CE⁴L: Continual Ego, Exo, and Ego-Exo Learning
Introduces a multi-view continual learning benchmark spanning four video tasks and VISTA, a lightweight adapter method with training-free subspace routing for simultaneous task and viewpoint shifts.
Publications
* Equal contribution. My name is shown in bold.
2026
FlyPrompt: Brain-Inspired Random-Expanded Routing with Temporal-Ensemble Experts for General Continual Learning
Combines brain-inspired random-expanded routing with temporal-ensemble experts to learn from single-pass, evolving data streams without explicit task boundaries.
MePo: Meta Post-Refinement for Rehearsal-Free General Continual Learning
Refines pretrained representations using pseudo task sequences and meta-learning, then uses a meta covariance reference to align outputs during rehearsal-free continual learning.
Spacing Effect Improves Generalization in Biological and Artificial Systems
Connects spaced temporal ensembling in neural networks with memory generalization in fruit flies, showing how learning intervals and experience variability jointly shape generalization.
ComMem: Complementary Memory Systems for Test-Time Adaptation of Vision-Language Models
Pairs a fast-updating visual memory with slowly refined textual prototypes to accumulate complementary knowledge and maintain cross-modal consistency at test time.
2025
Right Time to Learn: Promoting Generalization via Bio-inspired Spacing Effect in Knowledge Distillation
Introduces a time interval between teacher and student updates in online and self-distillation, connecting improved generalization to flatter optimization landscapes.
MindPainter: Efficient Brain-Conditioned Painting of Natural Images via Cross-Modal Self-Supervised Learning
Uses cross-modal self-supervision, pseudo-brain signals and a brain adapter to condition image inpainting and outpainting on visual brain recordings.
2024
Orchestrate Latent Expertise: Advancing Online Continual Learning with Multi-Level Supervision and Reverse Self-Distillation
Coordinates experts at multiple network depths with multi-level supervision and reverse self-distillation to balance underfitting and overfitting in online continual learning.
Preprints & work under review
Brain-Inspired Hierarchical Modularity for General Continual Learning
Combines expert specialization with integration across spatial and temporal scales, adapting pretrained models to evolving streams in vision, multimodal, video and embodied learning.
Domain Generalizable Continual Learning
Introduces continual learning across tasks and domains, using adaptive Domain Transformation to separate semantic and domain information and align predictions across distributions.
Ongoing work
CoSSR · AI for multiphoton calcium imaging
Manuscript in preparation
Co-designing acquisition and self-supervised reconstruction for high-throughput calcium imaging. Related Chinese patent application: CN122018128A, published May 12, 2026.