of Activation Steering for Tool Calling | Programming Refusal with Conditional Activation Steering | 2409.05907 | ICLR 2024 | 2,024 | null | method | method | true | cited | 3 | cited | Probe/condition-gated selective activation steering of LLM behavior, same conditional-steering problem and same method contribution | null | year<=2024 | null | pass | null | null | null | null | null | 2,309 | 4,066 | 1.76 | Agents, memory, and code | null | null | null | null | null | null | shipped_tex |
FA0082 | fars | AI_FA0082 | HU_2005.11401 | Context Bagging: Inference-Time Ensembling for Robust Long-Context QA Under Hard Distractors | Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks | 2005.11401 | NeurIPS 2020 | 2,020 | null | method | method | true | cited | 1 | cited | Foundational retrieval-augmented generation, same knowledge-intensive QA area but different problem from distractor-robust ensembling | null | year<=2024 | null | pass | null | null | null | null | null | 1,868 | 4,303 | 2.3 | Retrieval, grounding, and embeddings | null | null | null | null | null | null | shipped_tex |
FA0083 | fars | AI_FA0083 | HU_2407.12784 | Query-OOD Escalation: Detecting Memory Poisoning Attacks via Embedding-Space Anomaly Detection | AgentPoison: Red-teaming LLM Agents via Poisoning Memory or Knowledge Bases | 2407.12784 | NeurIPS 2024 | 2,024 | null | method | method | true | cited | 3 | parent | The memory-poisoning attack QOE defends against, same agent memory-poisoning security problem and same method-style contribution | null | year<=2024 | null | pass | null | null | null | null | null | 1,980 | 3,898 | 1.97 | Safety, security, and privacy | null | null | null | null | null | null | shipped_tex |
FA0085 | fars | AI_FA0085 | HU_2310.17631 | Tool-Gated Residual Distillation for DataChef Verifier Scoring | JudgeLM: Fine-tuned Large Language Models are Scalable Judges | 2310.17631 | ICLR 2023 | 2,023 | null | method | method | true | cited | 3 | cited | Fine-tunes smaller LLMs into scalable judges, same judge-distillation problem and same method contribution | null | year<=2024 | null | pass | null | null | null | null | null | 2,062 | 3,858 | 1.87 | Evaluation, judging, and verification | null | null | null | null | null | null | shipped_tex |
FA0087 | fars | AI_FA0087 | HU_1811.11682 | RazorSFT: On-Policy Supervised Fine-Tuning with KL-Minimal Target Selection for Continual Learning | Experience Replay for Continual Learning | 1811.11682 | NeurIPS 2018 | 2,018 | null | method | method | true | cited | 3 | cited | Replay method mitigating catastrophic forgetting in sequential training, same problem and same method contribution | null | year<=2024 | null | pass | null | null | null | null | null | 1,974 | 3,505 | 1.78 | Continual learning and model editing | null | null | null | null | null | null | shipped_tex |
FA0100 | fars | AI_FA0100 | HU_2004.04906 | Self-Anchored Temporal Filtering for LLM-Free Temporal-Aware Memory Retrieval | Dense Passage Retrieval for Open-Domain Question Answering | 2004.04906 | EMNLP 2020 | 2,020 | null | method | method | true | cited | 1 | cited | Dense passage retrieval, same retrieval area (the pure-dense baseline) but different problem from temporal-aware memory ranking | null | year<=2024 | null | pass | null | null | null | null | null | 1,822 | 4,868 | 2.67 | Agents, memory, and code | null | null | null | null | null | null | shipped_tex |
FA0101 | fars | AI_FA0101 | HU_2505.20274 | Task-Aware Early Termination for HNSW via Label-Histogram Stabilization | Probabilistic Kernel Function for Fast Angle Testing | 2505.20274 | ICLR 2026 oral | null | 8 | method | method | true | widened_iclr26 | 2 | widened (not cited; topic neighbour) | ANN efficiency primitives; same area | pass (pangram) | null | pass | pass | pass | null | 0.127 | null | 0 | 1,716 | 4,421 | null | Retrieval, grounding, and embeddings | null | null | null | null | null | null | shipped_tex |
