3/9/2026
Conditioning LLMs to Generate Code-Switched Text
Framework for conditioning LLMs to generate code-switched text focusing on English-Spanish language mixing capabilities.
3/9/2026
Generative Predictive Control: Flow Matching Policies for Dynamic and Difficult-to-Demonstrate Tasks
Generative control policies using flow matching for robot tasks, addressing expert demonstration requirements and dynamic task execution.
3/9/2026
FragFM: Hierarchical Framework for Efficient Molecule Generation via Fragment-Level Discrete Flow Matching
FragFM framework for efficient molecular generation using fragment-level discrete flow matching with hierarchical autoencoder.
3/9/2026
Aligning Compound AI Systems via System-level DPO
System-level DPO method for aligning compound AI systems with multiple interacting components via policy optimization.
3/9/2026
FindAnything: Open-Vocabulary and Object-Centric Mapping for Robot Exploration in Any Environment
FindAnything open-world mapping framework using vision-language models for real-time semantic understanding in robot exploration.
3/9/2026
From Tokenizer Bias to Backbone Capability: A Controlled Study of LLMs for Time Series Forecasting
Controlled study analyzing tokenizer bias and LLM backbone capabilities for time series forecasting tasks.
3/9/2026
Federated Learning: A Survey on Privacy-Preserving Collaborative Intelligence
Survey of federated learning as distributed ML paradigm enabling collaborative training across clients while preserving data privacy.
3/9/2026
HCT-QA: A Benchmark for Question Answering on Human-Centric Tables
HCT-QA benchmark for question answering on human-centric tables in PDFs and web pages with complex structural layouts.
3/9/2026
RM-R1: Reward Modeling as Reasoning
RM-R1 integrates chain-of-thought reasoning into reward models for improved LLM alignment with human preferences via RLHF.
3/9/2026
Software Development Life Cycle Perspective: A Survey of Benchmarks for Code Large Language Models and Agents
Survey of benchmarks for code LLMs and agents across software development lifecycle stages with tiered taxonomy framework.
3/9/2026
Maximizing Asynchronicity in Event-based Neural Networks
Event-based neural networks optimized for asynchronous event camera data using improved asynchronous-to-synchronous encoding methods.
3/9/2026
AdAEM: An Adaptively and Automated Extensible Measurement of LLMs' Value Difference
AdAEM proposes automated measurement of LLMs' value differences and biases through adaptive evaluation methodology.
3/9/2026
KramaBench: A Benchmark for AI Systems on Data-to-Insight Pipelines over Data Lakes
KramaBench benchmark evaluates AI systems on end-to-end data-to-insight pipelines over real-world data lakes with unstructured data.
3/9/2026
SPARC: Concept-Aligned Sparse Autoencoders for Cross-Model and Cross-Modal Interpretability
SPARC introduces concept-aligned sparse autoencoders enabling cross-model and cross-modal interpretability for AI systems.
3/9/2026
MAP: Mitigating Hallucinations in Large Vision-Language Models with Map-Level Attention Processing
MAP mitigates hallucinations in vision-language models using map-level attention on hidden state semantics to improve factual consistency.
3/9/2026
VLMQ: Token Saliency-Driven Post-Training Quantization for Vision-language Models
VLMQ applies token saliency-driven post-training quantization to vision-language models for efficient compression without retraining.
3/9/2026
A Geometric Perspective on the Difficulties of Learning GNN-based SAT Solvers
Geometric analysis using graph Ricci curvature to explain why GNN-based SAT solvers degrade on harder constraint problems.
3/9/2026
Performance Assessment Strategies for Language Model Applications in Healthcare
Framework for assessing performance of language models in healthcare applications, accounting for variability in real clinical environments.
3/9/2026
Reasoned Safety Alignment: Ensuring Jailbreak Defense via Answer-Then-Check
Answer-Then-Check safety alignment method enhances LLM robustness against jailbreak attacks by applying reasoning before generating final responses.
3/9/2026
LikePhys: Evaluating Intuitive Physics Understanding in Video Diffusion Models via Likelihood Preference
LikePhys evaluates physical plausibility understanding in video diffusion models, separating physics correctness from visual appearance.
3/9/2026
Phys2Real: Fusing VLM Priors with Interactive Online Adaptation for Uncertainty-Aware Sim-to-Real Manipulation
Phys2Real combines vision-language models with reinforcement learning for robust sim-to-real transfer in robotic manipulation tasks.
3/9/2026
CanvasMAR: Improving Masked Autoregressive Video Prediction With Canvas
CanvasMAR improves masked autoregressive video generation by adding structured global priors to reduce distortion in frame sampling.
3/9/2026
Just-In-Time Objectives: A General Approach for Specialized AI Interactions
Framework that infers user objectives in real-time to specialize LLM outputs and create customized tools, interfaces, and responses on-the-fly.
3/9/2026
Think with 3D: Geometric Imagination Grounded Spatial Reasoning from Limited Views
Method for 3D spatial reasoning in vision-language models, improving performance on tasks requiring understanding of spatial relationships from limited views.
3/9/2026
Automated Coding of Communication Data Using ChatGPT: Consistency Across Subgroups
Evaluates ChatGPT's consistency in coding communication data across different demographic subgroups, assessing reliability for large-scale analysis tasks.
3/9/2026
Shoot First, Ask Questions Later? Building Rational Agents that Explore and Act Like People
Methods to evaluate and improve AI agents' information-seeking behavior under uncertainty, drawing from human cognition research for high-stakes applications.
