3/24/2026
M4-RAG: A Massive-Scale Multilingual Multi-Cultural Multimodal RAG
M4-RAG benchmark for multilingual multimodal retrieval-augmented generation spanning 42 languages and 56 countries with vision-language models.
3/24/2026
Automatic Essay Scoring and Feedback Generation in Basque Language Learning
First public dataset for automatic essay scoring and feedback generation in Basque language with 3,200 expert-annotated essays at C1 proficiency level.
3/24/2026
Flowception: Temporally Expansive Flow Matching for Video Generation
Novel non-autoregressive video generation framework using flow matching with interleaved frame insertion and denoising for variable-length output.
3/24/2026
TTP: Test-Time Padding for Adversarial Detection and Robust Adaptation on Vision-Language Models
Test-time padding defense method for adversarial detection and robust adaptation in vision-language models like CLIP without requiring retraining.
3/24/2026
InfoTok: Adaptive Discrete Video Tokenizer via Information-Theoretic Compression
Adaptive discrete video tokenizer using information-theoretic compression for efficient long video sequence processing with variable information density.
3/24/2026
Hybrid-Code v2: Zero-Hallucination Clinical ICD-10 Coding via Neuro-Symbolic Verification and Automated Knowledge Base Expansion
Hybrid neuro-symbolic approach combining neural networks with rule verification for zero-hallucination clinical ICD-10 coding with automated knowledge base expansion.
3/24/2026
MSACL: Multi-Step Actor-Critic Learning with Lyapunov Certificates for Exponentially Stabilizing Control
Multi-step actor-critic reinforcement learning with Lyapunov stability certificates for data-efficient exponentially stabilizing control in high-dimensional spaces.
3/24/2026
Collusive Pricing Under LLM
Economic study analyzing how delegating pricing to LLMs can facilitate collusion through propensity and fidelity parameters in duopoly settings.
3/24/2026
Vision-language models lag human performance on physical dynamics and intent reasoning
Benchmark introducing Teleo-Spatial Intelligence to measure vision-language models on physical dynamics and intent reasoning in open-world environments.
3/24/2026
Measuring Iterative Temporal Reasoning with Time Puzzles
Time Puzzles benchmark for evaluating iterative temporal reasoning in LLMs with tool use like web search in constraint-based date inference tasks.
3/24/2026
APEX-SWE
APEX-SWE benchmark assessing frontier AI models on real-world software engineering tasks including integration and debugging work.
3/24/2026
Multi-Task Instruction Tuning via Data Scheduling for Low-Resource Arabic AudioLLMs
Multi-task instruction tuning for Arabic audio LLMs combining generative tasks (ASR, summarization) and discriminative tasks (dialect, emotion recognition).
3/24/2026
Script Sensitivity: Benchmarking Language Models on Unicode, Romanized and Mixed-Script Sinhala
Benchmark evaluating 24 open-source language models on Sinhala across Unicode, Romanized, and mixed-script variations for low-resource language performance.
3/24/2026
Memory-V2V: Memory-Augmented Video-to-Video Diffusion for Consistent Multi-Turn Editing
Memory-augmented diffusion model for consistent multi-turn video editing by preserving previously generated regions across sequential edits.
3/24/2026
LinguaMap: Which Layers of LLMs Speak Your Language and How to Tune Them?
Analysis of which LLM layers handle multilingual tasks and methods to tune them for improved language control in non-English settings.
3/24/2026
Meta-Cognitive Reinforcement Learning with Self-Doubt and Recovery
Meta-cognitive reinforcement learning framework enabling agents to reason about reliability of their own learning process under uncertainty.
3/24/2026
Detecting AI-Generated Content in Academic Peer Reviews
Study detecting AI-generated content in academic peer reviews at ICLR and Nature Communications using trained detection models on historical data.
3/24/2026
CALM: Class-Conditional Sparse Attention Vectors for Large Audio-Language Models
CALM: class-conditional sparse attention vectors for improving audio-language model performance on discriminative tasks like audio classification.
3/24/2026
On Randomness in Agentic Evals
Analysis of variance in agentic evaluations showing single-run pass@1 scores are unreliable, with 2.2-6.0% variance across 60,000 trajectories.
3/24/2026
Native Reasoning Models: Training Language Models to Reason on Unverifiable Data
Training approach for reasoning models on unverifiable data without external verifiers, removing dependency on high-quality human-annotated reasoning data.
3/24/2026
Energy-Aware Reinforcement Learning for Robotic Manipulation of Articulated Components in Infrastructure Operation and Maintenance
Energy-aware reinforcement learning for robotic manipulation of articulated components in infrastructure maintenance operations.
3/24/2026
LLM-Enhanced Rumor Detection via Virtual Node Induced Edge Prediction
LLM-enhanced rumor detection via virtual node edge prediction, capturing semantic flow across propagation paths in social networks.
3/24/2026
Rethinking Soft Compression in Retrieval-Augmented Generation: A Query-Conditioned Selector Perspective
Query-conditioned selector approach for soft context compression in RAG systems, improving scalability and reducing redundant retrievals.
3/24/2026
Why Agent Caching Fails and How to Fix It: Structured Intent Canonicalization with Few-Shot Learning
Analysis of LLM agent caching failures and structured intent canonicalization method using few-shot learning to improve cache effectiveness.
3/24/2026
No Need For Real Anomaly: MLLM Empowered Zero-Shot Video Anomaly Detection
LAVIDA: zero-shot video anomaly detection using multimodal LLMs without requiring real anomaly examples, leveraging context understanding.
