Search papers, labs, and topics across Lattice.
6
0
12
FreeStyle achieves a remarkable balance between style alignment and content preservation while effectively suppressing semantic leakage in dual-reference image generation.
A fully integrated robot learning stack that bridges the gap from simulation to real-world deployment, enhancing the efficacy of vision-language-action models.
Sparse rewards can be transformed into actionable turn-level feedback, enabling agents to learn from both successful and misleading actions in long-horizon tasks.
Video codecs can slash LLM perplexity by over 1.5x while boosting task accuracy by 21%, revolutionizing model compression strategies.
A new framework for EEG decoding boosts generalization across datasets while maintaining low training overhead and no extra inference costs.
Forget specialized architectures: StepAudio 2.5 proves a single audio-language foundation, shaped by RLHF, can dominate ASR, TTS, and real-time dialogue simultaneously.
