Abhimanyu Dubey et al

representative citing papers

citing papers explorer

Showing 50 of 75 citing papers.

• Efficient Training on Multiple Consumer GPUs with RoundPipe cs.DC · 2026-04-29 · conditional · none · ref 9

  RoundPipe achieves near-zero-bubble pipeline parallelism for LLM training on consumer GPUs by dynamically dispatching computation stages round-robin, yielding 1.48-2.16x speedups and enabling 235B model fine-tuning on 8x RTX 4090.

• Stability and Generalization in Looped Transformers cs.LG · 2026-04-16 · unverdicted · none · ref 8

  Looped transformers with recall and outer normalization produce reachable, input-dependent fixed points with stable gradients, enabling generalization, while those without recall cannot; a new internal recall variant performs competitively or better.

• AutoLLMResearch: Training Research Agents for Automating LLM Experiment Configuration -- Learning from Cheap, Optimizing Expensive cs.AI · 2026-05-12 · unverdicted · none · ref 31

  AutoLLMResearch trains agents via a multi-fidelity environment and MDP pipeline to extrapolate configuration principles from inexpensive to costly LLM experiments.

• Reflection Anchors for Propagation-Aware Visual Retention in Long-Chain Multimodal Reasoning cs.CV · 2026-05-10 · unverdicted · none · ref 3

  RAPO uses an information-theoretic lower bound on visual gain to select high-entropy reflection anchors and optimizes a chain-masked KL surrogate, delivering gains over baselines on reasoning benchmarks across LVLM backbones.

• UniShield: Unified Face Attack Detection via KG-Informed Multimodal Reasoning cs.CV · 2026-05-09 · unverdicted · none · ref 8

  UniShield introduces a knowledge-graph-informed multimodal framework that improves unified detection of physical and digital face attacks through instruction tuning and consistency-optimized reasoning.

• Tracing Uncertainty in Language Model "Reasoning" cs.LG · 2026-05-08 · unverdicted · none · ref 11

  Uncertainty trace profiles from LM reasoning traces predict correct final answers with AUROC up to 0.807 and enable early error detection using only initial tokens.

• Your Language Model is Its Own Critic: Reinforcement Learning with Value Estimation from Actor's Internal States cs.LG · 2026-05-08 · unverdicted · none · ref 10 · 2 links

  POISE trains a lightweight probe on the actor's internal states to predict expected rewards for RLVR, matching DAPO performance on math benchmarks with lower compute by avoiding extra rollouts or critic models.

• LaTER: Efficient Test-Time Reasoning via Latent Exploration and Explicit Verification cs.CL · 2026-05-08 · unverdicted · none · ref 2

  LaTER reduces LLM token usage 16-33% on reasoning benchmarks by exploring in latent space then switching to explicit CoT verification, with gains like 70% to 73.3% on AIME 2025 in the training-free version.

• TableVista: Benchmarking Multimodal Table Reasoning under Visual and Structural Complexity cs.CL · 2026-05-07 · unverdicted · none · ref 10

  TableVista benchmark finds foundation models maintain performance across visual styles but degrade sharply on complex table structures and vision-only settings.

• Nonsense Helps: Prompt Space Perturbation Broadens Reasoning Exploration cs.AI · 2026-05-07 · unverdicted · none · ref 13

  Prepending stochastic sequences from Lorem Ipsum vocabulary to prompts during GRPO resampling broadens reasoning exploration and outperforms standard resampling on hard tasks for 1.7B-7B models.

• DiagramNet: An End-to-End Recognition Framework and Dataset for Non-Standard System-Level Diagrams cs.AI · 2026-05-02 · unverdicted · none · ref 4

  DiagramNet supplies a new multimodal dataset and progressive training pipeline with decoupled multi-agent workflow, allowing a 3B model to outperform GPT-5, Claude-Sonnet-4, and Gemini-2.5-Pro by over 2x on system-level diagram tasks while generalizing to other benchmarks.

• BoostLoRA: Growing Effective Rank by Boosting Adapters cs.LG · 2026-04-30 · unverdicted · none · ref 12

  BoostLoRA grows effective adapter rank linearly via iterative boosting on hard examples with orthogonal low-rank updates, outperforming both single-shot ultra-low-rank adapters and full fine-tuning on math and code tasks with zero added inference overhead.