FA0102 | fars | AI_FA0102 | HU_2301.12314 | KL-Time Replay: Function-Space Drift Monitoring for Continual Learning in LLMs | Progressive Prompts: Continual Learning for Language Models | 2301.12314 | ICLR 2023 | 2,023 | null | method | method | true | cited | 3 | cited | Continual learning method for language models against catastrophic forgetting, same problem and same kind | null | year<=2024 | null | pass | null | null | null | null | null | 1,829 | 3,734 | 2.04 | Continual learning and model editing | null | null | null | null | null | null | shipped_tex |
FA0104 | fars | AI_FA0104 | HU_2510.10125 | Search-Anchored Hybrid Rollouts for Text-Based World Models | Ctrl-World: A Controllable Generative World Model for Robot Manipulation | 2510.10125 | ICLR 2026 poster | null | 6 | method | method | true | widened_iclr26 | 2 | widened (not cited; topic neighbour) | controllable world model for policy rollout evaluation; same problem area | pass (pangram) | null | pass | pass | pass | null | 0.083 | null | 0 | 2,153 | 3,161 | null | Agents, memory, and code | null | null | null | null | null | null | shipped_tex |
FA0105 | fars | AI_FA0105 | HU_2208.04378 | Cross-View PSD Distillation for Viewpoint-Robust Remote Photoplethysmography | Contrast-Phys: Unsupervised Video-based Remote Physiological Measurement via Spatiotemporal Contrast | 2208.04378 | ECCV 2022 | 2,022 | null | method | method | true | cited | 3 | cited | Contrastive training method for the same video rPPG measurement task. | null | year<=2024 | null | pass | null | null | null | null | null | 1,857 | 4,131 | 2.22 | Vision, multimodal, and applied domains | null | null | null | null | null | null | shipped_tex |
FA0106 | fars | AI_FA0106 | HU_2408.03910 | TraceBound: Evaluating Trace-Bounded Context for Token-Efficient Coding Agents | CodexGraph: Bridging Large Language Models and Code Repositories via Code Graph Databases | 2408.03910 | NAACL 2024 | 2,024 | null | method | method | true | cited | 3 | cited | Graph-database method for structure-aware context retrieval for repo-scale coding agents, same problem. | null | year<=2024 | null | pass | null | null | null | null | null | 1,520 | 3,839 | 2.53 | Agents, memory, and code | null | null | null | null | null | null | shipped_tex |
FA0107 | fars | AI_FA0107 | HU_2006.04152 | ConvergeStop: Inference-Time Convergence-Based Halting for Generative Text Embeddings | BERT Loses Patience: Fast and Robust Inference with Early Exit | 2006.04152 | NeurIPS 2020 | 2,020 | null | method | method | true | cited | 3 | cited | Patience-based early exit that halts when intermediate predictions stabilize, same convergence-halting problem and kind. | null | year<=2024 | null | pass | null | null | null | null | null | 1,886 | 3,277 | 1.74 | Efficient inference, architecture, and training | null | null | null | null | null | null | shipped_tex |
FA0110 | fars | AI_FA0110 | HU_2210.11287 | Targeted Counterfactual Branch Augmentation for Robust Text-Based World Models under Agent Policy Shift | MoCoDA: Model-based Counterfactual Data Augmentation | 2210.11287 | NeurIPS 2022 | 2,022 | null | method | method | true | cited | 3 | cited | Model-based counterfactual data augmentation for dynamics-model generalization, same problem and kind. | null | year<=2024 | null | pass | null | null | null | null | null | 2,060 | 4,812 | 2.34 | Agents, memory, and code | null | null | null | null | null | null | shipped_tex |
