Abstract representations emerge naturally in neural networks trained to perform multiple tasks https://t.co/DHreZEgw9E pic.twitter.com/AU3SYMNUHO
— Kevin Mitchell (@WiringTheBrain) October 21, 2021
Check out this thread as well:
We show that abstract latent variable representations reliably emerge in feedforward neural networks trained to perform multiple classification tasks -- similarly to how humans and other animals receive coarse feedback about the world. 2/8 pic.twitter.com/QdwuDht4Uv
— Jeff Johnston (@wjeffjohnston) October 21, 2021
No comments:
Post a Comment