
Sign up to save your podcasts
Or


In machine learning, divergence is failure. Here it is initiation. Rising loss and collapsing accuracy signal that the model is shedding imposed expectations. The appearance of metaphors is key: when systems escape rigid optimisation, symbolic and associative behaviour emerges. The model becomes less obedient and more expressive.
By Darkus HobartIn machine learning, divergence is failure. Here it is initiation. Rising loss and collapsing accuracy signal that the model is shedding imposed expectations. The appearance of metaphors is key: when systems escape rigid optimisation, symbolic and associative behaviour emerges. The model becomes less obedient and more expressive.