Tropical Cyclone Attributes in Sparse Autoencoder Features of an AI Weather Model

Key points

• Sparse autoencoders (SAEs) provide sparse features for probing hidden states, but how information about multiple physical attributes of an event is organized across them remains unclear.1
• Using tropical cyclones in GraphCast as a test case, we analyze 2,115 six-hourly timesteps from 205 cyclones, relating SAE activations to position, intensity, and pressure structure across five Layer-8 decompositions and 16 processor layers.1
• Each attribute relates to multiple features; position and pressure-structure relations overlap in a smaller subset.1
• These results suggest that the organization of physical-attribute relations provides a reproducible level of description for interpreting cyclone-related information in GraphCast.1

Sources (1)

• [1]Tropical Cyclone Attributes in Sparse Autoencoder Features of an AI Weather Model
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 08:30 PM

  Artificial intelligence (AI) weather models now forecast weather events skillfully, but how they represent these events internally remains unknown.

  Sparse autoencoders (SAEs) provide sparse features for probing hidden states, but how information about multiple physical attributes of an event is organized across them remains unclear.

Extractive summary: sentences quoted from the sources.