Cognitive Neural Dynamics
To learn dynamical approach to brain computation and cognition.
Students learn how the dynamics of neural networks, especially recurrent neural networks, contribute to cognitive computation in the brain circuitry, such as memory, decision making, inference, and language processing. The environment external to the brain keeps changing dynamically; therefore, the brain's principles of information processing are likely to be dynamic. We focus on the theoretical treatments of neural network dynamics of binary, analog, and spiking neurons. Both classic and advanced models are studied. Computational implications of biological and artificial learning rules are also explored.
The course targets students who are interested in computational neuroscience, artificial neural networks and Ai, learning theory, statistical physics, and nonlinear dynamical systems.
This course focuses on theoretical approaches. The course consists of one class every week. We select a set of original papers at the beginning of the course. I will recommend some papers but students can also nominate papers of their interest. After lectures to overview the field, students are asked to read these articles and explain the essential results during the class. In addition, each student is asked to consider a small project that extends the findings of any paper discussed in the class. The students should present their results in the final one or two classes.
The course will cover a broad range of models related to recurrent neural network dynamics within time limitations. The following are typical examples.
1. Memory processing models (Associative memory, Hippocampal circuit models, etc)
2. Reservoir computing
3. Random neural networks
4. Learning rules to train recurrent networks
5. Excitation-inhibition balance in computation
6. Cortical oscillations and synchrony
7. Navigation and cognitive processes
8. Dendritic computation
9. Reinforcement learning
Paper presentation and discussion during each class (50%). Project proposal (25%) and presentation (25%).
Students are encouraged to have basic knowledge of statistical physics, stochastic dynamics, and machine learning. Basic skills in mathematics, programming, and computer simulations are required.
Theoretical Neuroscience: Computational and Mathematical Modeling of Neural Systems. Peter Dayan and L. F. Abbott. The MIT Press (Paperback, 2005)
Neuronal Dynamics: From Single Neurons to Networks and Models of Cognition. Wulfram Gerstner, Winner M. Kistler et al. Cambridge University Press 2014.
NEW for AY2024