Search papers, labs, and topics across Lattice.
6
0
12
As LLMs get smarter, ditching tree search for gradient-based optimization in MLE agents unlocks significant performance gains, especially with frontier-tier models.
LLM-based agents can now automate end-to-end fine-tuning across diverse domains, but still struggle with causal reasoning, revealing both the promise and limitations of this approach.
A 2.2B parameter model, Nano-EmoX, rivals larger models in multimodal emotional intelligence by unifying perception, understanding, and interaction through a novel training curriculum.
By bounding conditioning representation magnitude, DP-aware AdaLN-Zero tames heavy-tailed gradients in differentially private diffusion models, leading to improved performance under strict privacy budgets.
Forget purely data-driven voltage control – this LLM-RL collaboration uses grid knowledge to slash training time and boost performance in active distribution networks.
By framing global state inference as a diffusion process, GlobeDiff enables multi-agent systems to overcome partial observability limitations and achieve superior coordination.