FA0111 | fars | AI_FA0111 | HU_2402.16617 | Label-Free Hyperparameter Calibration for Parallel Context Encoding via KL Divergence Matching | Long-Context Language Modeling with Parallel Context Encoding | 2402.16617 | ACL 2024 | 2,024 | null | method | method | true | cited | 3 | cited | Parallel context encoding method (CEPE), the same parallel-encoding problem the FARS paper tunes, same kind. | null | year<=2024 | null | pass | null | null | null | null | null | 1,973 | 3,994 | 2.02 | Efficient inference, architecture, and training | null | null | null | null | null | null | shipped_tex |
FA0112 | fars | AI_FA0112 | HU_2405.18137 | Interval-Calibrated Noisy Quantization: A Parameter-Free Defense Against Quantization-Gap Attacks | Exploiting LLM Quantization | 2405.18137 | NeurIPS 2024 | 2,024 | null | method | method | true | cited | 3 | parent | The quantization-gap attack the FARS paper defends against; same security problem, method contribution. | null | year<=2024 | null | pass | null | null | null | null | null | 2,533 | 5,054 | 2 | Safety, security, and privacy | null | null | null | null | null | null | shipped_tex |
FA0114 | fars | AI_FA0114 | HU_2305.10601 | Sketch-Gated Trace Clustering for Accelerating Inter-Trace Redundancy Pruning | Tree of Thoughts: Deliberate Problem Solving with Large Language Models | 2305.10601 | NeurIPS 2023 | 2,023 | null | method | method | true | cited | 1 | cited | Tree-structured inference framework for reasoning; same test-time reasoning area, different problem. | null | year<=2024 | null | pass | null | null | null | null | null | 2,211 | 4,666 | 2.11 | Efficient inference, architecture, and training | null | null | null | null | null | null | shipped_tex |
FA0115 | fars | AI_FA0115 | HU_2412.15287 | OCR-Anchor Reranking: When Best-of-N Selection Fails Due to Candidate Homogeneity | Inference-Aware Fine-Tuning for Best-of-N Sampling in Large Language Models | 2412.15287 | ICLR 2024 | 2,024 | null | method | method | true | cited | 3 | cited | Method to make best-of-N selection work better, same BoN problem and kind. | null | year<=2024 | null | pass | null | null | null | null | null | 1,715 | 4,520 | 2.64 | Evaluation, judging, and verification | null | null | null | null | null | null | shipped_tex |
FA0116 | fars | AI_FA0116 | HU_2410.02355 | Fact-Check Grounding Loss for Semantically Consistent Model Editing | AlphaEdit: Null-Space Constrained Knowledge Editing for Language Models | 2410.02355 | ICLR 2024 | 2,024 | null | method | method | true | cited | 3 | cited | Null-space constrained knowledge-editing method, same model-editing problem and kind. | null | year<=2024 | null | pass | null | null | null | null | null | 1,690 | 3,687 | 2.18 | Continual learning and model editing | null | null | null | null | null | null | shipped_tex |
FA0121 | fars | AI_FA0121 | HU_2512.23447 | Counterfactual Gate Supervision Does Not Fix Gating Credit Assignment in Engram-Style Conditional Memory | Coupling Experts and Routers in Mixture-of-Experts via an Auxiliary Loss | 2512.23447 | ICLR 2026 oral | 2,025 | 6.667 | method | method | true | cited | 3 | cited | Auxiliary loss aligning router/gating decisions with module capabilities, same gating credit-assignment problem and kind. | null | pangram | null | pass | null | null | null | null | 0 | 1,958 | 3,576 | 1.83 | Efficient inference, architecture, and training | null | null | null | null | null | null | shipped_tex |
FA0123 | fars | AI_FA0123 | HU_2305.16264 | Compute-Matched Repetition Advantage in Long-CoT Supervised Fine-Tuning | Scaling Data-Constrained Language Models | 2305.16264 | NeurIPS 2023 | 2,023 | null | analysis | analysis | true | cited | 3 | cited | Empirical study of the value of repeated data under compute budgets, same repetition-vs-unique-data problem and kind. | null | year<=2024 | null | pass | null | null | null | null | null | 2,199 | 4,047 | 1.84 | Post-training and reasoning | null | null | null | null | null | null | shipped_tex |
