AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
Different approaches to optimizing edge connections in neural networks
The challenge is to optimize the wiring of edges in a computation graph to avoid over squashing. By selectively adding or removing edges, the over squashing problem can be reduced. This is similar to cellular automata, where discreet rules are transformed into a continuous version. The process is called morpha genesis, and it involves training a CNN to reconstruct an image while damaging and destroying parts of it. The image will then recreate itself as an emergent property.