Development of Visual Functions in the Brain and in Artificial Networks
Prof. Se-Bum Paik
October 8(Thu) - October 8(Thu), 2020
12PM
ZOOM (ID: 728-142-6028 )
Neuro@noon Seminar
Date: 12PM, Thursday, October 8th
Place: ZOOM
ZOOM 회의 참가 ID: 728-142-6028
Speaker: Prof. Se-Bum Paik
Department of Bio and Brain Engineering, KAIST, Republic of Korea
Title: Spontaneous emergence of visual functions in the brain and in deep neural networks
Abstract:
Model studies with biologically inspired artificial neural networks have provided insight into the underlying mechanisms of brain functions, particularly regarding the development of various functional circuits for visual information processing. Recent studies using artificial neural networks revealed that various visual functions can emerge from supervised and unsupervised learning, suggesting a possible mechanism of how visual object recognition in the brain arises initially. However, the ability to perform various cognitive functions is often observed in naïve animals, and this raises questions about the origin of early cognitive functions in the brain. In this talk, I will introduce our recent findings that visual cognitive functions such as number sense or face recognition can emerge spontaneously in hierarchical neural networks in the complete absence of visual training. Using a biologically inspired deep neural network, we found that neurons tuned to face images or stimulus numerosity arise in untrained random feedforward networks. These neurons also showed single- and multi-neuron characteristics of the types observed in biological brains, such as Weber-Fechner law. The responses of these neurons enable the network to perform a visual comparison task, even under the condition that the information in the stimulus is incongruent with low-level visual cues. These results suggest that cognitive functions can emerge from the statistical properties of bottom-up projections in hierarchical neural networks, and provide new insight into the origin of early cognitive functions in biological brains, as well as in artificial deep neural networks.