Spatio-temporally efficient coding : A computational principle of biological neural networks
- 주제(키워드) Efficient coding , Hierarchical structure , Neural coding , Neural representation , Neuronal noise , Sensory system , Spike distance , Temporal coding
- 발행기관 Ulsan National Institute of Science and Technology
- 지도교수 Kim, Sung-Phil
- 발행년도 2022
- 학위수여년월 2022. 8
- 학위명 박사
- 학과 및 전공 Ulsan National Institute of Science and Technology Department of Biomedical Engineering
- 세부전공 Human Factors Engineering
- 원문페이지 82
- 실제URI http://www.dcollection.net/handler/unist/200000641211
- UCI I804:31001-200000641211
- 본문언어 영어
초록/요약
One of the major goals of neuroscience is to understand how the external world is represented in the brain. This is a neural coding problem: the coding from the external world to its neural representations. There are two different kinds of problems with neural coding. One is to study the types of neuronal activity that represent the external world. Representative examples here are rate coding and temporal coding. In this study, we will present the spike distance method that reads temporal coding-related information from neural data. Another is to study what principles make such neural representations possible. This is an approach to the computational principle and the main topic of the present study. The brain sensory system has hierarchical structures. It is important to find the principles assigning functions to the hierarchical structures. On the one hand, the hierarchical structures of the brain sensory system contain both bottom-up and top-down pathways. In this bidirectional hierarchical structure, two types of neuronal noise are generated. One of them is noise generated as neural information fluctuates across the hierarchy according to the initial condition of the neural response, even if the external sensory input is static. Another is noise, precisely error, caused by coding different information in each hierarchy because of the transmission delay of information when external sensory input is dynamic. Despite these noise problems, it seems that sensory information processing is performed without any major problems in the sensory system of the real brain. Therefore, a neural coding principle that can overcome these noise problems is needed; How can the brain overcome these noise problems? Efficient coding is one of representative neural coding principles, however, existing efficient coding does not take into account these noise problems. To treat these noise problems, as one of efficient coding principles, we devised spatio-temporal efficient coding, which was inspired by the efficient use of given space and time resources, to optimize bidirectional information transmission on the hierarchical structures. This optimization is to learn smooth neural responses on time domain. In simulations, we showed spatio-temporal efficient coding was able to solve above two noise problems. We expect that spatio-temporal efficient coding helps us to understand how the brain computes.
more목차
Abstract .......... ⅰ
Contents .......... ⅱ
List of Figures .......... ⅲ
1. Introduction .......... 1
2. Scheme of neural coding .......... 5
2.1. Rate coding and temporal coding .......... 5
2.2. Spike distances: measures for rate coding and temporal coding .......... 8
2.2.1. Spike distance focused on temporal coding .......... 9
3. Principle of neural coding .......... 33
3.1. Efficient coding and predictive coding .......... 34
3.2. Spatio-temporally efficient coding .......... 35
3.3. Implementation of spatio-temporally efficient coding .......... 39
4. Spatio-temporally efficient coding: a case of static sensory input .......... 45
5. Spatio-temporally efficient coding: a case of dynamic sensory input .......... 57
6. Discussion .......... 63
References .......... 68
Acknowledgment .......... 82