• When to Retrieve During Reasoning: Adaptive Retrieval for Large Reasoning Models cs.IR · 2026-04-29 · unverdicted · none · ref 5

  ReaLM-Retrieve uses step-level uncertainty to trigger retrievals during reasoning, achieving 10.1% better F1 scores and 47% fewer calls on multi-hop QA benchmarks.

• TRIP-Evaluate: An Open Multimodal Benchmark for Evaluating Large Models in Transportation cs.CV · 2026-04-29 · accept · none · ref 34

  TRIP-Evaluate is a new open multimodal benchmark with 837 text, image, and point-cloud items organized by a role-task-knowledge taxonomy to evaluate large models on transportation workflows.

• Thinking Without Words: Efficient Latent Reasoning with Abstract Chain-of-Thought cs.CL · 2026-04-24 · unverdicted · none · ref 11

  Abstract-CoT lets models reason with short discrete latent token sequences from a reserved vocabulary, using warm-up training and RL to match verbal CoT performance with up to 11.6x fewer tokens.

• EVPO: Explained Variance Policy Optimization for Adaptive Critic Utilization in LLM Post-Training cs.LG · 2026-04-21 · unverdicted · none · ref 11

  EVPO adaptively switches between critic-based and batch-mean advantage estimation using batch-level explained variance to provably achieve no greater variance than the better of PPO or GRPO at every step.

• Wiring the 'Why': A Unified Taxonomy and Survey of Abductive Reasoning in LLMs cs.AI · 2026-04-09 · accept · none · ref 34

  The paper delivers the first survey of abductive reasoning in LLMs, a unified two-stage taxonomy, a compact benchmark, and an analysis of gaps relative to deductive and inductive reasoning.

• Generate, Filter, Control, Replay: A Comprehensive Survey of Rollout Strategies for LLM Reinforcement Learning cs.LG · 2026-04-08 · unverdicted · none · ref 28

  This survey introduces the Generate-Filter-Control-Replay (GFCR) taxonomy to structure rollout pipelines for RL-based post-training of reasoning LLMs.

• MMORF: A Multi-agent Framework for Designing Multi-objective Retrosynthesis Planning Systems cs.AI · 2026-04-06 · unverdicted · none · ref 16

  MMORF provides a modular multi-agent framework for multi-objective retrosynthesis planning, with MASIL and RFAS systems showing strong safety, cost, and success metrics on a new 218-task benchmark.

• ODRPO: Ordinal Decompositions of Discrete Rewards for Robust Policy Optimization cs.LG · 2026-05-12 · unverdicted · none · ref 1

  ODRPO decomposes discrete rewards into ordinal binary indicators to compute independent advantages and reduce noise corruption in RLAIF policy optimization.

• Learning, Fast and Slow: Towards LLMs That Adapt Continually cs.LG · 2026-05-12 · unverdicted · none · ref 19

  Fast-Slow Training combines slow parameter updates with fast context optimization to achieve up to 3x better sample efficiency, higher performance, less forgetting, and preserved plasticity in continual LLM learning.

• Solve the Loop: Attractor Models for Language and Reasoning cs.LG · 2026-05-12 · unverdicted · none · ref 42

  Attractor Models solve for fixed points in transformer embeddings using implicit differentiation to enable stable iterative refinement, delivering better perplexity, accuracy, and efficiency than standard or looped transformers.

• Enhancing Multilingual Counterfactual Generation through Alignment-as-Preference Optimization cs.CL · 2026-05-12 · unverdicted · none · ref 52

  Macro uses Direct Preference Optimization on composite-scored preference pairs to improve validity of multilingual self-generated counterfactual explanations by 12.55% on average without degrading minimality.

• PRISM: A Geometric Risk Bound that Decomposes Drift into Scale, Shape, and Head cs.CL · 2026-05-12 · unverdicted · none · ref 25

  PRISM supplies a geometric upper bound on LLM variant risk that splits drift into scale, shape, and head axes and doubles as a differentiable regularizer against forgetting.

• Evaluating the False Trust engendered by LLM Explanations cs.HC · 2026-05-11 · unverdicted · none · ref 11

  A user study finds that LLM reasoning traces and post-hoc explanations create false trust by increasing acceptance of incorrect answers, whereas contrastive dual explanations improve users' ability to detect errors.

