FAST-CAD: A Fairness-Aware Framework for Non-Contact Stroke Diagnosis

Project Page  | MIT Tech Review

Stroke is an acute cerebrovascular disease, so we propose FAST-CAD, a DAT + Group-DRO framework that jointly enforces demographic-invariant representations and worst-group robustness for non-contact stroke diagnosis. Built on a 12-subgroup multimodal dataset, it couples adversarial domain discrimination with self-supervised encoders and optimizes worst-group risk, delivering 91.2% AUC and tight fairness bounds backed by domain adaptation and minimax theory.

Role: Collaborating Author.

Multi-Cache enhanced Prototype Learning for Test-Time Generalization of Vision-Language Models

Project Page

We observed that cache-based test-time adaptation performance is positively correlated with intra-class compactness. To address the unreliability of low-entropy samples under distribution shifts, we propose MCP, which uses an entropy cache for prototype initialization, an align cache to fuse visual and textual information and tighten intra-class distributions, and a negative cache to calibrate high-entropy predictions. We further extend this into the MCP++ framework by introducing cross-modal prototype alignment and residual learning, achieving state-of-the-art generalization on 15 downstream tasks.

Role: Co-first Author.

Mitigating Cache Noise in Test-Time Adaptation for Large Vision-Language Models

Also accepted at ICLR 2025 FM-Wild Workshop

We analyzed the root causes of the performance gap between zero-shot and few-shot TTA, identifying noisy cache labels as a critical bottleneck. We then propose the CRG framework, which maintains positive and negative visual prototypes alongside text prototypes, employs learnable residuals to align modalities, and leverages Gaussian Discriminant Analysis to dynamically model class distributions and suppress noisy samples. Finally, by jointly minimizing prediction entropy and maximizing inter-prototype distances, CRG achieves superior robustness and generalization across 13 benchmarks..

Role: First Author.

“Can We Make Feature-3DGS Faster, Better, and Smaller?”

We accelerate and shrink Feature-3DGS with semantic-aware Gaussian pruning and consistency loss, keeping fine details while boosting FPS and mIoU across Replica and Gopher/LindHall scenes.

Abstain-R1: Absent Recognition and Calibration in Post-Training of LLMs

Project Page

We propose Abstain-R1 that explicitly learns when to abstain on unanswerable queries and generates semantically meaningful post-refusal clarifications, improving refusal calibration while preserving strong performance on answerable prompts.

ActionLMM: A Large Multimodal Model for Detailed Action Description in Long Videos

We observe that existing video-language models (e.g., Video-LaMA, VideoChat) fail to capture fine-grained human actions because of data scarcity, modality misalignment, and the difficulty of modeling long clips. Introduces a dual-branch Q-Former that jointly learns from raw video frames and 3-D pose sequences, supported by video- and motion-memory banks, and trained on the new 30 k-video ActionCap-30k dataset. ActionLMM delivers fine-grained action captions and surpasses Video-LLaMA and other baselines on VQA, video captioning, and action captioning benchmarks.