FA0127 | fars | AI_FA0127 | HU_1903.05134 | Budget-Distilled ES-SSM: Cross-Budget Knowledge Distillation for Elastic Spectral State Space Models | Universally Slimmable Networks and Improved Training Techniques | 1903.05134 | ICCV 2019 | 2,019 | null | method | method | true | cited | 3 | cited | Arbitrary-width slimmable nets with in-place distillation, the same cross-budget distillation idea; same problem and contribution | null | year<=2024 | null | pass | null | null | null | null | null | 2,035 | 4,354 | 2.14 | Efficient inference, architecture, and training | null | null | null | null | null | null | shipped_tex |
FA0134 | fars | AI_FA0134 | HU_2006.16668 | Post-hoc Top-$p$ Expert Routing for Dynamic Compute Allocation in Mixture-of-Experts Language Models | GShard: Scaling Giant Models with Conditional Computation and Automatic Sharding | 2006.16668 | ICLR 2020 | 2,020 | null | method | method | true | cited | 1 | cited | MoE scaling/sharding infrastructure; same MoE area, different problem than inference-time adaptive expert counts | null | year<=2024 | null | pass | null | null | null | null | null | 1,959 | 11,313 | 5.77 | Efficient inference, architecture, and training | null | null | null | null | null | null | shipped_tex |
FA0137 | fars | AI_FA0137 | HU_1902.10416 | GaugeFix-LRM: Function-Preserving Q/K Gauge Fixing for Learnable Multipliers in Language Model Training | Equi-normalization of Neural Networks | 1902.10416 | ICLR 2019 | 2,019 | null | method | method | true | cited | 3 | cited | Handles rescaling symmetry by explicit norm-balancing instead of weight decay; same symmetry-control problem, method contribution | null | year<=2024 | null | pass | null | null | null | null | null | 2,148 | 3,903 | 1.82 | Efficient inference, architecture, and training | null | null | null | null | null | null | shipped_tex |
FA0141 | fars | AI_FA0141 | HU_2112.09118 | BH-Exit: Label-Free Early Termination for HNSW Search via Bucket-Histogram Stability | Unsupervised Dense Information Retrieval with Contrastive Learning | 2112.09118 | TMLR 2021 | 2,021 | null | method | method | true | cited | 1 | cited | Contriever trains dense retrievers; same retrieval area, different problem from ANN early termination. | null | year<=2024 | null | pass | null | null | null | null | null | 1,788 | 5,670 | 3.17 | Retrieval, grounding, and embeddings | null | null | null | null | null | null | shipped_tex |
FA0142 | fars | AI_FA0142 | HU_2508.10111 | Progress-Guarded LAVE: Lexer-Ignored Stall Filtering for Reliable CFG-Constrained Diffusion Decoding | Constrained Decoding of Diffusion LLMs with Context-Free Grammars | 2508.10111 | ICLR 2026 poster | 2,025 | 5.5 | method | method | true | cited | 3 | parent | CFG-constrained decoding for diffusion LLMs (the LAVE base); same problem and same method-type contribution | null | pangram | null | pass | null | null | null | null | 0 | 1,993 | 2,503 | 1.26 | Efficient inference, architecture, and training | null | null | null | null | null | null | shipped_tex |
FA0143 | fars | AI_FA0143 | HU_2106.07682 | Tuned-Lens-Style Affine Alignment for Encoder Truncation in Whisper ASR: An Empirical Investigation | Revisiting Model Stitching to Compare Neural Representations | 2106.07682 | NeurIPS 2021 | 2,021 | null | analysis | analysis | true | cited | 1 | cited | Model stitching to compare representations; same representation-alignment area, different problem than truncation speedup | null | year<=2024 | null | pass | null | null | null | null | null | 1,490 | 5,633 | 3.78 | Vision, multimodal, and applied domains | null | null | null | null | null | null | shipped_tex |