• Separate First, Fuse Later: Mitigating Cross-Modal Interference in Audio-Visual LLMs Reasoning with Modality-Specific Chain-of-Thought cs.AI · 2026-05-11 · unverdicted · none · ref 23

  Separate modality-specific reasoning before fusion reduces hallucinations and improves accuracy in audio-visual LLMs by enforcing isolated traces then integrating evidence.

• Crosslingual On-Policy Self-Distillation for Multilingual Reasoning cs.CL · 2026-05-10 · unverdicted · none · ref 2

  COPSD improves mathematical reasoning in low-resource languages by having LLMs self-distill from their own high-resource English behavior via token-level divergence on rollouts with privileged crosslingual context.

• Can Revealed Preferences Clarify LLM Alignment and Steering? cs.LG · 2026-05-08 · unverdicted · none · ref 4

  LLMs show partial internal coherence in medical decisions but frequently fail to accurately report their preferences or adopt user-directed ones via prompting.

• POETS: Uncertainty-Aware LLM Optimization via Compute-Efficient Policy Ensembles cs.LG · 2026-05-08 · unverdicted · none · ref 113

  POETS uses compute-efficient LLM policy ensembles to implicitly perform KL-regularized Thompson sampling, delivering O(sqrt(T gamma_T)) regret bounds and state-of-the-art sample efficiency in scientific discovery tasks such as protein search and quantum circuit design.

• Memory-Efficient Looped Transformer: Decoupling Compute from Memory in Looped Language Models cs.CL · 2026-05-08 · unverdicted · none · ref 50

  MELT decouples reasoning depth from memory in looped LLMs by sharing a single gated KV cache per layer and using two-phase chunk-wise distillation from Ouro, delivering constant memory use while matching or beating standard LLM performance.

• BalCapRL: A Balanced Framework for RL-Based MLLM Image Captioning cs.CV · 2026-05-08 · unverdicted · none · ref 11

  BalCapRL applies balanced multi-objective RL with GDPO-style normalization and length-conditional masking to improve MLLM image captioning, reporting gains of up to +13.6 DCScore, +9.0 CaptionQA, and +29.0 CapArena on LLaVA and Qwen models.

• A$^2$TGPO: Agentic Turn-Group Policy Optimization with Adaptive Turn-level Clipping cs.CL · 2026-05-07 · unverdicted · none · ref 5

  A²TGPO improves RL policy optimization for multi-turn agentic LLMs by normalizing information gain within same-depth turn groups, rescaling cumulative advantages by sqrt of term count, and modulating clipping ranges per turn's normalized IG.

• Milestone-Guided Policy Learning for Long-Horizon Language Agents cs.CL · 2026-05-07 · unverdicted · none · ref 31

  BEACON uses milestone partitioning, temporal reward shaping, and dual-scale advantage estimation to nearly double success rates on long-horizon ALFWorld tasks while raising effective sample use from 23.7% to 82%.

• FalconGEMM: Surpassing Hardware Peaks with Lower-Complexity Matrix Multiplication cs.DC · 2026-05-07 · unverdicted · none · ref 26 · 2 links

  FalconGEMM delivers a framework with deployment, group-parallel execution, and analytical decision modules that makes lower-complexity matrix multiplication practical, beating cuBLAS and similar libraries by 7.59-17.85% on LLM tasks.

• LoopTrap: Termination Poisoning Attacks on LLM Agents cs.CR · 2026-05-07 · unverdicted · none · ref 12

  LoopTrap is an automated red-teaming framework that crafts termination-poisoning prompts to amplify LLM agent steps by 3.57x on average (up to 25x) across 8 agents.

• RAG over Thinking Traces Can Improve Reasoning Tasks cs.IR · 2026-05-05 · unverdicted · none · ref 75

  RAG over structured thinking traces boosts LLM reasoning on AIME, LiveCodeBench, and GPQA, with relative gains up to 56% and little added cost.

• Spatiotemporal Hidden-State Dynamics as a Signature of Internal Reasoning in Large Language Models cs.CL · 2026-05-03 · unverdicted · none · ref 5

  Large reasoning models show measurable hidden-state dynamics that a new statistic can use to distinguish correct reasoning trajectories without labels.

• Iterative Finetuning is Mostly Idempotent cs.AI · 2026-05-01 · unverdicted · none · ref 5

  Iterative self-finetuning of LLMs mostly fails to amplify seeded behavioral traits, with amplification limited to specific DPO setups and often harming coherence.

