This is a project based course, focusing on the connection between neuroscience and AI technology. The brain is a strong, robust, and highly efficient adaptive controller which seamlessly combines electric bursts and analog chemicals.  It is capable of contextual perception, prediction, and generation, with multi-scale and multi-level activity. Many of the large leaps in AI come from the understanding of human’s cognition and animal behavior.  The course is a must background for research in the intersection of AI and neuroscience.  We will discuss additional learning architectures beyond the famous deep networks, such as Kohonen self-organizing networks, Hopfield memory networks, and networks that change activity by context; involve the most advanced ideas that improve AI including lifelong learning and low-energy analog computing; and further introduce neuroscientific findings to improve future AI – such as neuromodulation, dendritic trees, diversity of components, sequence generations, and multi-objectivity. While the course has some homeworks, a chief part of the grade will come from active participation, project, and presentation. 

The course requires successful completion of previous courses in AI and Machine Learning (311, 389 or 589) or permission by teacher. Delivery method: Students can come to class or be on the zoom – both are allowed. Presentations will be in person.

  

3 Credits

Class Hours: Mondays 4-6:30PM, LGRC A104A

 

Schedule

1. Single Neuron Models
February 2	McCulloch & Pittz, Hubert & Wiesel, Learning and Adaptation, Perceptron
2. Hierarchical Structures
February 9	Sensory Hierarchies (e.g. visual), Backpropagation Learning
February 16	Holiday – President’s Day
February 19	Forward Propagation Learning, Generalized Hierarchies for Abstraction and Generalization
February 23	Generalization Predictions in AI, Generating Neural Structures
3. Recurrent and Lifelong Learning
March 2	Turing Capabilities and Super Turing Adaptivity
March 9	Time Series Prediction
March 16	Holiday – Spring Break

March 23a	Lifelong Learning, Adaptive Architectures, Neuromodulations for Continual Adaptivity
4. Memory in Biology and Engineering
March 23b	Stable Memories and Dynamical Systems, Hopfield Memories, Generalized Hopfield for Large Capacity
March 30	Dynamic Energy Landscape and Sequence-Memories, Lifelong Updates of Memories, and Biophysics
5. Unlearning
April 6	Problem Learning and Hardship of Unlearning, Reconsolidation and Relaxing Fears
April 13a	Unlearning in AI and Data Science
April 13b	Projects: Concepts and Presentations
April 20	Holiday – Patriot’s Day

6. Low-energy Adaptive Reinforcement
April 24	Reinforcement Learning (RL), Low-energy (RL) with Multi-Step-Size, Fast RL on Sequences, RL with Adaptive Goals
7. Advanced Generators
April 27	Attentions, Sequences, State-space Modeling
May 4	Diffusion Models and Brain’s Chemicals, Traveling Waves in Biology

There is no textbook required. However, some recommended books can be used that cover the first few classes:

• “Neural networks: A comprehensive foundation”. Simon Haykin
• “Introduction to the theory of neural computation”. Hertz, Krogh, Palmer
• (Fun historic read) “Parallel Distributed processing”. Rumelheart, McCelland.