Sveriges mest populära poddar
Embodied AI 101
VetenskapTeknologi

One Learning Rate Doesn't Fit All: Layerwise Spectral Scheduling for Transformers

27 min•31 maj 2026
Shows that modern transformers are highly heterogeneous across layers and proposes layerwise learning rates based on weight spectrum shape, yielding up to 1.5× training speedup on LLaMA/GPT-style models.

Fler avsnitt av Embodied AI 101

Visa alla avsnitt av Embodied AI 101
Hy-Embodied-0.5-VLA: A Massive Bimanual Teleoperation Dataset for Vision-Language-Action
•21 min
Q-Guided Flow: Test-Time Gradient Guidance of Flow Policies
•35 min
Flow Reversal Steering: Guiding Diffusion-Based Robot Policies with High-Level Reasoning
•38 min
Test-Time Compute Scaling for Robot Policies (DIRECT)
•25 min
LabVLA: Bringing Vision-Language-Action to the Chemistry Lab
•42 min
Humanoid-GPT: A Foundation Model for Zero-Shot Humanoid Control
•26 min
CHORUS: Decentralized Multi-Robot Collaboration with a Single Shared VLA Model
•37 min
RISE: Self-Improving Robot Policy with Compositional World Model
•40 min
EmbodiedOneVision: Interleaved Vision-Text-Action Pretraining for General Robot Control
•35 min
Robix: A Unified Model for Robot Interaction, Reasoning and Planning
•35 min

Embodied AI 101 med Shaoqing Tan finns tillgänglig på flera plattformar. Informationen på denna sida kommer från offentliga podd-flöden.