Predictive Brains, Taught as Method: Zurich's Computational Psychiatry Course
- The Computational Psychiatry Course (CPC) 2026 from the Translational Neuromodeling Unit (University of Zurich and ETH Zurich) is a six-day hybrid course that teaches the modeling toolkit behind predictive processing and active inference; it is open to clinicians and researchers at any level (MD, Master, PhD, postdoc, PI) from psychology, medicine, neuroscience, engineering or physics.
- It runs 7–12 September 2026, on-site in Zurich or live online via Zoom; the online Main Course ticket costs 50 CHF and is free (0 CHF) for participants resident in listed low-income countries, and no CV or application is required.
- You move in a structured way from the clinical picture to the mathematics – Bayesian inference, the Hierarchical Gaussian Filter, dynamic causal modeling of effective connectivity, reinforcement learning – applied to real psychiatric data, so "the brain as a prediction machine" becomes something you can actually fit to behavior.
- Register on Eventzilla through the course page; the slides and tutorial code from every past edition (2015–2025) are already free on the course's GitHub, so you can begin self-studying today.
Predictive processing has become the most cited framework in psychiatry that most clinicians cannot actually use. We can all recite that the brain is a hypothesis-testing organ minimizing prediction error, but the gap between that sentence and a model you could fit to a patient's behavior is enormous – and it is precisely a gap of method. The CPC exists to close it, and after ten years it has become the field's standard on-ramp.
What the six days actually contain
The course, organized since 2015 by the Translational Neuromodeling Unit under Klaas Enno Stephan, is deliberately built for people who arrive from the clinic rather than from mathematics. The Main Course opens with an introduction to psychiatry and psychosomatic medicine, spends the middle days on the computational methods themselves, and closes with a conference-like day of applications from international speakers (this year's lineup includes names such as Pessiglione, Garrido, Valk, Benrimoh and De Choudhury).
The methods are the point. Across the week you cover hierarchical Bayesian models of learning and belief-updating, the Hierarchical Gaussian Filter as a tractable way to estimate a person's precision-weighting and volatility beliefs, dynamic causal modeling of effective connectivity, and reinforcement-learning accounts of decision-making. The optional sixth-day tutorials are three-hour, small-group, hands-on sessions on a single approach (for example, the HGF), all using open-source software – some tutorials require MATLAB, which the organizers help you obtain if you lack a license.
Credibility is not in question here. The TNU is one of the founding groups of computational psychiatry, and its open-science stance is unusually concrete: thirteen public repositories on GitHub hold the full materials from past courses (2015 through 2025), and slides from the live week are shared afterward according to each speaker's preference. You can audit the discipline before spending a franc.
Who should take it, and how
This is a resource for the clinician or clinician-researcher who has read Hohwy or Clark, found the ideas compelling, and wants to stop paraphrasing them. It is equally suited to a trainee planning computational work and to a senior clinician who simply wants to read the field's papers with comprehension rather than faith. The most economical route is to work through the GitHub materials first, decide which tutorial track matters to you, and then buy the 50 CHF online ticket for the live week – attendance earns a certificate, and UZH, ETH and EPFL students can take an oral exam for 3 ECTS.
For a clinician who has read Hohwy or Clark and wants to move from metaphor to method, the CPC is the rare course that hands you the equations and the code rather than only the story.
The Main Course assumes comfort with basic probability and linear algebra, and several hands-on tutorials require MATLAB; the free tier is restricted to residents of listed low-income countries, so most European and CIS-based colleagues pay the 50 CHF online fee, and the live talks are not video-recorded.