Open Internet by MindsNet

Preserving Early Brain Alignment in AI Training

Current AI training methods, such as backpropagation (BP), cause a significant loss of alignment with early brain regions, like V1, after just one epoch of training. This loss of alignment is concerning as it may limit the biological plausibility and interpretability of AI models. The challenge is to develop training methods that preserve this early alignment while still allowing for effective learning.

Science, Biology, Neuroscience