March 25–28, 2023

CNS 2023 | Invited-Symposium Sessions






1 Electrophysiological Studies of Human Memory Retrieval Sunday, March 26, 10:00 AM - 12:00 PM (PT) Grand Ballroom A
2 Learning and Generalization in Humans and Machines Sunday, March 26, 10:00 AM - 12:00 PM (PT) Grand Ballroom B/C
3 Fulfilling the Promise of Inhibitory Control: Bridging the Gap Between Motor and Cognitive Inhibition Tuesday, March 28, 10:00 AM - 12:00 PM (PT) Grand Ballroom A
4 Paths to Increased Brain-Behavior Reproducibility Tuesday, March 28, 10:00 AM - 12:00 PM (PT) Grand Ballroom B/C


Invited Symposium Session 1

Electrophysiological Studies of Human Memory Retrieval

Sunday, March 26, 2023, 10:00AM - 12:00PM (PT), Grand Ballroom A

Chair: Nanthia Suthana, UCLA
Speakers: Neal Morton , Nicholas Turk-Browne, Nanthia Suthana, Bradley Lega

Memory retrieval is an essential cognitive function that allows for individuals to access information stored in the past, make decisions, and anticipate future events. By using rare intracranial electrophysiological recordings from patients with implanted electrodes, researchers have been able to investigate the single neuron and oscillatory phenomena that underlie memory processes. This symposium aims to explore data obtained from these clinical opportunities to better understand the complexity of memory, including how it interacts with decision making, episodic boundaries, and spatial and temporal context.

TALK 1: Hippocampal-Prefrontal Representations Differentiate Outcomes that Vary by Context

Neal Morton, University of Texas

Decision-making in everyday life depends on the context in which choices are made. For instance, answering the phone in your own home is socially appropriate. However, answering your co-worker’s phone while visiting their house might be considered less conventional. To support successful decision making, an efficient memory system must thus form representations that code when stimulus-action-outcome relationships vary by context. Here, we leverage both high-resolution fMRI and intracranial electrophysiology across two studies to quantify the emergence of context-dependent coding in hippocampus and prefrontal cortex. In both studies, participants learned about objects with context-dependent reward values in a naturalistic, virtual environment consisting of an elongated, contextually-varying hallway with decision points on either end. During learning, participants learned how the reward value of objects varied as a function of the spatial context during the hallway period.  Intracranial electrophysiological data reveal that context dependent memory codes emerge in hippocampus and orbitofrontal cortex during learning. Specifically, we find that theta-gamma phase-amplitude coupling differentiates object-reward relationships by context, with such differentiation emerging from early to late learning. Furthermore, we find that later in learning, such codes are predictively reactivated in the hallway in anticipate of choice. Our fMRI data further reveal that context-dependent memory codes formed in hippocampus and orbitofrontal cortex support rapid generalization of knowledge to new object-reward relationships. Collectively, these findings show how memory representations formed in hippocampus and orbitofrontal cortex support adaptive decision making, promoting selection of the right action in any given context.

TALK 2: Memory Prediction and Reactivation in Human Electrophysiology

Nicholas Turk-Browne, Yale University

Memory retrieval allows us to both reminisce about the past and to anticipate the future. In this talk, I will present two recent studies from our lab that illustrate these functions of retrieval, respectively. Each relies on a unique strength of human electrophysiology in epilepsy patients. The first study exploits the temporal precision and broad coverage of hospital-based intracranial EEG from acute implants to explore interactions between statistical learning and episodic memory. We show that learned predictions, reflected in multivariate evidence for the category of an upcoming stimulus immediately before its appearance, interferes with episodic encoding of the current stimulus that triggered the prediction. The second study exploits the portability and flexibility of ambulatory intracranial EEG from chronic implants to explore spatial navigation and memory reactivation. We show that the hippocampus represents various features of real-world navigation along a linear track and that these features can be associated with sounds and later reinstated during rest. These studies reveal the richness and complexity of memory representations and the fundamentally interactive nature of memory systems.

TALK 3: Dynamic Neurophysiological Representations of Memory During Real-World Navigation in Humans

Nanthia Suthana, UCLA

The ability to recall memories of personal experiences is critical for everyday behavior and requires retrieval of the associated spatial context in which the memories are formed. Based on data from freely-moving animals, theta activity in the hippocampus is thought to be important for spatial navigation and memory. However, free recall studies in humans show mixed findings with regards to the specific role of theta oscillations in memory retrieval. This talk will present intracranial electroencephalographic (iEEG) data recorded from human participants navigating real and virtual environments while they recalled specific memories. Altogether, results demonstrate how human hippocampal theta oscillations are dynamically modulated by successful memory, saccadic eye movements, and spatial position depending on momentary task goals during freely-moving spatial navigation.