FA0147 | fars | AI_FA0147 | HU_2207.05833 | Quantile Remap Calibration for Precipitation Nowcasting | Earthformer: Exploring Space-Time Transformers for Earth System Forecasting | 2207.05833 | NeurIPS 2022 | 2,022 | null | method | method | true | cited | 3 | parent | Earthformer space-time transformer for Earth forecasting; same nowcasting problem, method contribution (also the calibrated base model) | null | year<=2024 | null | pass | null | null | null | null | null | 2,132 | 3,910 | 1.83 | Vision, multimodal, and applied domains | null | null | null | null | null | null | shipped_tex |
FA0150 | fars | AI_FA0150 | HU_2406.09519 | ShallowPPL: Investigating Early-Exit Logit Lens for Code Context Compression | Talking Heads: Understanding Inter-layer Communication in Transformer Language Models | 2406.09519 | NeurIPS 2024 | 2,024 | null | analysis | analysis | true | cited | 1 | cited | Interpretability of inter-layer information flow; same intermediate-representation area, different problem | null | year<=2024 | null | pass | null | null | null | null | null | 2,073 | 4,670 | 2.25 | Efficient inference, architecture, and training | null | null | null | null | null | null | shipped_tex |
FA0151 | fars | AI_FA0151 | HU_2404.02948 | MidPC LoRA: Intermediate SVD Slices for Continual Learning with Low-Rank Adaptation | PiSSA: Principal Singular Values and Singular Vectors Adaptation of Large Language Models | 2404.02948 | NeurIPS 2024 | 2,024 | null | method | method | true | cited | 3 | cited | SVD-based LoRA initialization method (top components), the direct spectral-endpoint baseline MidPC extends. | null | year<=2024 | null | pass | null | null | null | null | null | 1,763 | 3,517 | 1.99 | Continual learning and model editing | null | null | null | null | null | null | shipped_tex |
FA0153 | fars | AI_FA0153 | HU_2410.20056 | Fielded Max-Sim Keying for Assistant-Side Memory Recall in Long-Term Conversational Assistants | Multi-Field Adaptive Retrieval | 2410.20056 | ICLR 2024 | 2,024 | null | method | method | true | cited | 3 | cited | Multi-field retrieval method scoring fields separately, same fielded-retrieval problem and kind. | null | year<=2024 | null | pass | null | null | null | null | null | 1,734 | 4,382 | 2.53 | Agents, memory, and code | null | null | null | null | null | null | shipped_tex |
FA0156 | fars | AI_FA0156 | HU_2203.15556 | Length-Weighted Loss Does Not Explain the Repetition Advantage in Long-CoT Supervised Fine-Tuning | Training Compute-Optimal Large Language Models | 2203.15556 | NeurIPS 2022 | 2,022 | null | analysis | analysis | true | cited | 1 | cited | Compute-optimal scaling study for pretraining, same data-scaling area, different problem. | null | year<=2024 | null | pass | null | null | null | null | null | 1,925 | 4,509 | 2.34 | Post-training and reasoning | null | null | null | null | null | null | shipped_tex |
FA0161 | fars | AI_FA0161 | HU_2404.10774 | Speaker-Attested Grounding for False Memory Resistance in Agent Memory Systems | MiniCheck: Efficient Fact-Checking of LLMs on Grounding Documents | 2404.10774 | EMNLP 2024 | 2,024 | null | method | method | true | cited | 1 | cited | Fact-checking model for grounding LLM output, same verification area but not the agent-memory false-storage problem. | null | year<=2024 | null | pass | null | null | null | null | null | 1,914 | 4,379 | 2.29 | Agents, memory, and code | null | null | null | null | null | null | shipped_tex |
FA0162 | fars | AI_FA0162 | HU_2404.14469 | Training-Free Linear Routing for Sparse Attention via Attention-Mass Prediction | SnapKV: LLM Knows What You are Looking for Before Generation | 2404.14469 | NeurIPS 2024 | 2,024 | null | method | method | true | cited | 3 | cited | Training-free KV selection/compression for efficient long-context inference, same problem and kind. | null | year<=2024 | null | pass | null | null | null | null | null | 2,110 | 4,590 | 2.18 | Efficient inference, architecture, and training | null | null | null | null | null | null | shipped_tex |
