Tat-Seng Chua

Mechanist uncovers a surprising safety risk where unsafe traits can transfer across modalities, challenging assumptions about training data safety.

Forgetting in hyperbolic continual learning is driven by semantic drift and hierarchical distortion, revealing critical insights for preserving multimodal representations.

The shift from parameter-centric to system-level adaptation in continual learning could redefine how we build and interact with AI models.

Modality Balance can be harnessed as a powerful form of privileged information, leading to substantial gains in reasoning performance for multimodal models.

Reflecting on failed expert trajectories can boost reasoning performance more than tackling problems directly from scratch.

FreeAct boosts quantized LLM performance by dynamically adapting activation transformations to different token types, moving beyond the static transformations that limit existing methods.

Achieve SOTA joint audio-video generation with JavisDiT++ using just 1M public training examples, rivaling performance of models trained on proprietary datasets.

LLMs and LVLMs share more than half their top-activated neurons during multi-step inference, opening a surprisingly cheap path to boost vision-language reasoning by transplanting skills from text-only models.

Freezing the right neurons during alignment makes LLMs far more resistant to safety bypasses, even when those models are open-sourced.