3/30/2026
ViGoR-Bench: How Far Are Visual Generative Models From Zero-Shot Visual Reasoners?
ViGoR-Bench evaluates visual generative models on reasoning tasks requiring physical, causal, and spatial understanding beyond generation fidelity.
3/30/2026
A Compression Perspective on Simplicity Bias
Theoretical analysis of simplicity bias in neural networks using Minimum Description Length principle and compression-based learning framework.
3/30/2026
GazeQwen: Lightweight Gaze-Conditioned LLM Modulation for Streaming Video Understanding
GazeQwen integrates gaze information into multimodal LLMs for video understanding via parameter-efficient hidden-state modulation.
3/30/2026
Why Safety Probes Catch Liars But Miss Fanatics
Analysis of activation-based safety probes for detecting AI misalignment, proving polynomial-time probes fail on coherent misalignment cases.
3/30/2026
Methods for Knowledge Graph Construction from Text Collections: Development and Applications
Survey of knowledge graph construction methods from unstructured text across domains including healthcare, news, and social media.
3/30/2026
GUIDE: A Benchmark for Understanding and Assisting Users in Open-Ended GUI Tasks
GUIDE benchmark for evaluating GUI agents on open-ended tasks emphasizing user collaboration and iterative refinement over automation.
3/30/2026
On Integrating Resilience and Human Oversight into LLM-Assisted Modeling Workflows for Digital Twins
Design principles for integrating resilience and human oversight into LLM-assisted digital twin modeling workflows.
3/30/2026
Good Scores, Bad Data: A Metric for Multimodal Coherence
Multimodal Coherence Score metric evaluates fusion quality in multimodal AI systems independent of downstream task accuracy.
3/30/2026
Reinforcing Structured Chain-of-Thought for Video Understanding
Method reinforces structured chain-of-thought in multimodal LLMs for video understanding using reinforcement learning without costly CoT annotation.
3/30/2026
Can Small Models Reason About Legal Documents? A Comparative Study
Comparative evaluation of sub-10B parameter models on legal reasoning benchmarks using five prompting strategies, testing practical alternatives to frontier LLMs.
3/30/2026
Collision-Aware Vision-Language Learning for End-to-End Driving with Multimodal Infraction Datasets
Vision-language learning approach for autonomous driving using multimodal infraction datasets to reduce collision rates in end-to-end models.
3/30/2026
When Chain-of-Thought Backfires: Evaluating Prompt Sensitivity in Medical Language Models
Study evaluates MedGemma LLM robustness on medical QA benchmarks, finding chain-of-thought prompting reduces accuracy by 5.7% versus direct answering.
3/30/2026
Policy-Guided World Model Planning for Language-Conditioned Visual Navigation
PiJEPA framework combining learned navigation policies with world model planning for language-conditioned visual navigation in embodied AI.
3/30/2026
FairLLaVA: Fairness-Aware Parameter-Efficient Fine-Tuning for Large Vision-Language Assistants
Parameter-efficient fine-tuning method for multimodal LLMs addressing fairness disparities across demographic groups in clinical settings.
3/30/2026
VLAgeBench: Benchmarking Large Vision-Language Models for Zero-Shot Human Age Estimation
Benchmark evaluating large vision-language models on zero-shot human age estimation from facial images without task-specific training.
3/30/2026
H-Node Attack and Defense in Large Language Models
Mechanistic framework to identify and defend against hallucinations in LLMs by analyzing hidden-state dimensions of transformer models.
3/30/2026
Selective Deficits in LLM Mental Self-Modeling in a Behavior-Based Test of Theory of Mind
Behavior-based test revealing selective deficits in LLM theory of mind capabilities, specifically mental self-modeling.
3/30/2026
Dynamic Tokenization via Reinforcement Patching: End-to-end Training and Zero-shot Transfer
Method for learning variable-sized, data-driven patches via reinforcement learning for compact representation of long-horizon sequence data.
3/30/2026
"Oops! ChatGPT is Temporarily Unavailable!": A Diary Study on Knowledge Workers' Experiences of LLM Withdrawal
Diary study with 10 knowledge workers examining dependency on LLMs and workflow disruption when LLM access is unavailable.
3/30/2026
SkinGPT-X: A Self-Evolving Collaborative Multi-Agent System for Transparent and Trustworthy Dermatological Diagnosis
Multi-agent LLM system for dermatological diagnosis offering interpretability and traceability for clinical reasoning on rare skin diseases.
3/30/2026
Finding Distributed Object-Centric Properties in Self-Supervised Transformers
Analysis of how self-supervised Vision Transformers discover objects, addressing spurious activations in [CLS] token attention for improved object localization.
3/30/2026
SWE-PRBench: Benchmarking AI Code Review Quality Against Pull Request Feedback
SWE-PRBench: Benchmark of 350 pull requests showing frontier LLMs detect only 15-31% of human-flagged code review issues, demonstrating gap between code generation and code review performance.
3/30/2026
ATime-Consistent Benchmark for Repository-Level Software Engineering Evaluation
Time-consistent benchmark methodology for evaluating repository-level software engineering systems, addressing synthetic task design and temporal contamination issues.
3/30/2026
Sparse Auto-Encoders and Holism about Large Language Models
Philosophical exploration of whether LLMs' sparse auto-encoders suggest a meta-semantic picture of how language models capture meaning.