3/9/2026
LA-MARRVEL: A Knowledge-Grounded, Language-Aware LLM Framework for Clinically Robust Rare Disease Gene Prioritization
LA-MARRVEL is an LLM framework for clinical rare disease diagnosis that matches genes to patient phenotypes, improving diagnostic speed by 12-15 percentage points.
3/9/2026
The Persistence of Cultural Memory: Investigating Multimodal Iconicity in Diffusion Models
Study of cultural memory and memorization in text-to-image diffusion models through multimodal iconicity phenomenon.
3/9/2026
Diffusion Fine-Tuning via Reparameterized Policy Gradient of the Soft Q-Function
SQDF: KL-regularized RL method for diffusion model fine-tuning via soft Q-function to mitigate reward over-optimization.
3/9/2026
Whatever Remains Must Be True: Filtering Drives Reasoning in LLMs, Shaping Diversity
Analysis of reverse KL divergence in RL-tuned LLMs explaining diversity loss, proposing filtering-based reasoning improvements.
3/9/2026
A-3PO: Accelerating Asynchronous LLM Training with Staleness-aware Proximal Policy Approximation
A-3PO: accelerated asynchronous LLM training with PPO using staleness-aware proximal policy approximation.
3/9/2026
Understanding and Improving Hyperbolic Deep Reinforcement Learning
Analysis and improvements for hyperbolic deep reinforcement learning agents, identifying gradient factors affecting optimization success.
3/9/2026
Agent Tools Orchestration Leaks More: Dataset, Benchmark, and Mitigation
Tools Orchestration Privacy Risk: systematic study of how multi-tool agents leak sensitive information through tool aggregation.
3/9/2026
CASA: Cross-Attention over Self-Attention for Efficient Vision-Language Fusion
CASA: efficient vision-language fusion using cross-attention over self-attention to reduce memory and compute for multi-image conversations.
3/9/2026
CARE What Fails: Contrastive Anchored-REflection for Verifiable Multimodal
CARE: failure-centric reinforcement learning framework for multimodal reasoning with contrastive learning from incorrect rollouts.
3/9/2026
LLMTM: Benchmarking and Optimizing LLMs for Temporal Motif Analysis in Dynamic Graphs
LLMTM: benchmark for evaluating LLMs on temporal motif analysis in dynamic graphs with optimization techniques.
3/9/2026
Window-based Membership Inference Attacks Against Fine-tuned Large Language Models
WBC: membership inference attack against fine-tuned LLMs using window-based localized memorization signals instead of global loss.
3/9/2026
Classroom AI: Large Language Models as Grade-Specific Teachers
Framework for fine-tuning LLMs to generate grade-appropriate educational content across elementary to adult education levels.
3/9/2026
Accelerating Scientific Research with Gemini: Case Studies and Common Techniques
Case studies demonstrating how Google's Gemini LLMs assist researchers with mathematical discovery and routine scientific tasks.
3/9/2026
Towards Autonomous Mathematics Research
Aletheia: an autonomous AI agent for mathematics research that generates, verifies, and revises proofs at professional research level.
3/9/2026
Why Human Guidance Matters in Collaborative Vibe Coding
Research on AI-assisted code generation via natural language instructions, examining human collaboration and productivity impacts.
3/9/2026
DataChef: Cooking Up Optimal Data Recipes for LLM Adaptation via Reinforcement Learning
DataChef uses reinforcement learning to automatically optimize data recipe design for LLM adaptation, treating data curation pipeline design as learnable process.
3/9/2026
SWE-MiniSandbox: Container-Free Reinforcement Learning for Building Software Engineering Agents
SWE-MiniSandbox enables scalable RL training of software engineering agents without containers, reducing storage and setup overhead.
3/9/2026
Peak + Accumulation: A Proxy-Level Scoring Formula for Multi-Turn LLM Attack Detection
Peak+Accumulation formula detects multi-turn prompt injection attacks by aggregating per-turn risk scores at proxy layer without invoking LLM.
3/9/2026
IntelliAsk: Learning to Ask High-Quality Research Questions via RLVR
IntelliAsk uses RLVR to train LLMs to generate high-quality research review questions by learning from expert annotations across effort, evidence, and grounding dimensions.
3/9/2026
FLoRG: Federated Fine-tuning with Low-rank Gram Matrices and Procrustes Alignment
FLoRG enables federated fine-tuning of LLMs using low-rank adaptation with Procrustes alignment for distributed collaborative training without data sharing.
3/9/2026
The Cascade Equivalence Hypothesis: When Do Speech LLMs Behave Like ASR$\rightarrow$LLM Pipelines?
Research evaluates when speech LLMs behave identically to ASR→LLM cascades using matched-backbone testing and mechanistic analysis.
3/9/2026
Exploratory Memory-Augmented LLM Agent via Hybrid On- and Off-Policy Optimization
EMPO² framework combines on/off-policy RL optimization with memory-augmentation to improve exploration in LLM agents, addressing key bottleneck in novel state discovery.
3/9/2026
Modality Collapse as Mismatched Decoding: Information-Theoretic Limits of Multimodal LLMs
Information-theoretic analysis of multimodal LLM failures, framing inference as mismatched decoding that ignores non-text-aligned information.
3/9/2026
CoME: Empowering Channel-of-Mobile-Experts with Informative Hybrid-Capabilities Reasoning
CoME: Mobile agent architecture with four experts for decoupled yet integrated hybrid-capabilities reasoning including planning and action decision.