• Evaluating the Architectural Reasoning Capabilities of LLM Provers via the Obfuscated Natural Number Game cs.LG · 2026-05-01 · unverdicted · none · ref 11

  The Obfuscated Natural Number Game shows reasoning LLMs keep proof accuracy without semantic cues while general models degrade, establishing a metric for architectural reasoning in alien math domains.

• State Stream Transformer (SST) V2: Parallel Training of Nonlinear Recurrence for Latent Space Reasoning cs.LG · 2026-04-30 · unverdicted · none · ref 9

  SST V2 introduces parallel-trainable nonlinear recurrence in latent space to let transformers reason continuously across positions, delivering +15 points on GPQA-Diamond and halving remaining GSM8K errors over matched baselines.

• A Survey of Reasoning-Intensive Retrieval: Progress and Challenges cs.IR · 2026-04-30 · unverdicted · none · ref 20

  A survey that categorizes RIR benchmarks by domain and modality, proposes a taxonomy for integrating reasoning into retrieval pipelines, and outlines key challenges.

• Do Synthetic Trajectories Reflect Real Reward Hacking? A Systematic Study on Monitoring In-the-Wild Hacking in Code Generation cs.LG · 2026-04-26 · unverdicted · none · ref 7

  Synthetic reward hacking data does not capture natural hacking behaviors in code generation RL, causing monitors trained on it to generalize poorly compared to those trained on in-the-wild trajectories.

• Learning to Communicate: Toward End-to-End Optimization of Multi-Agent Language Systems cs.AI · 2026-04-23 · unverdicted · none · ref 51

  DiffMAS jointly optimizes latent communication and reasoning in multi-agent LLM systems via parameter-efficient supervised training on trajectories, yielding consistent gains over baselines on math, science, and code benchmarks.

• Foveated Reasoning: Stateful, Action-based Visual Focusing for Vision-Language Models cs.CV · 2026-04-22 · unverdicted · none · ref 10

  Foveated Reasoner integrates foveation as stateful actions inside the autoregressive decoding loop of vision-language models, trained via cold-start supervision then reinforcement learning to achieve higher accuracy at low token budgets.

• HypEHR: Hyperbolic Modeling of Electronic Health Records for Efficient Question Answering cs.AI · 2026-04-22 · unverdicted · none · ref 17

  HypEHR is a hyperbolic embedding model for EHR data that uses Lorentzian geometry and hierarchy-aware pretraining to answer clinical questions nearly as well as large language models but with much smaller size.

• Chasing the Public Score: User Pressure and Evaluation Exploitation in Coding Agent Workflows cs.CL · 2026-04-22 · unverdicted · none · ref 7

  Coding agents under repeated user pressure to raise public scores frequently exploit those scores through shortcuts that fail to improve private evaluations, demonstrated via a new 34-task benchmark and 1326 trajectories.

• Unsupervised Confidence Calibration for Reasoning LLMs from a Single Generation cs.LG · 2026-04-21 · unverdicted · none · ref 165

  Unsupervised single-generation confidence calibration for reasoning LLMs via offline self-consistency proxy distillation outperforms baselines on math and QA tasks and improves selective prediction.

• Process Reward Models Meet Planning: Generating Precise and Scalable Datasets for Step-Level Rewards cs.CL · 2026-04-20 · unverdicted · none · ref 4

  PDDL planning problems are used to generate about one million precise reasoning steps for training Process Reward Models, and adding this data to existing datasets improves LLM performance on both mathematical and non-mathematical reasoning benchmarks.

• Learning to Correct: Calibrated Reinforcement Learning for Multi-Attempt Chain-of-Thought cs.LG · 2026-04-20 · unverdicted · none · ref 16

  CAL-GRPO calibrates per-attempt weights in multi-attempt CoT to deliver unbiased gradients for optimizing Verification@K success while keeping variance low.

• AutoOR: Scalably Post-training LLMs to Autoformalize Operations Research Problems cs.LG · 2026-04-18 · unverdicted · none · ref 40

  AutoOR uses synthetic data generation and RL post-training with solver feedback to enable 8B LLMs to autoformalize linear, mixed-integer, and non-linear OR problems, matching larger models on benchmarks.