Philip Torr

Learning to See Before Seeing: Demystifying LLM Visual Priors from Language Pre-training

A Guardrail for Safety Preservation: When Safety-Sensitive Subspace Meets Harmful-Resistant Null-Space

Beyond Linear Probes: Dynamic Safety Monitoring for Language Models

Agent-X: Evaluating Deep Multimodal Reasoning in Vision-Centric Agentic Tasks

Eigen-Agent: Adaptive Multi-Agent Scientific Reasoning with Monitor-Based RAG

Single LLM Debate, MoLaCE: Mixture of Latent Concept Experts Against Confirmation Bias

BiasBusters: Uncovering and Mitigating Tool Selection Bias in Large Language Models

Interleaving Reasoning for Better Text-to-Image Generation

TRACEDET: HALLUCINATION DETECTION FROM THE DECODING TRACE OF DIFFUSION LARGE LANGUAGE MODELS

Safe Autoregressive Image Generation with Iterative Self-Improving Codebooks

Learnable Sparsity for Vision Generative Models

CoMAS: Co-Evolving Multi-Agent Systems via Interaction Rewards

It's a TRAP! Task-Redirecting Agent Persuasion Benchmark for Web Agents

DIVA: Discrete Diffusion Vision-Language-Action Models for Parallelized Action Generation

Can Large Language Models Express Uncertainty Like Human?

Better know nothing than half-know anything: A Precise and Efficient Dataset for Scientific Reasoning in Language Models

Towards Understanding Multimodal Fine-Tuning: A Case Study into Spatial Features

PostAlign: Multimodal Grounding as a Corrective Lens for MLLMs

FORCE: Transferable Visual Jailbreaking Attacks via Feature Over-Reliance CorrEction

Measuring Sparse Autoencoder Feature Space Similarities Across Large Language Models

Learning to Fly Camera Drones by Watching Internet Videos

Purifying Task Vectors in Knowledge-Aware Subspace for Model Merging

OMNI-LEAK: Orchestrator Multi-Agent Network Induced Data Leakage

h1: Bootstrapping Models to Reason over Longer Horizons via Reinforcement Learning

GLARE: Towards Graph-less Retrieval for Retrieval Augmented Generation on Million-scale Knowledge Graphs

LikePhys: Evaluating Intuitive Physics Understanding in Video Diffusion Models via Likelihood Preference

Ego to World: Collaborative Spatial Reasoning in Embodied Systems via Reinforcement Learning

Base Models Know How to Reason, Thinking Models Learn When

Bag of Tricks for Subverting Reasoning-based Safety Guardrails

RoboMonster: Compositional Generalization of Heterogeneous Multi-End Effector Embodied Agents

Predicting Weak-to-Strong Generalization from Latent Representations

Computer-Use Agents as Judges for Automatic GUI Design

VCode: a Multimodal Coding Benchmark with SVG as Symbolic Visual Representation

On the Diminishing Reliability of Reference-Free Memorization Detection in Modern Diffusion Models

ToolTweak: An Attack on Tool Selection in LLM-based Agents

Causal Discovery under Changing Mechanisms: A Unified Graphical Approach

Towards Causal Fine-Tuning under Latent-Confounded Shift

Building Scalable Real-World Robot Data Generation via Compositional Simulation

AN ITERATIVE PROMPTING FRAMEWORK FOR LLM-BASED DATA PREPROCESSING

CrystalSeg: Automating Synchrotron Tomographic Reconstruction Segmentation for Crystallography with Physically Guided Simulations

Towards Agents That Know When They Don't Know: Uncertainty as a Control Signal for Structured Reasoning

MATRIX: Multimodal Agent Tuning for Robust Tool-Use Reasoning

Can Knowledge-Graph-based Retrieval Augmented Generation Really Retrieve What You Need?

Bi-Factorial Preference Optimization: Balancing Safety-Helpfulness in Language Models

Direct3D-S2: Gigascale 3D Generation Made Easy with Spatial Sparse Attention

On the Coexistence and Ensembling of Watermarks

Towards Interpreting Visual Information Processing in Vision-Language Models

Focus On This, Not That! Steering LLMs with Adaptive Feature Specification

MALT: Improving Reasoning with Multi-Agent LLM Training

Towards Reliable Identification of Diffusion-based Image Manipulations

Towards Certification of Uncertainty Calibration under Adversarial Attacks

Shh, don't say that! Domain Certification in LLMs

Mixture of Experts Made Intrinsically Interpretable

On the Coexistence and Ensembling of Watermarks

True Multimodal In-Context Learning Needs Attention to the Visual Context

MIP against Agent: Malicious Image Patches Hijacking Multimodal OS Agents

Too Late to Recall: Explaining the Two-Hop Problem in Multimodal Knowledge Retrieval

Rethinking Safety in LLM Fine-tuning: An Optimization Perspective

Sparse Autoencoders Reveal Universal Feature Spaces Across Large Language Models

Do as I do (Safely): Mitigating Task-Specific Fine-tuning Risks in Large Language Models

Semantic Score Distillation Sampling for Compositional Text-to-3D Generation

Focus On This, Not That! Steering LLMs With Adaptive Feature Specification

Flex3D: Feed-Forward 3D Generation with Flexible Reconstruction Model and Input View Curation

Flex3D: Feed-Forward 3D Generation with Flexible Reconstruction Model and Input View Curation

MAD-Sherlock: Multi-Agent Debates for Out-of-Context Misinformation Detection

PoisonBench: Assessing Large Language Model Vulnerability to Data Poisoning

REVIP: Rethinking Visual Prompting for Multimodal Large Language Models with External Knowledge

Minimalist Concept Erasure in Generative Models

SynthCLIP: Are We Ready for a Fully Synthetic CLIP Training?

Incrementally Adapting Generative Vision-Language Models with Task Codebook

Can Editing LLMs Inject Harm?

PoisonBench: Assessing Language Model Vulnerability to Poisoned Preference Data

LLM Jailbreak Detection for (Almost) Free!

FedBiP: Heterogeneous One-Shot Federated Learning with Personalized Latent Diffusion Models

CRAB: Cross-environment Agent Benchmark for Multimodal Language Model Agents

OASIS: Open Agents Social Interaction Simulations on a Large Scale

SAGE: Scalable Ground Truth Evaluations for Large Sparse Autoencoders

Cracking the Collective Mind: Adversarial Manipulation in Multi-Agent Systems

Learning Visual Prompts for Guiding the Attention of Vision Transformers

A Scalable Communication Protocol for Networks of Large Language Models

FAIRMINDSIM: ALIGNMENT OF BEHAVIOR, EMO- TION, AND BELIEF IN HUMANS AND LLM AGENTS AMID ETHICAL DILEMMAS

Toward Robust Real-World Audio Deepfake Detection: Closing the Explainability Gap

Questioning Simplicity Bias Assumptions

Language Models' Internal Conflicts: Layer-wise Usable Information For Detecting Model (Un)answerability