What are common principles of natural and artificial intelligence?

The core challenge of intelligence is generalization. Neuroscience, cognitive science, and AI are all questing for principles that help generalization. Major system features that affect generalization include: task structure (multitasking, multiple inputs with same output and vice versa), microcircuitry (nonlinearities, canonical motifs and their operations, sparsity), macrocircuitry or architecture (e.g. modules for memory, information segregation, weight sharing by input symmetry or common development), learning rules (synaptic plasticity, modulation), and data stream (e.g. curriculum).

We aim to present current understanding of how these issues arise in both natural and artificial intelligence, comparing how these system features affect representations, computations, and learning. We provide case studies and coding exercises that illustrate these issues in neuroscience, cognitive science and AI.

• Learning Goal 1: A common understanding and vocabulary to describe challenges faced by naturally intelligent systems
  • Describe core shared concepts in neuroscience, cognitive science and machine learning and how they differ to each other
  • Describe and implement different ways in which an ANN can be compared to a BNN
  • Describe multiple scales of computation, and multiple scales of study (e.g. Marr’s levels, what/how/why?)
• Learning Goal 2: Experience a multiplicity of approaches and interests at the intersection of neuro and AI; be able to describe some of these approaches and interests
• Learning Goal 3: Be able to practically implement NeuroAI models
  • Coding and training models
  • Adding more features to existing models
  • Debugging (within guardrails)
  • Interpreting, analyzing and critiquing existing models
• Learning Goal 4: Complete research that deals with difficulties in NeuroAI
  • Writing down a problem in a way that makes it tractable
  • Interacting with other people from other disciplines fruitfully
  • Do research (reading papers, implementing previous SOTA, coding new methods, evaluating diff methods) in NeuroAI
  • Communicating their research in ways that are comprehensible to their target audience

All our content is open source, you can see the NeuroAI course book here.

Day 1: Big Picture and Introduction to NeuroAI Field

Day 2: Comparing Tasks

Day 3: Comparing Networks

Day 4: Projects Day

Day 5: Microcircuits

Day 6: Macrocircuits: Architecture and Modules

Day 7: Cognitive Structures

Day 8: Micro-learning: Synaptic Plasticity

Day 9: Macro-learning: Objectives

Day 10: NeuroAI Mysteries & Project Presentations

In parallel with course work, students will apply the techniques they are learning on real research projects matching their interests. These are done with a smaller set of peers and a Project TA that will help groups formulate research questions and explore available open science datasets.

Research Project Data Sets

• Exploring local learning rules and credit assignment in artificial and natural intelligence
  • Development Lead by Colleen Gillon, Imperial College London
• Leveraging neural architectural priors and modularity in embodied agents
  • Development Lead by Divyansha Lachi, Georgia Tech
• Characterizing computational similarity in task-trained dynamical models
  • Development Lead by Chris Versteeg, Emory University

It is our priority to provide affordable, quality education in computational sciences to anyone anywhere in the world, and we must also make sure that our teaching assistants receive fair compensation for their work.

While we are supported by generous donations from a variety of foundations and industry partners, we strive to make our live courses sustainable by charging a small, regionally-adjusted tuition fee. This fee is (1) substantially lower than those of traditional summer schools, (2) determined by the location, career position, and funding status of each student individually, and (3) essentially in its entirety is used to pay our teaching assistants.

See fee calculator here.

We do not want tuition fees to be a barrier for anybody, so fee waivers are available to students that need them without any impact on their admission. However, if you can pay even a part of your fee without hardship, we kindly ask that you do so. There is also an opportunity to pay more than your fee if you would like to help subsidize fee waivers for students with less financial means than yourself.

These roles are also outlined on our general courses page and in our Portal.

Students

Students have some experience in computational neuroscience as well as deep learning or AI research, both of which require a familiarity with Python. Students work in small learning groups (“pods”) to complete coding tutorials and develop a research project – all under the guidance of our teaching assistants. See more about the student role on the Courses page.

Teaching Assistants

Teaching assistants have some knowledge of AI research as well as familiarity with concepts in computational neuroscience and are experienced in Python. Teaching assistants guide students through the tutorials and help them develop projects. As part of the application for teaching assistants, you are required to make a 5-minute video using a pre-made tutorial. We want as many qualified teaching assistants as possible. Our best advice is to let your personality shine through and take your time moving through the material.  This will ensure the content is clear and easy to follow. Please follow this link to the instructions on what to include in your video as well as detailed recording instructions.

Teaching assistants are paid, full-time, temporary, contracted roles. TA compensation is provided by this calculator. See the Courses page for more on TA roles.

Link