Zhe Jin

Re-encoding a compact set of distilled synthetic images during test-time adaptation dynamically realigns source anchors with shifting model weights, eliminating both catastrophic forgetting and pseudo-label drift under severe distribution shifts.

Pure semantic similarity is insufficient for multi-user agent privacy, but pre-filtering retrieval candidate pools with physical biometrics creates an effective 27% performance barrier against non-owner data access.

Shared KV caches inherently leak private context when gated by semantic relevance alone, but binding biometric probes directly to cache blocks caps unauthorized memory access below 2% without sacrificing sub-30-token prefill speeds.

Coarse indexing can dramatically speed up long-video retrieval without sacrificing answer quality, achieving up to 1.7x acceleration.

Breaching one robustness-optimized defense can expose an entire family of defenses, with purification methods showing an alarming 80.4% transfer attack success rate.

Achieving precise control over human motion and camera trajectories in 3D scene animation could redefine the realism and interactivity of video generation.

Event cameras can now achieve surprisingly high face recognition accuracy, even under poor lighting, by transferring knowledge from standard RGB face models.