Seeing the Mind, Educating the Brain (29) - Stanislas Dehaene

The talk provides a clear and insightful demonstration of how cognitive computational models can be used to dissect the mechanisms of human decision-making. Wyart’s framework of ‘what’, ‘how’, and ‘why’ questions is a useful heuristic for structuring research in cognitive neuroscience. The presentation is grounded in a specific study from his group, which adds credibility and allows for a detailed walkthrough of the methodology. The use of Bayesian observer models as a benchmark is appropriate and well-explained. The MEG decoding results are compelling, showing a progressive selection of relevant information across the cortical hierarchy, and the finding that integration is abstract and predictive of accuracy is particularly noteworthy. The comparison of two possible integration strategies for 2D decisions is elegantly designed, and the conclusion that the brain projects evidence onto a tilted axis is supported by the decoding results. The final section on artificial neural networks is brief but raises interesting questions about the normative reasons for this strategy. However, the talk is concise and some details are glossed over, such as the specifics of the neural network optimization and the exact statistical methods used. The speaker assumes a certain level of familiarity with the concepts, which might limit accessibility for a general audience. The sources cited are primarily the speaker’s own work and a few key references, but the talk would benefit from more extensive referencing to the broader literature. Overall, the talk is scientifically rigorous and provides valuable insights into the neural basis of perceptual decisions.

The title accurately reflects the content: a presentation within a colloquium on cognitive neuroscience, focusing on computational models of human cognition.

8/10

The talk is delivered by a recognized researcher in cognitive neuroscience, based on a published study from his group, and presented at a prestigious institution (Collège de France). The methodology is clearly explained, and the results are contextualized within existing literature. However, as a conference talk, it lacks full methodological details and peer-review transparency.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 00:00Introduction and motivation for using cognitive computational models.
• 02:00Presentation of the 'what', 'how', 'why' framework.
• 05:00Description of the 2D perceptual decision task and Bayesian observer model.
• 10:00MEG decoding results for 1D decisions, showing progressive selection.
• 15:00MEG decoding for 2D decisions, revealing tilted integration.
• 18:00Use of artificial neural networks to explore the 'why' question.

• Collège de France - Seeing the Mind, Educating the Brain colloquium — Official page of the colloquium where this talk was given.
• Stanislas Dehaene's chair page at Collège de France — Information about the chair holder and related resources.
• YouTube playlist of Stanislas Dehaene's lectures — Playlist containing recordings of related lectures.

• Dehaene, S., Sergent, C., & Changeux, J.-P. (2003). A neuronal network model linking subjective reports and objective physiological data during conscious perception. PNAS, 100(14), 8520-8525. — The 2003 PNAS paper mentioned by Wyart as an inspiration for cognitive computational modeling.
• Mante, V., Sussillo, D., Shenoy, K. V., & Newsome, W. T. (2013). Context-dependent computation by recurrent dynamics in prefrontal cortex. Nature, 503(7474), 78-84. — The Nature paper from the Newsome lab that studied neural geometry of feature selection, as referenced in the talk.

• www.college-de-france.fr
• www.fondation-cdf.fr
• fr.linkedin.com
• www.threads.net