Bernard Ghanem

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

Structured-Noise Masked Modeling for Video, Audio and Beyond

Turning the Spell Around: Lightweight Alignment Amplification via Rank-One Safety Injection

SoCo: Progressive Spectrum Optimization for Large Language Model Compression

Transformers from Compressed Representations

Motion-Aware Concept Alignment for Consistent Video Editing

MoDA: Modulation Adapter for Fine-Grained Visual Understanding in Instructional MLLMs

FloorplanQA: A Benchmark for Spatial Reasoning in LLMs using Structured Representations

FedFFT: Taming Client Drift in Federated SAM via Spectral Perturbation Filtering

Purifying Task Vectors in Knowledge-Aware Subspace for Model Merging

Train Long, Think Short: Curriculum Learning for Efficient Reasoning

Imitate Optimal Policy: Prevail and Induce Action Collapse in Policy Gradient

Beyond the Last Answer: Your Reasoning Trace Uncovers More than You Think

ADVMEM: Adversarial Memory Initialization for Realistic Test-Time Adaptation via Tracklet-Based Benchmarking

Efficient Correspondence Learning for Dense Semantic Label Propagation

EAGLE: Enhanced Visual Grounding Minimizes Hallucinations in Instructional Multimodal Models

Reasoning Vectors: Transferring Chain-of-Thought Capabilities via Task Arithmetic

CoLaP: Contrastive Learning with Adaptive Prompts for Continual Learning

GOTTA be diverse

Multimodal Safety Evaluation in Generative Agent Social Simulations

OWL: Optimized Workforce Learning for General Multi-Agent Assistance in Real-World Task Automation

RMP-SAM: Towards Real-Time Multi-Purpose Segment Anything

Mind-the-Glitch: Visual Correspondence for Detecting Inconsistencies in Subject-Driven Generation

OmniResponse: Online Multimodal Conversational Response Generation in Dyadic Interactions

Rethinking Safety in LLM Fine-tuning: An Optimization Perspective

Test-Time Adaptation for Combating Missing Modalities in Egocentric Videos

ResidualViT for Efficient Zero-Shot Natural Language Temporal Video Grounding

Behind the Magic, MERLIM: Multi-modal Evaluation Benchmark for Large Image-Language Models

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

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

OASIS: Open Agents Social Interaction Simulations on a Large Scale

Compressed-Language Models for Understanding Compressed File Formats: a JPEG Exploration