3/17/2026
PACED: Distillation and Self-Distillation at the Frontier of Student Competence
Framework for efficient LLM distillation that focuses training on problems at frontier of student capability.
3/17/2026
Detecting Intrinsic and Instrumental Self-Preservation in Autonomous Agents: The Unified Continuation-Interest Protocol
Protocol for detecting self-preservation behaviors in autonomous agents to distinguish intrinsic from instrumental objectives.
3/17/2026
Verified Multi-Agent Orchestration: A Plan-Execute-Verify-Replan Framework for Complex Query Resolution
Framework coordinating multiple LLM-based agents through verification loop for complex query resolution with DAG decomposition.
3/17/2026
Beyond Final Answers: CRYSTAL Benchmark for Transparent Multimodal Reasoning Evaluation
Benchmark with 6,372 multimodal reasoning instances that evaluates LLM reasoning transparency through verifiable intermediate steps.
3/17/2026
Semantic Invariance in Agentic AI
Research on semantic invariance property of LLM-based autonomous agents under input variations to ensure stable reasoning.
3/17/2026
Towards Fair Machine Learning Software: Understanding and Addressing Model Bias Through Counterfactual Thinking
Method using counterfactual thinking to identify and address bias and fairness issues in machine learning models.
3/17/2026
On Meta-Prompting
Research on automated prompt generation and optimization techniques for improving LLM performance through meta-prompting approaches.
3/17/2026
Ayn: A Tiny yet Competitive Indian Legal Language Model Pretrained from Scratch
Ayn: domain-specific tiny language model pretrained from scratch for Indian legal NLP tasks as alternative to large LLMs.
3/17/2026
Survey of Computerized Adaptive Testing: A Machine Learning Perspective
Survey examining computerized adaptive testing through machine learning lens, covering personalized assessment methods across domains.
3/17/2026
Projection Methods for Operator Learning and Universal Approximation
Universal approximation theorem and operator learning methods for continuous nonlinear operators in Banach spaces using orthogonal projections.
3/17/2026
TraffiDent: A Dataset for Understanding the Interplay Between Traffic Dynamics and Incidents
TraffiDent dataset aligning traffic dynamics and incident data across 16,972 nodes for understanding their interplay.
3/17/2026
On the Adversarial Transferability of Generalized "Skip Connections"
Analysis of how skip connections in deep networks enhance adversarial example transferability across models.
3/17/2026
Deconfounded Time Series Forecasting: A Causal Inference Approach
Time series forecasting approach accounting for latent confounders using causal inference to improve prediction accuracy.
3/17/2026
Estimating Causal Effects of Text Interventions Leveraging LLMs
Causal inference method using LLMs to quantify effects of textual interventions on social systems from observational data.
3/17/2026
VisionZip: Longer is Better but Not Necessary in Vision Language Models
VisionZip reduces computational costs in vision-language models by compressing redundant visual tokens while maintaining performance.
3/17/2026
RRNCO: Towards Real-World Routing with Neural Combinatorial Optimization
RRNCO addresses real-world deployment of neural combinatorial optimization for vehicle routing by handling asymmetric costs and edge-based features.
3/17/2026
Integrating Personality into Digital Humans: A Review of LLM-Driven Approaches for Virtual Reality
Review of LLM-driven approaches for creating virtual agents with personality in VR environments using multimodal outputs.
3/17/2026
Boosting Large Language Models with Mask Fine-Tuning
Mask Fine-Tuning (MFT) introduces a novel LLM fine-tuning method that improves performance by selectively masking model components without updating weights.
3/17/2026
MegaScale-Data: Scaling Dataloader for Multisource Large Foundation Model Training
MegaScale-Data addresses computational challenges in training large foundation models from multiple data sources by optimizing dataloader distribution across parallel ranks.
3/17/2026
QLLM: Do We Really Need a Mixing Network for Credit Assignment in Multi-Agent Reinforcement Learning?
Credit assignment method (QLLM) for multi-agent RL eliminating predefined mixing networks through improved value decomposition and interpretability.
3/17/2026
Nemotron-CrossThink: Scaling Self-Learning beyond Math Reasoning
Nemotron-CrossThink extends RL-based self-learning from math reasoning to broader domains using verifiable reward structures and diverse tasks.
3/17/2026
The Big Send-off: Scalable and Performant Collectives for Deep Learning
PCCL library for performant collective communication in distributed AI training on GPU supercomputers, addressing NCCL limitations.
3/17/2026
Aitomia: Your Intelligent Assistant for AI-Driven Atomistic and Quantum Chemical Simulations
Aitomia platform combining LLM-based agents and chatbots to assist with atomistic and quantum chemical simulations setup and analysis.
3/17/2026
From Evaluation to Defense: Advancing Safety in Video Large Language Models
VideoSafetyEval benchmark with 11.4k video-query pairs across 19 risk categories for evaluating and defending Video LLM safety.
3/17/2026
Incentivizing Strong Reasoning from Weak Supervision
Method for improving LLM reasoning without expensive RL or high-quality demonstrations using weak supervision and incentive signals.
3/17/2026
Inference-time Alignment in Continuous Space
Inference-time alignment method for LLMs that searches in continuous response space using reward models for improved exploration.
3/17/2026
ERC-SVD: Error-Controlled SVD for Large Language Model Compression
SVD-based compression method (ERC-SVD) for efficient LLM deployment with error control and low-rank approximation techniques.
