Neural Models of Cognitive Processes

Course Objective

Computational modelling is an important tool for cognitive neuroscience. Conversely, neuroscience has proven to be a very rich source of inspiration for AI. This course will teach you the background knowledge on the core principles of computational modelling, and dive into the links between AI and (cognitive) neuroscience.

The course is intended to offer insight(s) into what different types of models exist in in cognitive neuroscience, how they can be (and are) used to enrich the field, and it explores what questions arise when evaluating modelling work in this field. We will also investigate the parallels between current issues in computational cognitive neuroscience and open questions in AI.

Course Content

Computational models are an important tool in cognitive neuroscience. A large branch of research focuses on an experimental approach, testing

predictions by means of carefully designed experiments. Models, on the other hand, can integrate experimental results into complete and detailed theories that produce testable predictions. As such, they form a critical step in the empirical cycle by generating predictions for future experiments.

When used appropriately, a model allows for the integration of findings from a wide range of experiments. Rather than merely verbal theories, computational models are rich in detail and allow for a mechanistic view on how the brain produces its behavior.

An old adage from statistics is that ``all models are wrong, but some models are useful''. They are wrong because a model by definition is a simplification of reality, but they are useful when they generate testable predictions. However, it can be difficult to assess whether a model is too much of a simplification, and whether its predictions actually are useful. What makes a model good or bad? To what extent do models need to fit the data? And if multiple models fit the data, how do we choose which is the `better one'?

In addition, modeling papers can at times seem rather enigmatic, and for the untrained reader it is all too easy to get lost in the mathematical equations that make up computational models.

This course takes a learn-by-example approach to give an overview of different modeling approaches that are common in neuroscience. We will start at a high level of abstraction, with models that are used to mathematically describe experimental data, with relatively little regard

for their implementation in the brain. By means of practical sessions, you will get hands-on experience with some of these models and see how they are implemented. By means of debates, you will learn how to assess different models in terms of their strengths and weaknesses. Lastly, you will listen to neuroscience/AI podcast episodes, and present one highlighted researcher's work in class.

Additional Information Teaching Methods

Lectures and discussion, computer tutorial and practicals.

Method of Assessment

Grades are based on a weighted average of performance on a final exam (60%), weekly quizzes (15%), and class participation in the debate sessions (Perusall 10%, Podcast presentations 15%)

Entry Requirements

There is no explicit required knowledge. However, as the practicals have you work with Python code, it might be useful to familiarize oneself with the language. https://www.codecademy.com/learn/python offers a wonderful free online tutorial

Literature

Literature (articles, tutorials) will be provided through Canvas & Perusall.