5/12/2026
One Trajectory, One Token: Grounded Video Tokenization via Panoptic Sub-object Trajectory
Novel video tokenization approach using panoptic sub-object trajectories to reduce tokens for transformer-based long video processing.
5/12/2026
FeDa4Fair: Client-Level Federated Datasets for Fairness Evaluation
arXiv paper on federated learning fairness evaluation, addressing discrimination at client level despite global model fairness.
5/12/2026
Causal Discovery for Irregularly Time Series with Consistency Guarantees
Causal discovery method for irregularly sampled time series with consistency guarantees, addressing missing data challenges.
5/12/2026
Reinforcement Learning with Action Chunking
Q-chunking algorithm for improving reinforcement learning on long-horizon sparse-reward tasks in offline-to-online settings.
5/12/2026
TinyTroupe: An LLM-powered Multiagent Persona Simulation Toolkit
TinyTroupe toolkit for LLM-powered multiagent persona simulation with fine-grained specifications for realistic human behavior simulation.
5/12/2026
Temperature and Persona Shape LLM Agent Consensus With Minimal Accuracy Gains in Qualitative Coding
Experimental study of multi-agent LLM systems for qualitative coding, examining how persona and temperature affect consensus-building vs single agents.
5/12/2026
Position: Stop Evaluating AI with Human Tests, Develop Principled, AI-specific Tests instead
Argues LLM evaluations on human-designed tests are ontologically flawed; proposes AI-specific evaluation frameworks instead of anthropomorphizing model performance.
5/12/2026
MECAT: A Multi-Experts Constructed Benchmark for Fine-Grained Audio Understanding Tasks
MECAT benchmark for fine-grained audio understanding using multi-expert annotations to distinguish detailed model outputs.
5/12/2026
EventTSF: Event-Aware Non-Stationary Time Series Forecasting
EventTSF integrates natural language events with time series data for improved non-stationary forecasting.
5/12/2026
Assessing Trustworthiness of AI Training Dataset using Subjective Logic -- A Use Case on Bias
Subjective logic framework for assessing trustworthiness of AI training datasets regarding bias and fairness.
5/12/2026
Multimodal Representation Learning Conditioned on Semantic Relations
Multimodal representation learning conditioned on semantic relations beyond single shared embeddings like CLIP.
5/12/2026
GUARD: Guideline Upholding Test through Adaptive Role-play and Jailbreak Diagnostics for LLMs
GUARD testing framework translating ethics guidelines into actionable jailbreak diagnostics for LLM safety evaluation.
5/12/2026
Scam2Prompt: A Scalable Framework for Auditing Malicious Scam Endpoints in Production LLMs
Scam2Prompt framework auditing malicious endpoints in production LLMs for security vulnerabilities from training data.
5/12/2026
Top-H Decoding: Adapting the Creativity and Coherence with Bounded Entropy in Text Generation
Top-H decoding method balancing creativity and coherence in LLM text generation via bounded entropy.
5/12/2026
AU-Harness: An Open-Source Toolkit for Holistic Evaluation of Audio LLMs
AU-Harness open-source toolkit for standardized evaluation of audio language models with multi-turn dialogue support.
5/12/2026
RL Fine-Tuning Heals OOD Forgetting in SFT
Analysis of why SFT+RL post-training works: SFT peaks on OOD early then declines; RL recovers generalization.
5/12/2026
Positional Encoding via Token-Aware Phase Attention
Token-Aware Phase Attention (TAPA) positional encoding improving long-context modeling beyond RoPE limitations.
5/12/2026
Elastic MoE: Unlocking the Inference-Time Scalability of Mixture-of-Experts
Elastic MoE method enabling runtime scaling of activated experts for heterogeneous hardware and varying workloads.
5/12/2026
Benchmarking and Mitigating Sycophancy in Medical Vision Language Models
Benchmark and mitigation techniques for sycophancy in medical vision language models for visual QA tasks.
5/12/2026
GLAI: GreenLightningAI for Accelerated Training through Knowledge Decoupling
GLAI architectural block separating structural and quantitative knowledge for accelerated neural network training.
5/12/2026
You Have Been LaTeXpOsEd: A Systematic Analysis of Information Leakage in Preprint Archives Using Large Language Models
Security analysis of information leakage from LaTeX sources and metadata in arXiv preprints using LLMs.
5/12/2026
Turning Drift into Constraint: Robust Reasoning Alignment in Non-Stationary Multi-Stream Environments
Multi-source reasoning alignment for MLLMs addressing concept drift in non-stationary environments via constraint satisfaction.
5/12/2026
PAINET: A Principled Efficient Transformer for 3D Dynamics Modeling
PAINET transformer for 3D dynamics modeling in multi-body systems using geometric symmetries for trajectory prediction.
5/12/2026
EverydayMMQA: A Multilingual and Multimodal Framework for Culturally Grounded Spoken Visual QA
EverydayMMQA dataset for multilingual/multimodal visual QA with cultural grounding using OASIS framework for low-resource languages.
5/12/2026
Control-Augmented Autoregressive Diffusion for Data Assimilation
Framework augmenting auto-regressive diffusion models with offline-trained controllers for guided generation via stepwise corrections.
