4/21/2026
TIP: Token Importance in On-Policy Distillation
Token Importance analysis in on-policy knowledge distillation identifying which token positions provide most useful learning signals for student LLMs.
4/21/2026
Enhancing LLM-based Search Agents via Contribution Weighted Group Relative Policy Optimization
Contribution Weighted Group Relative Policy Optimization for training LLM-based search agents via reinforcement learning with improved credit assignment.
4/21/2026
Tight Sample Complexity Bounds for Best-Arm Identification Under Bounded Systematic Bias
Best-Arm Identification framework for autonomous reasoning and planning with LLM agents, addressing systematic evaluation biases in search expansion.
4/21/2026
Tabular foundation models for in-context prediction of molecular properties
Tabular foundation models for molecular property prediction using in-context learning without task-specific fine-tuning, improving drug discovery applications.
4/21/2026
Multimodal Sentiment Analysis with Missing Modality: A Knowledge-Transfer Approach
Knowledge-transfer network for multimodal sentiment analysis that reconstructs missing modalities during training and testing phases.
4/21/2026
Uncovering Logit Suppression Vulnerabilities in LLM Safety Alignment
Research on logit suppression vulnerabilities in LLM safety alignment, introducing SSAG method to systematically identify and manipulate safety mechanisms in large language models.
4/21/2026
ProTrain: Efficient LLM Training via Memory-Aware Techniques
Memory-aware techniques for efficient LLM training with automatic configuration in resource-constrained environments.
4/21/2026
Automatic Dataset Construction (ADC): Sample Collection, Data Curation, and Beyond
Automatic Dataset Construction methodology for large-scale data collection and curation with minimal human annotation.
4/21/2026
SFTMix: Elevating Language Model Instruction Tuning with Mixup Recipe
SFTMix method improving LLM instruction tuning using mixup data augmentation without requiring data filtering.
4/21/2026
Plasticity Loss in Deep Reinforcement Learning: A Survey
Survey of plasticity loss in deep reinforcement learning, examining causes and solutions for adaptation failures.
4/21/2026
Navigating Distribution Shifts in Medical Image Analysis: A Survey
Survey of distribution shift challenges and solutions in medical image analysis deep learning models.
4/21/2026
A Computational Method for Measuring "Open Codes" in Qualitative Analysis
Computational method for automating open code identification in qualitative analysis using generative AI tools.
4/21/2026
Enhancing Trust in Large Language Models via Uncertainty-Calibrated Fine-Tuning
Fine-tuning method to reduce LLM hallucinations through uncertainty calibration for improved trustworthiness.
4/21/2026
XOXO: Stealthy Cross-Origin Context Poisoning Attacks against AI Coding Assistants
Security research on context poisoning attacks against AI coding assistants, demonstrating vulnerabilities in automatic context gathering.
4/21/2026
RA-RRG: Multimodal Retrieval-Augmented Radiology Report Generation with Key Phrase Extraction
Multimodal retrieval-augmented system for chest X-ray report generation using key phrase extraction to reduce hallucinations.
4/21/2026
Sense and Sensitivity: Examining the Influence of Semantic Recall on Long Context Code Reasoning
Study evaluating 10 LLMs on code reasoning, distinguishing semantic understanding from lexical pattern matching in long context code understanding.
4/21/2026
Who Gets the Kidney? Human-AI Alignment, Indecision, and Moral Values
Study evaluating LLM alignment with human moral values in high-stakes decisions using kidney allocation scenarios.
4/21/2026
ReGA: Model-Based Safeguard for LLMs via Representation-Guided Abstraction
ReGA: representation-guided abstraction technique to safeguard LLMs against jailbreak attacks and harmful content generation.
4/21/2026
R3D2: Realistic 3D Asset Insertion via Diffusion for Autonomous Driving Simulation
R3D2 uses diffusion models to insert photorealistic 3D assets into autonomous driving simulations from real-world data.
4/21/2026
StableMTL: Repurposing Latent Diffusion Models for Multi-Task Learning from Partially Annotated Synthetic Datasets
StableMTL repurposes latent diffusion models for multi-task learning from partially annotated synthetic datasets in zero-shot setting.
4/21/2026
LLM Hypnosis: Exploiting User Feedback for Unauthorized Knowledge Injection to All Users
Security vulnerability where single users can persistently poison LLM knowledge through feedback upvoting/downvoting without external access.
4/21/2026
Lizard: An Efficient Linearization Framework for Large Language Models
Lizard linearization framework transforms Transformers into subquadratic architectures for efficient long-sequence LLM inference.
4/21/2026
MetaLint: Easy-to-Hard Generalization for Code Linting
MetaLint: meta-learning framework for code linting that generalizes to unseen best practices using instruction-following LLMs.
4/21/2026
Asymptotic behavior of eigenvalues of large rank perturbations of large random matrices
Random Matrix Theory analysis of deformed weight matrices in DNNs, relating to neural network pruning techniques.
4/21/2026
Culinary Crossroads: A RAG Framework for Enhancing Diversity in Cross-Cultural Recipe Adaptation
RAG framework with LLMs for diverse cross-cultural recipe adaptation considering dietary needs and cultural appropriateness.
