Lei Lei

Forget specialized architectures: StepAudio 2.5 proves a single audio-language foundation, shaped by RLHF, can dominate ASR, TTS, and real-time dialogue simultaneously.

Distributional counterfactual explanations are now possible for black-box tabular models, thanks to a novel sparse search algorithm that sidesteps the need for gradients.

A dual-level Mixture-of-Experts architecture can effectively unify diverse image restoration tasks within a single diffusion transformer model, achieving state-of-the-art results.

Ditch optical flow for image alignment: DMAligner leverages diffusion models to synthesize novel views, sidestepping traditional warping's occlusion and illumination woes.