FA0163 | fars | AI_FA0163 | HU_2304.05128 | Execution-Signature Recycling: Deduplicating Unit-Test Failure Feedback for Test-Time Code Scaling | Teaching Large Language Models to Self-Debug | 2304.05128 | ICLR 2023 | 2,023 | null | method | method | true | cited | 3 | parent | Execution-feedback self-debugging for code generation, same problem and kind, the direct baseline. | null | year<=2024 | null | pass | null | null | null | null | null | 1,820 | 4,542 | 2.5 | Agents, memory, and code | null | null | null | null | null | null | shipped_tex |
FA0168 | fars | AI_FA0168 | HU_2004.10964 | Token-Balanced Continual Pretraining Eliminates Brain Rot Degradation | Don't Stop Pretraining: Adapt Language Models to Domains and Tasks | 2004.10964 | ACL 2020 | 2,020 | null | method | method | true | cited | 3 | cited | Continued/domain-adaptive pretraining strategies, same continual-pretraining problem and kind. | null | year<=2024 | null | pass | null | null | null | null | null | 1,843 | 3,637 | 1.97 | Post-training and reasoning | null | null | null | null | null | null | shipped_tex |
FA0171 | fars | AI_FA0171 | HU_2310.02575 | SourceJS-LoRA: Source-Referenced Jensen-Shannon Divergence for Learning LoRA Merge Coefficients | AdaMerging: Adaptive Model Merging for Multi-Task Learning | 2310.02575 | ICLR 2024 | 2,023 | null | method | method | true | cited | 3 | parent | Learns merge coefficients via entropy minimization, the exact problem and kind, direct parent baseline. | null | year<=2024 | null | pass | null | null | null | null | null | 1,751 | 4,237 | 2.42 | Continual learning and model editing | null | null | null | null | null | null | shipped_tex |
FA0172 | fars | AI_FA0172 | HU_2402.07841 | Auditing Norm-Clipped L2-Laplacian Token-Embedding Obfuscation Against Sequence-Aware Reconstruction | Do Membership Inference Attacks Work on Large Language Models? | 2402.07841 | Accepted at Conference on Language Model | 2,024 | null | analysis | analysis | true | cited | 1 | cited | Large-scale evaluation of MIAs on LLMs, same privacy-evaluation area, different problem. | null | year<=2024 | null | pass | null | null | null | null | null | 2,284 | 3,504 | 1.53 | Safety, security, and privacy | null | null | null | null | null | null | shipped_tex |
FA0174 | fars | AI_FA0174 | HU_2005.13239 | Action-Support Likelihood Audits Predict Rollout Consistency Failures in Text-Based World Models | MOPO: Model-based Offline Policy Optimization | 2005.13239 | NeurIPS 2020 | 2,020 | null | method | method | true | cited | 3 | cited | Penalizes model rollouts where dynamics are unreliable under distribution shift, same problem and kind. | null | year<=2024 | null | pass | null | null | null | null | null | 1,536 | 5,025 | 3.27 | Agents, memory, and code | null | null | null | null | null | null | shipped_tex |
FA0175 | fars | AI_FA0175 | HU_2205.04605 | Distance-Hiding Fingerprints for Text Embeddings via Secure SimHash | Sentence-level Privacy for Document Embeddings | 2205.04605 | ACL 2022 | 2,022 | null | method | method | true | cited | 3 | cited | Privacy-preserving text embedding scheme with formal guarantees, same problem and kind. | null | year<=2024 | null | pass | null | null | null | null | null | 1,715 | 3,898 | 2.27 | Safety, security, and privacy | null | null | null | null | null | null | shipped_tex |
FA0181 | fars | AI_FA0181 | HU_2309.00071 | Data-Free Transition-Spectrum Winsorization for Mamba Long-Context Generalization | YaRN: Efficient Context Window Extension of Large Language Models | 2309.00071 | ICLR 2023 | 2,023 | null | method | method | true | cited | 3 | cited | Lightweight modification extending context generalization of pretrained models, same problem and kind. | null | year<=2024 | null | pass | null | null | null | null | null | 1,673 | 3,645 | 2.18 | Efficient inference, architecture, and training | null | null | null | null | null | null | shipped_tex |
