Abstract:We consider the information transmission problem in neurons and its possible implications for learning in neural networks. Our approach is based on recent developments in statistical physics and complexity science. We also develop a method to select statistically significant neural responses from the background activity and consider its wider applications. This would support temporal coding theory as a model for neural coding.
What problem does this paper attempt to address?
The main problems that this paper attempts to solve are understanding how neurons transmit information in the nervous system and exploring the impact of this information transmission on neural network learning. Specifically, the research focuses on the following aspects:
1. **Information Transmission Mechanism**: Researchers hope to understand the relationship between stimuli and the responses of single neurons or neural networks, and how these stimuli are encoded into the internal representations of neurons.
2. **Selection of Neuronal Responses with Statistical Significance**: Develop a method to select statistically significant neural responses from background activities (i.e., normal neural activities), which is crucial for supporting the temporal coding theory as a neural coding model.
3. **Micro - architecture of Cognitive Processes**: Study how the micro - architecture of brain circuits supports cognitive processes such as thinking, memory, and decision - making, especially how to coordinate information processing on large - scale spatio - temporal scales.
4. **Dynamics of Neuronal Bursting Activity**: Use methods in statistical physics and complexity science to study the dynamic characteristics of neuronal bursting activity, which may be related to consciousness.
5. **Application of Information Bottleneck and Renormalization Group**: Explore the applications of the information bottleneck method and the renormalization group theory in neuroscience, especially in understanding the information compression and transmission mechanisms of deep neural networks.
Through these studies, the paper aims to provide a unified understanding framework for perception and learning, which requires interdisciplinary cooperation, including the efforts of electrophysiologists, molecular biologists, and cognitive scientists.