自己紹介・研究目的
令和5年度入学/ ■SPRING事業 採択学生紹介
数理・ヒューマンシステム科学専攻
令和5年度 大学院入学
李 加一
リ カイチ
Research on the design and application of novel neural networks
Hello everyone, my name is Jiayi Li (李 加一). I am currently in my first year of a Ph.D. program in Advanced Mathematics and Human Mechanisms at the Artificial Intelligence Lab. My research interests include deep learning, machine learning, neural networks, and evolutionary computing. Currently, my research focus is on designing new neural models and constructing novel neural networks.
Artificial Intelligence (AI) has been rapidly developing, and tools such as AI drawing, AI writing, and AI programming have had a significant impact, changing the way people work and enabling them to be more creative. Although these AI tools are already quite powerful, they still rely on the most traditional McCulloch-Pitts neuron model, which has a simple structure, weak computational power, and poor interpretability. Inspired by biology, I aim to design neural models that are more realistic to biological neurons, thus enhancing the interpretability and computational power of neural networks. Currently, I have made improvements based on the dendritic neuron model and designed a dendritic neuron model with adjustable excitatory and inhibitory effects. This model increases the computational power of individual neurons and enhances their reasoning ability compared to traditional neural models. I have also successfully used the newly proposed neural model to construct novel neural networks and obtained excellent results in various real-world applications, such as in medical, environmental, and information sciences. This model outperforms traditional neural models such as multilayer perceptron, support vector machine, and Transformer. I will continue to explore using these novel neural models to construct better neural networks to tackle more complex real-world problems.
(旧フェローシップ事業採択学生)
Artificial Intelligence (AI) has been rapidly developing, and tools such as AI drawing, AI writing, and AI programming have had a significant impact, changing the way people work and enabling them to be more creative. Although these AI tools are already quite powerful, they still rely on the most traditional McCulloch-Pitts neuron model, which has a simple structure, weak computational power, and poor interpretability. Inspired by biology, I aim to design neural models that are more realistic to biological neurons, thus enhancing the interpretability and computational power of neural networks. Currently, I have made improvements based on the dendritic neuron model and designed a dendritic neuron model with adjustable excitatory and inhibitory effects. This model increases the computational power of individual neurons and enhances their reasoning ability compared to traditional neural models. I have also successfully used the newly proposed neural model to construct novel neural networks and obtained excellent results in various real-world applications, such as in medical, environmental, and information sciences. This model outperforms traditional neural models such as multilayer perceptron, support vector machine, and Transformer. I will continue to explore using these novel neural models to construct better neural networks to tackle more complex real-world problems.
(旧フェローシップ事業採択学生)