Tag
This paper introduces AgentCanvas, a typed-graph runtime for embodied agents, and KDLoop, a coding-agent search procedure, to automate the design of embodied agent architectures, evaluating across multiple embodied tasks and revealing challenges like rollout noise and local edit basins.
Hyung Won Chung shares at MIT that after testing 5760 architectures at Google, the original 2017 Transformer was the best, then he moved to OpenAI to train o1. He claims 99% of AI research is theater.
This paper presents Conv-VaDE, a variational deep embedding model for interpretable EEG microstate discovery that jointly learns topographic reconstruction and probabilistic soft clustering. It includes a systematic architecture search evaluated on resting-state EEG data to determine optimal model configurations for stability and interpretability.