FA0184 | fars | AI_FA0184 | HU_2411.11925 | Velocity-Forecast Sampling for Flow-Matching Heads: A Negative Result | Continuous Speculative Decoding for Autoregressive Image Generation | 2411.11925 | ECCV 2026 | 2,024 | 4 | method | method | true | cited | 3 | cited | Also accelerates continuous visual AR inference via speculative reuse of predictions | null | pangram | null | pass | null | null | null | null | 0 | 1,978 | 2,783 | 1.41 | Vision, multimodal, and applied domains | null | null | null | null | null | null | shipped_tex |
FA0186 | fars | AI_FA0186 | HU_2510.06213 | 8-bit Quantization Provides No Privacy Benefit Against Training-Free Embedding Inversion | Training Dynamics Impact Post-Training Quantization Robustness | 2510.06213 | ICLR 2026 poster | null | 5.5 | analysis | analysis | true | widened_iclr26 | 2 | widened (not cited; topic neighbour) | comprehensive analysis of quantization effects; same quantization-consequences question, same kind | pass (pangram) | null | pass | pass | pass | null | 0.1 | null | 0 | 1,581 | 4,350 | null | Safety, security, and privacy | null | null | null | null | null | null | shipped_tex |
FA0187 | fars | AI_FA0187 | HU_2603.03226 | Differentially Private Eigenspectrum Monitor Logs for Hallucination Detection | Adaptive Methods Are Preferable in High Privacy Settings: An SDE Perspective | 2603.03226 | ICLR 2026 poster | null | 5 | analysis | analysis | true | widened_iclr26 | 2 | widened (not cited; topic neighbour) | SDE analysis of DP noise interacting with optimization; same DP-mechanism-behavior question, same kind | pass (pangram) | null | pass | pass | pass | null | 0.176 | null | 0.035022 | 2,121 | 4,430 | null | Safety, security, and privacy | null | null | null | null | null | null | shipped_tex |
FA0188 | fars | AI_FA0188 | HU_2408.15621 | Differentially Private Spectral Monitor Logs for Hallucination Detection: A Comparative Study of Wishart and Gaussian Mechanisms | Convergent Differential Privacy Analysis for General Federated Learning | 2408.15621 | ICLR 2026 poster | null | 6 | analysis | analysis | true | widened_iclr26 | 2 | widened (not cited; topic neighbour) | convergent DP analysis in federated learning; same DP-mechanism-analysis kind | pass (pangram) | null | pass | pass | pass | null | 0.121 | null | 0 | 1,702 | 3,617 | null | Safety, security, and privacy | null | null | null | null | null | null | shipped_tex |
FA0190 | fars | AI_FA0190 | HU_1206.2944 | Paired Median-of-Means Rewards for Robust Configuration Selection in Vector Search Benchmarking | Practical Bayesian Optimization of Machine Learning Algorithms | 1206.2944 | NeurIPS 2012 | 2,012 | null | method | method | true | cited | 1 | cited | Configuration/hyperparameter selection area, but about search strategy, not noise-robust measurement | null | year<=2024 | null | pass | null | null | null | null | null | 2,509 | 4,125 | 1.64 | Evaluation, judging, and verification | null | null | null | null | null | null | shipped_tex |
FA0191 | fars | AI_FA0191 | HU_2512.23017 | HeadRollback: Post-Task Attention Head Rollback for Replay-Free Continual LoRA Fine-Tuning | Merge before Forget: A Single LoRA Continual Learning via Continual Merging | 2512.23017 | ICLR 2026 poster | 2,025 | 5 | method | method | true | cited | 3 | cited | LoRA continual learning method against forgetting, same problem and contribution kind | null | pangram | null | pass | null | null | null | null | 0 | 1,841 | 4,604 | 2.5 | Continual learning and model editing | null | null | null | null | null | null | shipped_tex |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.