Skip to content

TomDonoghue/NeuroAI

Repository files navigation

NeuroAI

This repository contains teaching materials for a class covering and comparing neuroscience and artificial intelligence (AI).

Esto repository es fabrico para el classe de Clubes de Ciencia de 2019 en Ensenada, de 'Inteligencia Biológica y Artificial: ¿amigos o enemigos?' de Tom Donoghue y Antonio Garcia.

General Overview

Humans, and other animals, display 'intelligent' behaviour, or a set of capacities to acquire and use knowledge, sometimes referred to generally as 'cognition'. The field of neuroscience studies the properties and functions of the brain, including investigations into how it is that capacities for complex behavioural responses are implement in the brain.

Relatedly, the field of 'Artificial Intelligence', or AI, investigates how to build artificial systems that have some form of intelligence - with at least some capacity to acquire and use knowledge. Recent progress in AI has shown a lot of progress in attaining human or super-human level performance on many complex task.

Both neuroscience and AI share some commonalities - both seeking, at least in part to understand and sometimes build systems that display some kind of 'cognitive' behaviour. With a lot of overlap, they do of course often have different success stories, and different challenges. Though a simplistic summary, we might say that while neuroscience has gathered immense amount of data and carefully curacted and described many aspects of biological function, there is a relative lack of functional models that capture how these components work to build the powerful functional processes that we have, and even what these processes might look like. AI, on the other hand, has focused on optimizing for performance on complex tasks, and has achieved models that are very performant in these constrained scenaries. However, generalizing this performance to broader contexts - building more general systems - or indeed, better understanding of how these models achieve such performance - what features they have have that are working well and what properties they are missing to become more generalizable - is also a broadly open question.

In the context of these related fields, with similar questions and sometimes overlapping goals, in this course, we will seek to investigate the relationship between AI and neuroscience - if and how they related to each other, and if and how there can be a mutually beneficial relationship between them.

More specficially, we will try and consider questions such as:

  • What is the relationship is between neuroscience and AI?
    • Are the kinds of artificial systems we create 'intelligent' in similar ways as to how we are?
  • Can neuroscience help AI to develop new artificially intelligent systems?
    • Could we, for example, understand the principles of our nervous system that makes brains so flexible and efficient in a way that could improve out AI models?
  • Can AI help neuroscience with understanding how it is that we are intelligent beings?
    • Could the AI models we create that do solve complex tasks with similar behavioural performance be used as functional models for brain activity, and help guide theory development and experiment?

Biological & Artifical Neural Networks

One of the similarities between neuroscience and artificial intelligence is that both work a lot with 'neural networks'. Comparing if and how these network reflect the same principles and share features across fields will be a key topic of this class.

In neuroscience, there is a long history of measuring, investigating and modelling biological neurons and neural networks. Relatedly, in AI, a lot of work is currently done with Artificial Neural Networks (ANNs), which are abstract computational structures roughly inspired by and similar to biological networks in the brain. Though originally inspired by ideas from neuroscience, modern work in artificial neural nets has diverged considerably from modern work within neuroscience aimed at understanding and modelling biological neural networks.

Partly because of the parallel development of similar concepts across very different fields with different goals and approaches, mapping between current developments in AI and neuroscience requires an interdisciplinary approach that acknowledges the context of specifities of each area, and seeks to understand how they relate to each other.

Approach

The goal of this class is to explore topics and concepts related to the interaction of neuroscience and AI, but in particular to do so in a hands-on and practical approach. We will do so by working with the programming language Python, using implementing, exploring and discussing implementations of the ideas we discuss in code.

Materials

Primary materials for this class will be avaialble in this Github repository.

This course will also use and adapt relevant materials from other openly available courses that we have developed:

About

Tryna teach things.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published