CCN 2026 | Keynote: Kenji Doya

Item: CCN 2026 | Keynote: Kenji Doya
Rating: 4
Author: PostSingularity Institute

by Kenji Doya

Life & Natural Systems Neuroscience PSANNeurosciences PSAN5Cognitive and behavioural neuroscience

🎙 Kenji Doya 👥 4K 📅 August 12, 2026 ⏱ 64 min 👁 72 📄 expert opinion 🧭 2026-08-15

Available in: English (current) Français

Keywords

reinforcement learningbasal gangliadopamineserotoninmental simulation

Summary

Kenji Doya, a professor at OIST, delivers a keynote on neural circuits for prediction and action. He begins by introducing his lab’s dual focus on robotics/machine learning and neurobiology, unified by reinforcement learning (RL). He reviews core RL concepts: value functions, temporal discounting, reward prediction error, and the trade-off between exploration and exploitation. He presents his early work on a robot learning to stand up, illustrating delayed reward problems. He then discusses the basal ganglia’s role in RL, highlighting dopamine’s encoding of temporal difference errors and his hypothesis that striatal neurons encode action values. He presents experimental evidence from rat binary choice tasks and calcium imaging, supporting this. He also explores serotonin’s role in regulating temporal discounting, showing that optogenetic stimulation of serotonin neurons increases patience in mice. Moving to mental simulation, he contrasts model-free and model-based RL, presenting a smartphone-based robot that learns to balance using an internal model. He proposes a framework where the cerebellum, basal ganglia, and cortex specialize in different learning types. He describes fMRI experiments on grid navigation, suggesting that mental simulation involves a global brain circuit including cerebellum and basal ganglia. Finally, he discusses neural decoding in mouse parietal cortex during auditory virtual navigation, showing internal model-based state estimation. He concludes by highlighting open questions and the importance of AI-neuroscience collaboration.

217 words

Critical Evaluation

Value of the Information & Strength of the Argument

The talk provides valuable insights into the neural mechanisms of reinforcement learning and mental simulation, synthesizing decades of research. Doya’s argumentation is solid, grounded in both theoretical frameworks and experimental evidence from his own lab. He clearly explains the logic behind each hypothesis and presents data that either supports or refines it. For instance, he shows that serotonin’s effect is not simply on temporal discounting but is context-dependent, highlighting the complexity of neuromodulation. The integration of robotics, machine learning, and neurobiology is compelling, demonstrating the bidirectional benefits of these fields.

Scientific Rigor, Source Quality, Title Accuracy

The talk demonstrates high scientific rigor, with references to key studies (e.g., Schultz et al. on dopamine, and his own publications). However, as a keynote, it does not provide a full literature review, and some claims are presented as hypotheses without exhaustive citation. The title accurately reflects the content, though the talk is broader than ‘prediction and action’, covering also learning and mental simulation. The speaker’s authority is unquestionable, given his extensive publication record and awards.

181 words

Title / Content Match

The title 'Neural circuits for prediction and action' accurately reflects the main theme, though the talk covers broader topics including reinforcement learning and mental simulation.

Quality & Reliability

8/10

The talk is given by a leading expert in computational neuroscience, with a strong publication record and honors. The content is based on established research and his own experimental work, but it is a keynote presentation, not a peer-reviewed article, so some claims are presented as hypotheses.

Key Moments

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

Cited Sources

Okinawa Institute of Science and Technology — Doya's affiliation and lab
Neural Networks (journal) — Doya served as Co-Editor in Chief

Concurring Sources

Schultz, W., Dayan, P., & Montague, P. R. (1997). A neural substrate of prediction and reward. — Foundational study on dopamine and reward prediction error.
Doya, K. (2008). Modulators of decision making. — Doya's own work on neuromodulators and reinforcement learning parameters.

Contribution & Novelties

The talk synthesizes Doya’s extensive research on reinforcement learning and mental simulation, offering a comprehensive framework linking computational models to neural circuits. It highlights novel findings on serotonin’s role in temporal discounting and the neural basis of mental simulation. The integration of robotics and neurobiology provides a unique perspective.

Pour aller plus loin :

Reinforcement learning — Foundational concept.
Basal ganglia — Key brain region discussed.
Dopamine — Neuromodulator central to reward prediction error.
Serotonin — Neuromodulator implicated in temporal discounting.
Temporal difference learning — Algorithm underlying dopamine responses.

89 words

Radar Profile

The radar profile shows high scores across all dimensions, indicating a well-rounded and reliable presentation. The talk is rich in information, technically sound, and based on credible sources, with a strong alignment between title and content.

Reliability 8/10