Search Papers | Poster Sessions | All Posters
Poster C98 in Poster Session C - Friday, August 9, 2024, 11:15 am – 1:15 pm, Johnson Ice Rink
Internal Neural Noise Progression for Emergent Classification Robustness
Suayb Arslan1 (), Hojin Jang2, Pawan Sinha1; 1MIT, 2Korea University
Mammalian perceptual development follows a largely consistent progression along several dimensions. Acuity and chromatic sensitivity improve over the first several months of life after birth. Recent evidence suggests that the progressions may have adaptive value by inducing the formation of receptive field structures that enable later resilience to spatial or chromatic degradations. Here we examine whether the developmental changes in neural noise may follow a similar logic, i.e. does the temporal progression in neural noise lead to benefits in classification performance, especially under challenging conditions? Our preliminary results in two distinct experimental settings indicate that progressions from high network noise to lower levels lead to phenomena similar to those associated with stochastic resonance. These findings not only provide a potential teleological account of noise progressions in biological systems, they also suggest useful training regimens for artificial vision systems to improve their robustness.
Keywords: Noise training biomimetic training brain noise progression stochastic resonance