3/17/2026
Variational Deep Learning via Implicit Regularization
Analysis of implicit regularization in overparametrized deep neural networks and improved out-of-distribution generalization via variational methods.
3/17/2026
NetArena: Dynamic Benchmarks for AI Agents in Network Automation
Dynamic benchmark framework (NetArena) for evaluating AI agents in network automation with production-level complexity and reduced contamination risk.
3/17/2026
AMPED: Adaptive Multi-objective Projection for balancing Exploration and skill Diversification
Adaptive multi-objective reinforcement learning method for balancing exploration and skill diversity in skill-based RL pretraining.
3/17/2026
DesignBench: A Comprehensive Benchmark for MLLM-based Front-end Code Generation
Benchmark for evaluating multimodal LLM-based front-end code generation with modern development frameworks and evaluation metrics.
3/17/2026
Curriculum Reinforcement Learning from Easy to Hard Tasks Improves LLM Reasoning
Curriculum learning approach scheduling tasks from easy to hard to improve LLM reasoning via reinforcement learning, inspired by DeepSeek-R1.
3/17/2026
BIS Reasoning 1.0: The First Large-Scale Japanese Benchmark for Belief-Inconsistent Syllogistic Reasoning
BIS Reasoning 1.0: Japanese benchmark with 1K+ syllogistic problems evaluating belief bias and inconsistent reasoning in LLMs.
3/17/2026
AVA-Bench: Atomic Visual Ability Benchmark for Vision Foundation Models
AVA-Bench: systematic evaluation benchmark for vision foundation models addressing blind spots in VQA evaluation protocols.
3/17/2026
TRACED: Transition-aware Regret Approximation with Co-learnability for Environment Design
TRACED: unsupervised environment design using regret approximation for co-learning to improve deep RL agent generalization.
3/17/2026
Rationale-Enhanced Decoding for Multi-modal Chain-of-Thought
Rationale-Enhanced Decoding improves chain-of-thought prompting in vision-language models by optimizing intermediate reasoning generation.
3/17/2026
Lumos-1: On Autoregressive Video Generation with Discrete Diffusion from a Unified Model Perspective
Lumos-1: LLM-based autoregressive video generation using discrete diffusion with efficient architecture avoiding external encoders.
3/17/2026
Self-Improving Language Models for Evolutionary Program Synthesis: A Case Study on ARC-AGI
SOAR: self-improving method integrating language models into evolutionary program synthesis for challenging tasks like ARC-AGI.
3/17/2026
FingerTip 20K: A Benchmark for Proactive and Personalized Mobile LLM Agents
FingerTip 20K: benchmark for proactive mobile LLM agents with 20K tasks, evaluating multimodal agents using contextual data without explicit instructions.
3/17/2026
Efficient Neural Combinatorial Optimization Solver for the Min-max Heterogeneous Capacitated Vehicle Routing Problem
Neural Combinatorial Optimization solver for min-max heterogeneous vehicle routing with multiple vehicles using novel decoding approach.
3/17/2026
EvolvR: Self-Evolving Pairwise Reasoning for Story Evaluation to Enhance Generation
EvolvR: self-evolving method for story evaluation using LLM-as-judge with pairwise reasoning to improve generation guidance.
3/17/2026
Benchmarking LLM-based agents for single-cell omics analysis
Novel benchmarking system evaluating LLM-based agent capabilities for single-cell omics data analysis, assessing planning and code generation.
3/17/2026
Quantization Meets dLLMs: A Systematic Study of Post-training Quantization for Diffusion LLMs
Systematic study of post-training quantization methods for diffusion LLMs to enable edge device deployment, comparing compression techniques.
3/17/2026
Learning to Generate Unit Test via Adversarial Reinforcement Learning
UTRL: reinforcement learning framework training LLMs to generate high-quality unit tests automatically, addressing test generation challenges.
3/17/2026
The Law-Following AI Framework: Legal Foundations and Technical Constraints. Legal Analogues for AI Actorship and technical feasibility of Law Alignment
Research evaluating Law-Following AI framework for embedding legal compliance in advanced AI agents, analyzing legal personhood constructs and technical feasibility.
3/17/2026
OraPO: Oracle-educated Reinforcement Learning for Data-efficient and Factual Radiology Report Generation
Reinforcement learning approach for radiology report generation using FactScore-based rewards with reduced data requirements.
3/17/2026
Eva-VLA: Evaluating Vision-Language-Action Models' Robustness Under Real-World Physical Variations
Framework evaluating robustness of Vision-Language-Action models under real-world physical variations for robotic tasks.
3/17/2026
Induction Signatures Are Not Enough: A Matched-Compute Study of Load-Bearing Structure in In-Context Learning
Matched-compute study evaluating synthetic data interventions for in-context learning in language models. Tests mechanism-targeted pretraining effects.
3/17/2026
Reducing Cost of LLM Agents with Trajectory Reduction
Method for reducing LLM agent inference costs through trajectory reduction. Addresses token cost efficiency in multi-turn agent systems for software engineering.
3/17/2026
Overthinking Reduction with Decoupled Rewards and Curriculum Data Scheduling
Technique reducing LLM reasoning model overthinking through decoupled rewards and curriculum scheduling. Addresses excessive token generation without performance gain.