The Neuroarchitecture of Learning and Memory

September 26-October 3, 2026

Director: Sheena Josselyn

SickKids, University of Toronto, Canada

Faculty:

Laura Colgin, University of Texas, Austin, USA (Co-Organizer)

Daniel Levenstein, Yale University, New Haven, USA

Joshua Johansen, RIKEN Center for Brain Science, Wako, Japan

Monika Schönauer, University of Freiburg, Germany

Michael Shadlen, Columbia University, New York, USA

Priyanka Rao-Ruiz, Vrije Universiteit Amsterdam, The Netherlands

Sheena Josselyn, SickKids, University of Toronto, Canada

Learning and memory are fundamental to adaptive behaviour across biological and artificial systems. Advances in neuroscience and artificial intelligence (AI) are converging to reveal shared principles of how information is encoded, stored, and retrieved. In rodent models, memory has been traced to sparse, distributed populations of neurons—engrams—that are both necessary and sufficient for memory expression. These discoveries have provided unprecedented mechanistic insight into how experiences are transformed into lasting neural representations. In parallel, deep learning and other AI approaches have modelled core features of memory processes, including pattern separation, generalisation, and flexible inference. Human neuroscience has further extended this framework, demonstrating that hippocampal and cortical circuits support abstract representations that enable reasoning, prediction, and complex decision-making.

This interactive Advanced Course will bring together researchers spanning systems neuroscience, cognitive psychology, computational modelling, and AI to address key questions: How do engrams support the richness of memory across different species? What computational principles link memory networks in the brain with artificial systems? How do oscillations in the brain contribute to memory? Do humans have engrams? By integrating insights from rodents to humans and from engrams to AI, we aim to chart a path toward a unified understanding of memory that bridges biological and artificial domains, with implications for treating memory disorders and building more adaptive intelligent systems.

Daniel Levenstein

Attractor-based models of the hippocampus and its contribution to learning and memory

Computational models are crucial for connecting the biological properties of the hippocampus with its functions in learning and memory, as well as formalizing our hypotheses on how these cognitive abilities might be implemented in the brain. During the lectures, I will introduce a widely-used class of models where memories and cognitive maps are stored through strong connections between similarly-tuned neurons, and I will explore their relationship with the hippocampus’ anatomical structure and physiology during wakefulness and sleep. I will also examine recent models inspired by modern AI systems, where the prediction of sensory stimuli is used to learn hippocampus-like cognitive maps within rich virtual environments, without relying on pre-existing knowledge of space.

Joshua Johansen

Constructing emotional memories in hierarchically organized brain circuits

Innately aversive experiences profoundly shape brain function, generating emotional states and instructing the formation of long-term emotional memories. Yet, more complex emotional memories emerge from internal brain models that evaluate sensory stimuli in the context of past experiences and the organism’s current physiological state. Our laboratory investigates the neural circuits and cell coding mechanisms that transform aversive experiences into both simple and complex emotional states, thereby regulating memory formation and guiding adaptive behavior. I will present recent work identifying a brainstem circuit that integrates external sensory and internal motor features of innately aversive experiences to establish a sensorimotor state in the amygdala, driving aversive memory formation. In contrast to this bottom-up pathway, we have found that the medial prefrontal cortex encodes more abstract emotional states by constructing an internal associative model that performs emotional inference through projections to the amygdala. Together, these findings support a new hierarchical circuit framework of emotion, in which sensory, bodily, and cognitive factors interact across distributed neural systems to support the formation and recall of different types of emotional memories.

Sheena Josselyn

Engrams, representations and memory in mice

Understanding how the brain encodes, stores, and uses information is a central goal of neuroscience. Many neuropsychiatric and neurodegenerative disorders, including autism spectrum disorder, post-traumatic stress disorder (PTSD), and Alzheimer’s disease, may arise from disrupted information processing. Thus, uncovering the neural mechanisms by which information is represented in the brain is not only key to understanding normal cognition but also essential for developing targeted therapeutic strategies.
Memory can be defined as the persistence of internal representations acquired through experience, and the capacity to reconstruct these representations across time. The enduring physical changes in the brain that encode such information are referred to as engrams. Although the idea of a physical memory trace can be traced back to ancient Greek philosophy, it was not formally articulated until 1904 when Richard Semon coined the term engram. Despite this long conceptual history, identifying the precise neural substrates of an engram has proven remarkably difficult, in part because memory is encoded across multiple levels, from epigenetic and synaptic modifications to coordinated patterns of neuronal ensemble activity.
Our laboratory seeks to understand how specific neurons are recruited, or allocated, to a given engram, and how membership within these neuronal ensembles may evolve with time, plasticity, or new learning. By combining molecular, imaging, and behavioral approaches in mice, we aim to link changes in neuronal excitability and network dynamics to the stability and flexibility of memory representations. In these lectures, we will delve into historic and recent findings toward mapping and manipulating memory engrams in the mammalian brain, and discuss their implications for understanding both healthy and disordered memory.