3/30/2026
Towards GUI Agents: Vision-Language Diffusion Models for GUI Grounding
Evaluation of discrete diffusion vision-language models for GUI grounding in agentic systems, comparing against autoregressive VLMs.
3/30/2026
Clawed and Dangerous: Can We Trust Open Agentic Systems?
Study of security challenges in open agentic systems combining LLMs with external capabilities, persistent memory, and privileged execution used in coding assistants and enterprise automation.
3/30/2026
Automating Domain-Driven Design: Experience with a Prompting Framework
LLM-based prompting framework automating domain-driven design activities including ubiquitous language, event storming, and bounded contexts.
3/30/2026
Distilling Conversations: Abstract Compression of Conversational Audio Context for LLM-based ASR
Method for incorporating conversational context into LLM-based speech recognition through abstract compression of multi-turn audio.
3/30/2026
Channelling, Coordinating, Collaborating: A Three-Layer Framework for Disability-Centered Human-Agent Collaboration
Framework for disability-centered human-agent collaboration proposing channelling, coordinating, and co-creating interaction layers.
3/30/2026
ARTA: Adaptive Mixed-Resolution Token Allocation for Efficient Dense Feature Extraction
Mixed-resolution vision transformer using adaptive token allocation for efficient dense feature extraction from images.
3/30/2026
Working Notes on Late Interaction Dynamics: Analyzing Targeted Behaviors of Late Interaction Models
Analysis of late interaction retrieval models, investigating length bias and similarity distribution behaviors in multi-vector scoring.
3/30/2026
Knowdit: Agentic Smart Contract Vulnerability Detection with Auditing Knowledge Summarization
AI agent system for detecting smart contract vulnerabilities by leveraging DeFi semantics and auditing knowledge summarization.
3/30/2026
PhysVid: Physics Aware Local Conditioning for Generative Video Models
Physics-aware conditioning scheme for generative video models to enforce physical constraints during generation.
3/30/2026
Preference-Aligned LoRA Merging: Preserving Subspace Coverage and Addressing Directional Anisotropy
Technique for merging multiple LoRA modules while preserving task-critical directions and addressing subspace misalignment in multi-task adaptation.
3/30/2026
Label-Free Cross-Task LoRA Merging with Null-Space Compression
Method for merging LoRA-adapted model checkpoints without joint training, handling heterogeneous tasks like classification and regression.
3/30/2026
From Human Cognition to Neural Activations: Probing the Computational Primitives of Spatial Reasoning in LLMs
Mechanistic study of how LLMs internally represent and use spatial information, examining whether spatial reasoning relies on structured representations or linguistic heuristics.
3/30/2026
Mitigating the Reasoning Tax in Vision-Language Fine-Tuning with Input-Adaptive Depth Aggregation
Input-adaptive depth aggregation method to mitigate reasoning performance degradation in vision-language model fine-tuning.
3/30/2026
CALRK-Bench: Evaluating Context-Aware Legal Reasoning in Korean Law
CALRK-Bench: context-aware legal reasoning benchmark for Korean law evaluating complex norm interactions.
3/30/2026
Reflect to Inform: Boosting Multimodal Reasoning via Information-Gain-Driven Verification
Information-gain-driven verification approach to reduce hallucinations in multimodal LLM long-form generation.
3/30/2026
Generative Score Inference for Multimodal Data
Generative score inference method for uncertainty quantification on multimodal supervised learning tasks.
3/30/2026
Automated near-term quantum algorithm discovery for molecular ground states
AI-driven quantum algorithm discovery using LLMs and evolutionary processes for molecular ground state problems.
3/30/2026
Generative Modeling in Protein Design: Neural Representations, Conditional Generation, and Evaluation Standards
Survey of generative modeling in protein design covering neural representations, conditional generation, and evaluation standards.
3/30/2026
Why Models Know But Don't Say: Chain-of-Thought Faithfulness Divergence Between Thinking Tokens and Answers in Open-Weight Reasoning Models
Analysis of chain-of-thought faithfulness in open-weight reasoning models comparing thinking tokens vs final answers.
3/30/2026
KMM-CP: Practical Conformal Prediction under Covariate Shift via Selective Kernel Mean Matching
Conformal prediction method under covariate shift using kernel mean matching for uncertainty quantification.
3/30/2026
CPUBone: Efficient Vision Backbone Design for Devices with Low Parallelization Capabilities
CPUBone: vision backbone architecture design optimized for CPU devices with low parallelization capabilities.
3/30/2026
Can AI Models Direct Each Other? Organizational Structure as a Probe into Training Limitations
ManagerWorker: two-agent pipeline study where expensive model directs cheap model for software engineering tasks.
3/30/2026
Neuro-Symbolic Process Anomaly Detection
Neuro-symbolic approach to process anomaly detection combining neural networks with domain knowledge from event logs.
3/30/2026
A Boltzmann-machine-enhanced Transformer For DNA Sequence Classification
Boltzmann-machine-enhanced Transformer for DNA sequence classification with latent structure discovery.
3/30/2026
UNIFERENCE: A Discrete Event Simulation Framework for Developing Distributed AI Models
UNIFERENCE: discrete-event simulation framework for developing and benchmarking distributed AI inference algorithms.
3/30/2026
Rocks, Pebbles and Sand: Modality-aware Scheduling for Multimodal Large Language Model Inference
Modality-aware scheduling system for efficient multimodal LLM inference handling text, images, and video workloads.