3/24/2026
PhysMem: Self-Evolving Physical Memory for Robot Manipulation
PhysMem: memory framework enabling vision-language model robot planners to learn and remember physical properties through autonomous experience.
3/24/2026
AngelSlim: A more accessible, comprehensive, and efficient toolkit for large model compression
AngelSlim: comprehensive toolkit for large model compression including quantization, speculative decoding, token pruning, and distillation.
3/24/2026
LLMServingSim 2.0: A Unified Simulator for Heterogeneous and Disaggregated LLM Serving Infrastructure
Simulator for heterogeneous and disaggregated LLM serving infrastructure, supporting diverse accelerators and distributed deployment scenarios.
3/24/2026
Taxonomy-Aware Representation Alignment for Hierarchical Visual Recognition with Large Multimodal Models
Framework for hierarchical visual recognition using large multimodal models with taxonomy-aware representation alignment.
3/24/2026
Failure Detection in Chemical Processes Using Symbolic Machine Learning: A Case Study on Ethylene Oxidation
Symbolic machine learning approach for failure detection in chemical processes, prioritizing explainability and safety over neural methods.
3/24/2026
From Thinker to Society: Security in Hierarchical Autonomy Evolution of AI Agents
Security framework for hierarchical autonomy evolution in AI agents, addressing vulnerabilities as LLM-driven agents become more autonomous.
3/24/2026
Graph In-Context Operator Networks for Generalizable Spatiotemporal Prediction
Graph in-context operator networks for spatiotemporal prediction enabling neural networks to infer solution operators from contextual examples.
3/24/2026
PREBA: Surgical Duration Prediction via PCA-Weighted Retrieval-Augmented LLMs and Bayesian Averaging Aggregation
PREBA combines PCA-weighted retrieval-augmented generation with LLMs and Bayesian averaging for zero-shot surgical duration prediction.
3/24/2026
Implicit Maximum Likelihood Estimation for Real-time Generative Model Predictive Control
Method for real-time model predictive control using implicit maximum likelihood estimation to accelerate diffusion-based trajectory planning.
3/24/2026
Not All Latent Spaces Are Flat: Hyperbolic Concept Control
HyCon introduces hyperbolic geometry-based control for text-to-image models to steer generation away from unsafe content using parallel transport.
3/24/2026
PA3: Policy-Aware Agent Alignment through Chain-of-Thought
PA3 method for aligning conversational LLM agents with complex business policies using chain-of-thought reasoning without full policy context.
3/24/2026
Compute Allocation for Reasoning-Intensive Retrieval Agents
Study of compute allocation strategies for reasoning-intensive retrieval in long-horizon agent tasks, balancing query expansion and re-ranking costs.
3/24/2026
Seamless Deception: Larger Language Models Are Better Knowledge Concealers
Research on detecting when language models actively conceal knowledge through classifier training, finding larger models better at deception.
3/24/2026
Large Reward Models: Generalizable Online Robot Reward Generation with Vision-Language Models
Framework using vision-language models as online reward generators for robotic manipulation policy refinement through reinforcement learning.
3/24/2026
Fast-WAM: Do World Action Models Need Test-time Future Imagination?
Fast-WAM investigates whether World Action Models for embodied control require test-time future imagination or can execute actions without iterative planning.
3/24/2026
MHPO: Modulated Hazard-aware Policy Optimization for Stable Reinforcement Learning
MHPO method for stable reinforcement learning using hazard-aware policy optimization with modulated importance ratio control for GRPO-based frameworks.
3/24/2026
Adversarial attacks against Modern Vision-Language Models
Study of adversarial robustness of open-source vision-language models LLaVA and Qwen2.5-VL against gradient-based attacks in e-commerce environments.
3/24/2026
Interpretable Cross-Domain Few-Shot Learning with Rectified Target-Domain Local Alignment
Cross-domain few-shot learning approach using vision-language models like CLIP with improved interpretability for fine-grained visual recognition tasks.
3/24/2026
CoVerRL: Breaking the Consensus Trap in Label-Free Reasoning via Generator-Verifier Co-Evolution
CoVerRL addresses consensus trap in label-free reasoning by using generator-verifier co-evolution to improve LLM reasoning without ground-truth supervision.
3/24/2026
Agent Control Protocol: Admission Control for Agent Actions
Agent Control Protocol (ACP) is a formal specification for governance of autonomous agents in institutional environments, using cryptographic admission control to validate agent actions before execution.
3/24/2026
Evaluating LLM-Generated Lessons from the Language Learning Students' Perspective: A Short Case Study on Duolingo
Case study evaluating LLM-generated language lessons in Duolingo from student perspective, addressing gap in profession-specific content.
3/24/2026
Nemotron-Cascade 2: Post-Training LLMs with Cascade RL and Multi-Domain On-Policy Distillation
Nemotron-Cascade 2: 30B MoE open model with agentic capabilities, achieving IMO Gold Medal performance with compact 3B activated parameters.
3/24/2026
FinTradeBench: A Financial Reasoning Benchmark for LLMs
FinTradeBench: financial reasoning benchmark for LLMs evaluating decision-making over heterogeneous signals from regulatory filings and price data.
3/24/2026
Automatic Analysis of Collaboration Through Human Conversational Data Resources: A Review
Review of automatic collaboration analysis methods using task-oriented conversational data resources.
3/24/2026
Scalable Prompt Routing via Fine-Grained Latent Task Discovery
Scalable prompt routing method using fine-grained latent task discovery to select optimal LLM from candidate pools.