4/21/2026
VocabTailor: Dynamic Vocabulary Selection for Downstream Tasks in Small Language Models
VocabTailor dynamically selects vocabulary for small language models to reduce memory footprint on edge devices.
4/21/2026
LEAF: Knowledge Distillation of Text Embedding Models with Teacher-Aligned Representations
Knowledge distillation framework for text embeddings enabling lightweight student models aligned to larger teacher representations.
4/21/2026
BEFT: Bias-Efficient Fine-Tuning of Language Models in Low-Data Regimes
Efficient fine-tuning method focusing on bias terms of LLMs for parameter efficiency in low-data scenarios.
4/21/2026
LayerNorm Induces Recency Bias in Transformer Decoders
Investigation of how LayerNorm causes recency bias toward later tokens in Transformer decoders via causal self-attention interaction.
4/21/2026
On the Shelf Life of Fine-Tuned LLM-Judges: Future-Proofing, Backward-Compatibility, and Question Generalization
Analysis of fine-tuned LLM judges' temporal robustness, backward-compatibility, and generalization across question types.
4/21/2026
On the Theory of Continual Learning with Gradient Descent for Neural Networks
Theoretical analysis of continual learning limitations using one-hidden-layer quadratic neural networks on sequential tasks.
4/21/2026
Parallel Test-Time Scaling for Latent Reasoning Models
Parallel test-time scaling for latent reasoning LLMs, comparing efficiency of continuous vector space reasoning versus explicit chain-of-thought.
4/21/2026
Neptune: Advanced ML Operator Fusion for Locality and Parallelism on GPUs
Tensor compiler enabling advanced operator fusion for reduction operators including attention mechanisms on GPUs.
4/21/2026
Constructive Distortion: Improving MLLMs with Attention-Guided Image Warping
AttWarp method using attention-guided image warping to improve MLLMs' fine-grained spatial reasoning and small detail detection.
4/21/2026
CLASP: Training-Free LLM-Assisted Source Code Watermarking via Semantic-Preserving Transformations
Training-free LLM-assisted watermarking for source code using semantic-preserving transformations to prevent unauthorized reuse.
4/21/2026
MTSQL-R1: Towards Long-Horizon Multi-Turn Text-to-SQL via Agentic Training
Agentic training approach for multi-turn text-to-SQL with execution, verification, and refinement for coherent dialogue grounding.
4/21/2026
What Makes AI Research Replicable? Executable Knowledge Graphs as Scientific Knowledge Representations
Executable knowledge graphs as representations to improve LLM agent replicability of AI research through better code generation and background knowledge.
4/21/2026
Demonstrating Real Advantage of Machine-Learning-Enhanced Monte Carlo for Combinatorial Optimization
ML-assisted approach for solving quadratic unconstrained binary optimization problems, demonstrating practical advantages over classical methods.
4/21/2026
ASTRA: An Automated Framework for Strategy Discovery, Retrieval, and Evolution for Jailbreaking LLMs
Automated framework for discovering and evolving jailbreak attack strategies against LLMs with continuous learning capabilities.
4/21/2026
Revisiting Entropy in Reinforcement Learning for Large Reasoning Models
Study of entropy collapse in reinforcement learning with verifiable rewards for improving LLM reasoning capabilities and avoiding premature convergence.
4/21/2026
LoRA on the Go: Instance-level Dynamic LoRA Selection and Merging
LoRA on the Go: Instance-level dynamic LoRA selection and merging for parameter-efficient fine-tuning across diverse domains without labeled data.
4/21/2026
REFLEX: Reference-Free Evaluation of Log Summarization via Large Language Model Judgment
REFLEX: Reference-free evaluation metric for log summarization using LLM judgment to assess quality across relevance and informativeness dimensions.
4/21/2026
SHRUG-FM: Reliability-Aware Foundation Models for Earth Observation
SHRUG-FM: Reliability-aware geospatial foundation models detecting out-of-distribution failures in Earth observation.
4/21/2026
VIDEOP2R: Video Understanding from Perception to Reasoning
VideoP2R: Process-aware reinforcement fine-tuning framework separating perception and reasoning for video language models.
4/21/2026
FireScope: Wildfire Risk Prediction with a Chain-of-Thought Oracle
FireScope: Large-scale dataset and chain-of-thought oracle for multimodal wildfire risk prediction from satellite imagery.
4/21/2026
The Impact of Off-Policy Training Data on Probe Generalisation
Systematic evaluation of off-policy training data impact on LLM behavior monitoring probes across eight behaviors.
4/21/2026
Beyond URLs: Metadata Diversity and Position for Efficient LLM Pretraining
Study investigating metadata types beyond URLs for efficient LLM pretraining, identifying document quality signals.
4/21/2026
Matrix: Peer-to-Peer Multi-Agent Synthetic Data Generation Framework
Matrix: Peer-to-peer multi-agent framework for coordinated synthetic data generation without centralized orchestrator.
4/21/2026
Rethinking Jailbreak Detection of Large Vision Language Models with Representational Contrastive Scoring
Representational contrastive scoring method for efficient, generalizable jailbreak detection in Vision-Language Models.
4/21/2026
Reliability-Aware Adaptive Self-Consistency for Efficient Sampling in LLM Reasoning
ReASC: Reliability-aware adaptive self-consistency method reducing inference cost in LLM reasoning by reweighting response samples.