Users especially liked connections between interpretability and unsupervised scientific simulation via surrogate models discussed during the ICML presentation on interpreting physics in video world models.
Based on 1 visible X reactions from 1 accounts; directional sample.
Ask a question below.
Published answers will appear here.
The other points I especially liked from the audience: - Connecting unsupervised scientific simulation via interpretability to existing surrogate models methods in physics - Bottleneck autoencoder or compression paradigms for video representations instead of sparse coder paradigms (some video representations seem to suffer from overly high-dimensionality and not superposition!)
Great turnout at our ICML poster for "Interpreting Physics in Video World Models"! I especially liked the questions on whitebox evals from the students studying manufacturing. I sadly couldn't meet everyone who reached out before I left Seoul. DM me if you'd like to chat on Zoom:)
The other points I especially liked from the audience: - Connecting unsupervised scientific simulation via interpretability to existing surrogate models methods in physics - Bottleneck autoencoder or compression paradigms for video representations instead of sparse coder paradigms (some video representations seem to suffer from overly high-dimensionality and not superposition!)
Users especially liked connections between interpretability and unsupervised scientific simulation via surrogate models discussed during the ICML presentation on interpreting physics in video world models.
Based on 1 visible X reactions from 1 accounts; directional sample.
Ask a question below.
Published answers will appear here.