A Day of Insightful Talks

The symposium kicked off with Nao Uchida (Harvard) delivering a thought-provoking keynote on the role of dopamine in reinforcement learning. His talk shed light on circuit motifs underlying reward prediction errors, proposing a mechanism involving feedback and sign reversal of ventral striatal input to midbrain dopamine neurons. This framework offers a compelling way to think about how the brain computes reward-related signals.

Following Uchida, John Murray (Dartmouth) introduced the concept of task generalization through neural kernels—a powerful approach to understanding common frameworks for behavioral and neural generalization across humans and artificial intelligence models. His talk highlighted how computational methods can bridge gaps in our understanding of cognition.

Shantanu Jadhav (Brandeis) then took us on a journey into hippocampal-prefrontal interactions in spatial learning and generalization. He presented compelling evidence that while frontal cortex representations generalize across tasks, the hippocampus maintains environment-specific maps, offering key insights into memory and decision-making processes.

Anne Collins (UC Berkeley) provided a thought-provoking counterpoint to standard reinforcement learning models. Her research suggests that certain human cognitive functions are better explained by a combination of working memory and habitual behaviors rather than classic reinforcement learning frameworks. This perspective challenges prevailing theories and opens new directions for understanding human learning.

Gina Kuperberg (Tufts) brought an exciting cognitive neuroscience perspective, exploring language learning through the lens of modern artificial intelligence. In an era dominated by large language models, her work examines how human linguistic processing aligns (or diverges) from AI-driven models—a particularly relevant topic in today’s rapidly evolving research landscape.

Closing the symposium, Sabine Kastner (Princeton) delivered the Shukart Lecture, offering a fascinating retrospective on her career studying the neural mechanisms of attention. She emphasized the critical role of the higher-order thalamus in attentional control, providing a synthesis of two decades of groundbreaking research.

More Than Just Talks

Beyond the scientific discussions, the symposium fostered community engagement, with students actively introducing speakers, networking opportunities, and valuable interactions with Tufts leadership. These moments underscore the importance of symposia not just as venues for presenting research but also as spaces for fostering collaboration, mentorship, and new ideas.

Looking Ahead

The success of this year’s symposium reaffirms the importance of interdisciplinary dialogue in neuroscience. As we push forward in understanding the brain, these events serve as a catalyst for new questions, collaborations, and discoveries. I look forward to seeing where these conversations lead and to many more engaging symposia in the future.

A Window into Uncertainty: The Prefrontal-Thalamic Connection

Our work focuses on how the prefrontal cortex and thalamus interact to manage uncertainty and enable flexible behavioral responses. Using tree shrews as a model, we designed a hierarchical rule-switching task to test how these animals adapt their decisions in the face of conflicting or ambiguous cues. This task mirrors real-world decision-making scenarios, such as deciding whether a failed strategy is due to poor execution or a fundamental change in circumstances.

Tree shrews demonstrated remarkable flexibility in these tasks, which correlated with dynamic activity in the transthalamic circuit. Specifically, the thalamus appears to mediate uncertainty by distinguishing between errors caused by sensory noise and those signaling environmental shifts. This “uncertainty filter” ensures that the brain efficiently determines whether to persist with a chosen strategy or adapt to a new one.

The complementarity of prefrontal and thalamic circuitry

This role for the thalamus complements that of the prefrontal cortex. Prefrontal neurons exhibit Mixed selectivity, the ability of neurons to respond to multiple task-relevant features, allowing the brain to integrate information from diverse sources efficiently. This property is ubiquitous across species and brain regions, supporting tasks from basic sensory discrimination to complex decision-making. By leveraging mixed selectivity, the prefrontal cortex achieves scalable and flexible computations. For example, neurons may simultaneously encode both the degree of conflict in a task and the expected reward, enabling rapid and context-appropriate responses. However, this encoding scheme may come with limitations, both in terms of controllability and signal propagation. The finding that the thalamus may demix cortical signals and thereby isolate different forms of uncertainty while also broadcasting these dimixed signals between prefrontal areas is the main finding of the paper. These distinct features of cortical and thalamic circuits are likely related to their architectural attributes—the cortex has internal recurrent excitatory connectivity, while the thalamus does not.

Implications for Mental Health and Beyond

Our findings extend beyond basic neuroscience, offering insights into cognitive disorders like schizophrenia and ADHD, where flexibility often breaks down. For instance, disruptions in transthalamic communication might underlie the rigid or maladaptive decision-making observed in these conditions. Understanding these mechanisms could inspire novel therapeutic interventions aimed at restoring adaptive decision-making in affected individuals.

In addition, this research highlights the thalamus as a critical node in cognitive networks—a stark contrast to its traditional view as a sensory relay center. By showing how the thalamus supports higher-order cognition, our study emphasizes the need for a paradigm shift in how we think about its role in the brain.

Broader Implications for Neuroscience

This study contributes to a growing recognition of the brain’s flexible networks—dynamic collaborations between regions that balance stability and adaptability. These findings align with previous research on thalamic contributions to attention and decision-making, suggesting that the thalamus might act as a “gatekeeper” for cognitive processes.