5/12/2026
Less Diverse, Less Safe: The Indirect But Pervasive Risk of Test-Time Scaling in Large Language Models
Test-time scaling in LLMs reduces safety and reliability when candidate diversity decreases, revealing failure mode.
5/12/2026
When a Robot is More Capable than a Human: Learning from Constrained Demonstrators
Learning from constrained demonstrations for robots with limited control interfaces via policy learning.
5/12/2026
Batch Bayesian Active Learning with Partial Batch Label Sampling
Batch Bayesian active learning with partial label sampling addressing scalability of acquisition functions.
5/12/2026
Users as Annotators: LLM Preference Learning from Comparison Mode
User-in-the-loop pairwise preference collection for LLM alignment training without professional annotators.
5/12/2026
GenCellAgent: Generalizable, Training-Free Cellular Image Segmentation via Large Language Model Agents
GenCellAgent multi-agent framework for training-free cellular image segmentation using vision-language models with planner-executor-evaluator loop.
5/12/2026
LILO: Bayesian Optimization with Natural Language Feedback
LILO framework uses LLMs to translate natural language feedback into structured preference signals for Bayesian optimization.
5/12/2026
ActivationReasoning: Logical Reasoning in Latent Activation Spaces
ActivationReasoning framework enables systematic logical reasoning in LLM latent spaces using sparse autoencoders for interpretability.
5/12/2026
SAID: Safety-Aware Intent Defense via Prefix Probing for Large Language Models
SAID defense mechanism against jailbreaks via prefix probing for safety-aligned LLMs without external filtering.
5/12/2026
No Mean Feat: Simple, Strong Baselines for Context Compression
BenchPress evaluation suite and strong simple baselines for context compression in retrieval-augmented generation systems.
5/12/2026
TetraJet-v2: Accurate NVFP4 Training for Large Language Models with Oscillation Suppression and Outlier Control
TetraJet-v2 enables 4-bit fully-quantized LLM training using NVFP4 format with oscillation suppression and outlier control.
5/12/2026
Forget BIT, It is All about TOKEN: Towards Semantic Information Theory for LLMs
Semantic Information Theory for LLMs replacing classical information theory bits with tokens, developing first-principles theory.
5/12/2026
Continuum: Efficient and Robust Multi-Turn LLM Agent Scheduling with KV Cache Time-to-Live
Continuum improves multi-turn LLM agent inference efficiency via KV cache time-to-live management across tool-interleaved calls.
5/12/2026
COGNOS: Universal Enhancement for Time Series Anomaly Detection via Constrained Gaussian-Noise Optimization and Smoothing
COGNOS addresses MSE loss limitations in time series anomaly detection via constrained Gaussian-noise optimization and smoothing.
5/12/2026
MURPHY: Feedback-Aware GRPO with Retrospective Credit Assignment for Multi-Turn Code Generation
MURPHY improves GRPO training with feedback-aware retrospective credit assignment for multi-turn agentic code generation tasks.
5/12/2026
Energy Consumption of Dataframe Libraries for End-to-End Deep Learning Pipelines:A Comparative Analysis
Comparative analysis of Pandas, Polars, and Dask dataframe libraries in deep learning pipelines focusing on energy consumption and GPU interaction.
5/12/2026
Auditing Data Membership in Reinforcement Learning With Verifiable Rewards
Membership inference attacks to audit unauthorized data use in RLVR training pipelines used in LLM post-training.
5/12/2026
Commanding Humanoid by Free-form Language: A Large Language Action Model with Unified Motion Vocabulary
Large Language Action model (Humanoid-LLA) that translates free-form natural language commands to robotic humanoid control with diverse motions.
5/12/2026
GraphBench: Next-generation graph learning benchmarking
GraphBench provides standardized benchmarking suite for graph machine learning with consistent evaluation protocols for foundation models.
5/12/2026
CARL: Criticality-Aware Agentic Reinforcement Learning
CARL improves multi-step reinforcement learning for agents by identifying and optimizing criticality-aware action choices rather than treating all steps equally.
5/12/2026
When Tables Leak: Attacking String Memorization in LLM-Based Tabular Data Generation
Security analysis showing LLM-based tabular data generation systems memorize and leak sensitive string information from training data.
5/12/2026
RAG-HAR: Retrieval Augmented Generation-based Human Activity Recognition
RAG-HAR applies retrieval-augmented generation with LLMs for human activity recognition without dataset-specific training.
5/12/2026
Can LLMs Estimate Student Struggles? Human-AI Difficulty Alignment with Proficiency Simulation for Item Difficulty Prediction
Analysis of whether LLMs can estimate question difficulty and perceive student learning struggles for educational assessment.
5/12/2026
Selective LoRA for Visual Tokens and Attention Heads
Parameter-efficient fine-tuning method applying LoRA selectively to visual tokens and attention heads in vision-language models.
5/12/2026
Schoenfeld's Anatomy of Mathematical Reasoning by Language Models
Framework (ThinkARM) for analyzing and abstracting LLM reasoning traces using episode theory into functional reasoning steps.
5/12/2026
LENS: LLM-Enabled Narrative Synthesis for Mental Health by Aligning Multimodal Sensing with Language Models
Framework aligning multimodal health sensor data with LLMs to generate clinical narratives for mental health assessment.