Chang Xu

Associate Professor@University of Sydney·澳大利亚·OpenReview

研究方向

Deep Neural Network Design and Optimisation · Adversarial Machine Learning and Applications · Multi-View/Label/Task Learning · Deep Generative Models

Chang Xu

VEAttack: Downstream-agnostic Vision Encoder Attack against Large Vision Language Models

Catching the Details: Self-Distilled RoI Predictors for Fine-Grained MLLM Perception

Sample Margin-Aware Recalibration of Temperature Scaling

Consistency-Driven Calibration and Matching for Few-Shot Class Incremental Learning

Ban&Pick: Ehancing Performance and Efficiency of MoE-LLMs via Smarter Routing

Rethinking Causal Mask Attention for Vision-Language Inference

Action-aware Dynamic Pruning for Efficient Vision-Language-Action Manipulation

WATS: Wavelet-Aware Temperature Scaling for Reliable Graph Neural Networks

From Language to Locomotion: Retargeting-free Humanoid Control via Motion Latent Guidance

Towards Robust Graph Unlearning via Gradient Consistency Control

Can Large Language Models Express Uncertainty Like Human?

Efficient Conditional Generation on Scale-based Visual Autoregressive Models

Fixing Model-Fitting: Compressing Guidance for Better Sampling

Enhancing Physical Plausibility in Video Generation by Reasoning the Implausibility

Diff-SSR: Diffusion Model with Structure-Modulated for Image Super-Resolution

Not All Tokens are Guided Equal: Elucidating Guidance in Autoregressive Modelling

Block Rotation is All You Need for MXFP4 Quantization

Circle-RoPE: Cone-like Decoupled Rotary Positional Embedding for Vision-Language Models

Building Flow Uniqueness in One-step Generative Modeling

Nash Policy Gradient: A Policy Gradient Method with Iteratively Refined Regularization for Finding Nash Equilibria

OmniMotion: Multimodal Motion Generation with Continuous Masked Autoregression

DeCoP: Enhancing Self-Supervised Time Series Representation with Dependency Controlled Pre-training

Adversarially Injected Diagnosis for Coherent Visual Autoregressive Generation

Attention Calibration for Reducing Hallucination in Large Vision-Language Models

Efficient Self-Guided Editing for Text-Driven Image-to-Image Translation

From Language to Action Streams: Bridging LLM Autoregression for Long-Horizon Robot Action Prediction

SRSR: Enhancing Semantic Accuracy in Real-World Image Super-Resolution with Spatially Re-Focused Text-Conditioning

Diffusion Attribution Score: Evaluating Training Data Influence in Diffusion Models

Representative Guidance: Diffusion Model Sampling with Coherence

Exploring Local Memorization in Diffusion Models via Bright Ending Attention

Anti-Exposure Bias in Diffusion Models

Adversarial Robustness via Deformable Convolution with Stochasticity

Stable Fair Graph Representation Learning with Lipschitz Constraint

Understanding and Mitigating Hallucination in Large Vision-Language Models via Modular Attribution and Intervention

Learning Mask Invariant Mutual Information for Masked Image Modeling

VLA-Cache: Efficient Vision-Language-Action Manipulation via Adaptive Token Caching

Beyond One-Hot Labels: Semantic Mixing for Model Calibration

SAN-Diff: Structure-aware noise for super-resolution diffusion model

Empirical Study on Enhancing Efficiency in Masked Image Modeling Pre-training

Training-free Long Video Generation with Chain of Diffusion Model Experts

The Devil is in the Word: Video-Conditioned Text Representation Refinement for Text-to-Video Retrieval

Weak-to-Strong Enhanced Vision Model

Compress Guidance in Conditional Diffusion Sampling

RED: Efficiently Boosting Ensemble Robustness via Random Sampling Inference

Investigating Memorization in Video Diffusion Models

Exploring the Design Space of Autoregressive Models for Efficient and Scalable Image Generation

FusionSAM: Visual Multimodal Learning with Segment Anything Model

Consistency Calibration: Improving Uncertainty Calibration via Consistency among Perturbed Neighbors