TALK 4: Schema, Drift, and Episodic Boundaries: A New Look at Primacy Effects in Free Recall

Bradley Lega, University of Texas Southwestern Medical Center

Recent evidence from human intracranial recordings have identified boundary—related activity in the MTL and neocortex, both during spatial navigation and verbal episodic processing.  Single unit recordings specifically include examples of boundary—type cells in the hippocampus.   The role of boundary representations in episodic construction theory supports a possible connection with classical properties of episodic memory behavior such as primacy and recency effects.  We test how boundary information may interact with key components of temporal episodic information, namely representational drift.  Using a dataset of 108 subjects performing free recall during the acquisition of intracranial recordings, we model both temporal drift and item—level reinstatement across encoding lists, by fitting general linear models.  This revealed differences in temporal drift (slope of the autocorrelation signal) for primacy, middle list, and recency items, consistent with boundary—like representations that "anchor" temporal context.  In posterior temporal regions, we also find item—level reinstatement that we interpret as consistent with "schema like" representations, in which individual serial positions exhibit heightened similarity across lists.  This finding lies in contrast to the drifting changes predicted by the temporal context model.  We find no evidence of phase coding for serial position.  We consider how these observations may suggest potential integration of complementary models of episodic processing.

Invited Symposium Session 2

Learning and Generalization in Humans and Machines

Sunday, March 26, 2023, 10:00AM - 12:00PM (PT), Grand Ballroom B/C

Chair: Anna Schapiro, University of Pennsylvania
Speakers: James Whittington, Judy Fan, Ellie Pavlick, Dan Yamins 

Humans have remarkable capacities for learning and generalization across environments and timescales. They can learn rapidly in novel environments, integrate information gracefully over long periods of time, and generalize acquired knowledge flexibly across disparate contexts. Imbuing machines with these crucial capabilities is of great recent interest both for artificial intelligence applications and for building neural simulations as tools to help us understand the mind and brain. This symposium will explore learning and generalization processes across the domains of memory, language, action, and perception. The speakers will examine what kinds of learning humans are capable of, what current machine learning systems are capable of, and what the matches and mismatches can tell us about mechanisms of learning and representation.

TALK 1: How to Organise Knowledge for Flexible Behaviour

James Whittington, Stanford University & University of Oxford

Animals behave flexibly, seamlessly generalising knowledge between apparently different scenarios. This is the hallmark of intelligence. To do this, representations and computations in the brain must also be flexible and generalise. Here we describe several pieces of work on understanding the representations of tasks that can be decomposed into separate building blocks. First, we describe a hippocampal model for learning building blocks and generalising them to novel situations. This model accounts for numerous cell types in hippocampus and entorhinal cortex, and can be related to transformer neural networks. Second, we describe a theoretical result that says different building blocks should be represented by different neural populations, and accounts for novel data such as grid cells warping. Last, we investigate how individual building blocks should be represented and in doing so provide a normative theory of entorhinal grid cells.

TALK 2: Discovering Abstractions that Bridge Perception, Action, and Communication

Judy Fan, University of California San Diego & Stanford University 

Humans display a remarkable capacity for discovering useful abstractions to make sense of and interact with the world. In particular, many of these abstractions are portable across behavioral domains, manifesting in what people see, do, and talk about. For example, people can visually decompose objects into parts; these parts can be rearranged to create new objects; the procedures for doing so can be encoded in language. What principles explain why some abstractions are favored by humans more than others, and what would it take for machines to emulate human-like learning of such "bridging" abstractions? In the first part of this talk, I'll discuss a line of work investigating how people learn to communicate about shared procedural abstractions during collaborative physical assembly, which we formalize by combining a model of linguistic convention formation with a mechanism for inferring recurrent subroutines within the motor programs used to build various objects. In the second part, I'll share new insights gained from extending this approach to understand why the kinds of abstractions that people learn and use varies between contexts. I will close by suggesting that embracing the study of such multimodal, naturalistic behaviors in humans at scale may also shed light on the mechanisms needed to support fast, flexible learning and generalization in machines.

TALK 3: Symbols and Compositionality in Large Artificial Neural Networks

Ellie Pavlick, Brown University

Recent advances in artificial intelligence have produced neural network models that achieve impressive results on tasks involving language and vision, and which often appear to exhibit human-like behaviors such as compositionality and productivity. This talk will probe the extent to which such models have representations that could arguably be viewed as "symbolic" or "compositional", and will discuss how the underlying representations contribute to the models ability (or lack thereof) to generalize outside of their training distributions.