Monika Schönauer

Material-specific memory engrams emerge rapidly in human neocortex

New memories are initially labile and have to be consolidated into stable long-term representations. Current theories assume that this is supported by a shift in the neural substrate that supports the memory, away from rapidly plastic hippocampal networks towards more stable representations in the neocortex. Rehearsal, i.e. repeated activation of the neural circuits that store a memory, is thought to crucially contribute to the formation of neocortical long-term memory representations. We investigate memory consolidation in the human brain by non-invasive in-vivo imaging of functional brain activity (fMRI, EEG) and microstructural plasticity (DW-MRI). During the lectures, we will demonstrate that active rehearsal of learning material during wakefulness can facilitate rapid systems consolidation, leading to an immediate formation of lasting memory engrams in the neocortex, even when participants study complex episodic narratives. These representations, observed in both brain activity and brain microstructure, satisfy general mnemonic criteria: They are long-term stable, drive behavior, and code the specific content of what has been learnt.

Laura Colgin

Coordinated place cell sequences and learning and memory processes

The hippocampus is a crucial brain network for learning and memory. Place cells are neurons within the hippocampus that fire in specific spatial locations known as place fields. This spatially selective firing is thought to encode the “where” component of episodic memory. During active behaviors, theta rhythms temporally coordinate firing of sequences of hippocampal place cells that represent successively visited locations. During subsequent periods of awake rest and sleep, these place cell sequences are replayed on a faster time scale during sharp wave-ripples. This presentation will discuss research studies linking coordinated place cell sequences to learning and memory processes.

Michael Shadlen

Neurobiology of decision making

A common framework, termed bounded evidence accumulation or bounded drift-diffusion, accounts for the speed, accuracy, and confidence of many perceptual decisions. However, until now, the drift-diffusion signal has eluded direct observation. Recent advances in high-density neural recording allow us to measure the stochastic drift-diffusion signal giving rise to a single decision. The technology also enables simultaneous recording from populations of functionally related neurons in the parietal cortex and the superior colliculus (SC)—two strongly interconnected nodes of the decision macro-making circuit.
I will contrast the drift-diffusion dynamics in the lateral intraparietal area with single-trial dynamics of neurons of the SC. The burst-like dynamics of SC neurons suggest they are not involved in distributed computation but instead play a distinct role in ending the decision process. We confirm this hypothesis by focal inactivation of the SC while recording in area LIP. The results show the potential of high-density neural recordings, but I will also express concerns about potential pitfalls, particularly the misleading appeal of high-dimensional neural representation. I will also offer perspectives on neural coding and cross-species research involving primates and mice, which introduce new ideas about the balance of processing and controlling operations in the brain.

Priyanka Rao-Ruiz

Molecular and structural dissection of synaptic engrams

Contextual memories are sparsely encoded by memory engram cells and are particularly salient when associated with a negative valence. While enhanced synaptic connectivity onto hippocampal engram cells is crucial for recent memory storage and recall, a time-dependent maturation of cortical engram cells and a strengthening of their synapses drive remote memory expression. However, the specific structural and molecular signatures defining these engram cell synaptic subpopulations, and how they interact, remain unclear. My lectures will focus on our eDorts to bridge this gap in our understanding of the neuroarchitecture of memory. I will present recent work characterizing how engram and putative non-engram synapses adapt at both structural and molecular levels following contextual fear learning. Additionally, I will highlight technological innovations that allow us to isolate engram-specific signatures from bulk tissue and correlate molecular and morphological features at the single-spine level. Finally, I will outline future directions aimed at manipulating these features in vivo to test their causal role in memory. Together, these insights bring us a step closer to understanding the synaptic connectivity patterns that support memory storage within engram networks.