Summary: A new study delves into the enigmatic realm of deep neural networks, discovering that while these models can identify objects akin to human sensory systems, their recognition strategies diverge from human perception. When prompted to generate stimuli similar to a given input, the networks often produced unrecognizable or distorted images and sounds.
This indicates that neural networks cultivate their distinct “invariances”, differing starkly from human perceptual patterns. The research offers insights into evaluating models that mimic human sensory perceptions.
Comments are closed.