Neural Maturation Provides the Stability of Representation and the Solution for Understanding Complex Concepts

Hiroyasu Watanabe

DOI: https://doi.org/10.1101/2024.09.22.614315

2024-09-25

Abstract:The mechanisms of the human mind remain a major mystery for science. Progress in these efforts has come from psychology, behavioral science, physiology, and biology. Especially in the past 40 years, advances in molecular biology have expanded significantly. Despite these numerous expansions in understanding the mechanisms of higher brain functions, there are still significant gaps in integrating each research result. In particular, the gaps between gene function and the overall phenotypes related to the mind have not been elucidated at all. Recent technological progress in realizing human-like brain functions provided by artificial intelligence has progressed day by day. However, major parts of the current astonishing progress in AI do not necessarily require an understanding of modern molecular biology. Here, we examine the significance of neural maturation in neural network-based information processing models. In both supervised and reinforcement learning paradigms, immature transmission could learn properties similar to normal high-fidelity conditions before reaching catastrophe points. However, continuous long-term learning induced a loss of learning results. Then, I looked for genes with physiological significance for neuronal transmission fidelity. I found that the candidate gene KCNH7 is expressed at higher levels during the development of the mouse brain. In the simulated neural model with the KCNH7 channel property, the excitation threshold increased, providing a linear response property. These properties enhanced the stability of representation recall and enabled the understanding of complex concepts in unsupervised learning models. These results demonstrate the significance of neural maturation in achieving higher recognition abilities in adults.

Neuroscience

What problem does this paper attempt to address?

The paper attempts to address the importance of neural maturation in neural network information processing models, particularly in enhancing advanced cognitive abilities in adults. Specifically, the paper explores the following aspects:

1. **Neural Maturation and Information Processing**: By introducing randomness to simulate immature neural transmission and studying its impact on supervised learning and reinforcement learning. The results show that under random conditions, long-term continuous learning leads to the loss of learning outcomes, mainly due to the divergence of activation values.

2. **Genetic Role**: Identifying candidate genes with physiological significance for the fidelity of neural transmission. The study found that the KCNH7 gene has a high expression level during mouse brain development and that this gene is involved in regulating the excitability threshold of neurons, thereby enhancing the stability of representation recall and aiding in the understanding of complex concepts.

3. **Neuron Simulation**: By incorporating KCNH7 channel characteristics into the neuron model, it was found that this characteristic can suppress noise-induced variations and maintain the firing of action potentials under high-frequency stimulation, thereby improving the fidelity of neural transmission.

4. **Complex Information Processing**: The study examined the learning stability under different activation threshold conditions in Spiking Neural Networks (SNN) and Restricted Boltzmann Machines (RBM). The results indicate that higher thresholds provide better stability for concept learning.

In summary, through theoretical modeling and experimental validation, this paper reveals the importance of neural maturation in achieving advanced cognitive abilities in adults and provides new insights into understanding the mechanisms of mental disorders.