Moving forward, our research aims to explore how these circuits are modulated by neuromodulators like dopamine and acetylcholine, which are known to play roles in attention and learning. We also plan to investigate whether similar mechanisms operate in humans using advanced imaging and computational modeling techniques.

From Laboratory to Life

The translational potential of this research is immense. By understanding how the prefrontal-thalamic circuit processes uncertainty, we can design targeted interventions to improve decision-making in psychiatric disorders. These findings also inspire broader applications in artificial intelligence, where mimicking the brain’s adaptability could enhance machine learning algorithms.

Closing Thoughts

Our work provides a glimpse into the neural mechanisms that make cognitive flexibility possible. By showing how the prefrontal cortex and thalamus collaborate to resolve uncertainty, we hope to inspire future research into how these circuits can be harnessed to improve both mental health and technology.

This paper reflects years of collaboration and exploration, highlighting the power of basic neuroscience to answer profound questions about the human experience.

References:

The paper: Lam, N. H., Mukherjee, A., Wimmer, R. D., Nassar, M. R., Chen, Z. S., & Halassa, M. M. (2024). Prefrontal transthalamic uncertainty processing drives flexible switching. Nature, 10.1038/s41586-024-08180-8. Advance online publication. https://doi.org/10.1038/s41586-024-08180-8

Media Coverage: https://now.tufts.edu/2024/11/13/teaching-ai-rules-brain

These questions have led us on a largely empirical journey over the last decade, recording from the thalamus and prefrontal cortex in animals solving various tasks. In collaboration with a number of colleagues, we have also tried to summarize our results and understanding of them in the form of neural models. I would like to highlight the latest of those—the paper, “Rapid contextual inference by a thalamocortical model using recurrent neural networks,” led by Wei-Long Zheng, a former postdoc in the lab and now Professor at Shanghai Jiao Tong University.

Wei-Long’s paper, published in Nature Communications, uses a biologically inspired architecture to mimic the prefrontal cortex (as an RNN) and the mediodorsal thalamus (as a single layer feedforward network). The resulting hybrid network also includes some biologically-inspired tricks, such as convergent inputs, a special cortico-thalamic learning rule and a winner take all mechanism in the thalamus (to capture the function of the thalamic reticular nucleus). All told, this hybrid structure outperforms RNNs on a number of decision making tasks, mainly in the ability to infer that tasks have changed and in the ability to solve multiple tasks sequentially without interference. The thalamus-like feedforward network partitions the prefrontal cortex, reducing interference across task representations. There was very nice press coverage on this by Shanghai Jiao Tong University which I encourage people to read– https://news.sjtu.edu.cn/jdzh/20241008/202588.html (It’s in mandarin, but google translate does a very good job). I should also emphasize that this paper was possible because of the very talented collaborator we had—Robert Yang who was amazing to work with and had so many insights throughout. The two trainees involved, Zhongxuan Wu (student at UT Austin) and Ali Hummos (postdoc at MIT) did a terrific job helping Wei-Long with this study.

The broader context: task optimized networks as tools in systems neuroscience

The 2024 Nobel Prize in Physics was awarded to Geoff Hinton and John Hopfield for their pioneering work on backpropagation and associative networks, respectively. These ingredients laid the foundation for much of the neural network tools we have today, including those we use to model the brain. RNNs, with their ability to maintain information over time, have been critical in advancing AI, from language models to reinforcement learning systems. But their potential in neuroscience is equally profound—They are uniquely positioned to answer why questions—such as the fundamental question of ‘why is the brain wired the way it is?’.

The narrower context: RNN-based PFC-thalamus models

Wei-Long’s paper follows and build on two other papers by our lab which I would like to highlight. The first is Rikhye, Gilra and Halassa (2018), where Aditya Gilra developed a similar model (in spirit) and showed for the first time a role for the thalamus in PFC partitioning and the mitigation of catastrophic forgetting. The second is by Hummos et al. (2022), who derived the cortico-thalamic learning rule and showed that it was capable of compressing cortical activity into a task context signal the thalamus carries. Importantly, the Hummos model was not only capable of solving complex human tasks, but also reproducing neural activity patterns seen in the scanner.

A place for NeuroAI in systems neuroscience

The intersection between Biological and Artificial Intelligence is quite exciting. Many institutions are now investing in what some call “NeuroAI”, with dedicated training programs that teach computer scientists about the brain and neuroscientists about computational methods. The future is quite exciting and just like causal tools (e.g. optogenetics) have become a hallmark of systems neuroscience studies (particularly in small animals), it is not unlikely that NeuroAI will be an equally prevalent approach for the interface between systems and cognitive neuroscience.

References:

• Rikhye RV, Gilra A, Halassa MM (2018). “Thalamic regulation of switching between cortical representations enables cognitive flexibility.” Nature Neuroscience.
• Hummos A, et al. (2022). “Thalamic regulation of frontal activity in human decision making” PLoS Computational Biology.
• Zheng W-L, et al. (2024) “Rapid contextual inference by a thalamocortical model using recurrent neural networks.”Nature Communications, 2024.