TALK 4: How Well Do Unsupervised Learning Algorithms Explain Actual Human Learning?

Dan Yamins, Stanford University

Humans learn from visual inputs at multiple timescales, both rapidly and flexibly acquiring visual knowledge over short periods, and robustly accumulating online learning progress over longer periods. Modeling these powerful learning capabilities is an important problem for computational visual cognitive science, and models that could replicate them would be of substantial utility in real-world computer vision settings. I'll discuss recent work we've done to establish benchmarks for both real-time and life-long continual visual learning. Our real-time learning benchmark measures a model's ability to match the rapid visual behavior changes of real humans over the course of minutes and hours, given a stream of visual inputs. Our life-long learning benchmark evaluates the performance of models in a purely online learning curriculum obtained directly from child visual experience over the course of years of development. We evaluate a spectrum of recent deep self-supervised visual learning algorithms on both benchmarks, finding that none of them perfectly match human performance, though some algorithms perform substantially better than others. We present analysis indicating that the better algorithms succeed primarily due to their ability to handle sparse low-diversity datastreams that naturally arise in the real world, and that actively leveraging memory through negative sampling appears useful for facilitating learning in such low-diversity environments.


Invited Symposium Session 3

Fulfilling the Promise of Inhibitory Control: Bridging the Gap Between Motor and Cognitive Inhibition

Tuesday, March 28, 2023, 10:00AM - 12:00PM (PT), Grand Ballroom A

Chair: Jan R. Wessel, University of Iowa
Speakers: Jan R. Wessel, Dace Apšvalka, Earl K. Miller, Ryan J. Hubbard & Lili Sahakyan

Inhibitory control is one of the core mechanisms through which the brain enables flexible, goal-directed behavior, and features prominently in many highly influential psychological theories. In those theories, inhibitory control purportedly regulates a wide range of lower-order processes, ranging from urges and emotions to memory and language. However, there is a substantial gap between these psychological theories on the one hand and cognitive neuroscience of inhibitory control on the other hand. The latter field has, for the last three decades, focused heavily on the inhibitory control of movement – for example, during the stopping of actions. This substantial body of work has outlined a cortico-subcortical circuit underlying motoric inhibition and its purported neural signatures – predominantly, beta-band activity in the local field potential. However, disproportionately little research has focused on whether this neural mechanism actually inhibits non-motor activity – including the processes central to the above-mentioned psychological theories. To fill this gap, this symposium will highlight recent converging evidence from fMRI, EEG, deep-brain stimulation, and intracranial studies, all suggesting that the same neural mechanisms involved in inhibiting action may indeed serve to inhibit non-motor activity. As such, this symposium should be of interest to cognitive neuroscientists interested in executive functions, psychologists interested in the neural mechanisms that regulate thoughts and actions, and clinical scientists working on disorders ranging from ADHD and PTSD to gambling disorder and Parkinson’s disease.

TALK 1: The Universal Role of Inhibitory Control in Flexible Behavior and Cognition

Jan R. Wessel, University of Iowa

Inhibitory control is one of the fundamental mechanisms by which the brain regulates behavior. Cognitive neuroscience work on inhibitory control has outlined the likely role of a cortico-subcortical mechanism involving frontal cortex and the subthalamic nucleus in the inhibition of movement. Most of this work stems from paradigms like the stop-signal task, in which prepotent actions have to be suddenly inhibited following an explicit signal. In this talk, I will outline work that crucially expands this picture in two ways. First, I will briefly describe a series of studies showing that fronto-subthalamic inhibitory control mechanisms are activated by a multitude of control-demanding stimuli – including many that do not explicitly require inhibitory control. Then, I will describe recent work suggesting that this same mechanism may be involved in the inhibitory regulation of active non-motoric representations, including active task set representations held in working memory. In doing so, I hope to promote a wider view of the fronto-subthalamic inhibitory control circuit as a ubiquitous mechanism for cognitive and motoric flexibility – one that enables goal-directed behavior in a wide array of scenarios that necessitate rapid adjustments to both ongoing thoughts and actions.

TALK 2: Top-Down Control by Beta Rhythms

Earl K. Miller, Massachusetts Institute of Technology

Working memory is the sketchpad of consciousness, the fundamental mechanism the brain uses to gain flexible, volitional control over its thoughts and actions. For the past 50 years, working memory has been thought to rely on cortical neurons that fire continuous impulses that keep thoughts “online”. However, new work from our lab has revealed more complex dynamics. The impulses fire sparsely and interact with brain rhythms of different frequencies. Higher frequency gamma (> 35 Hz) rhythms help carry the contents of working memory while lower frequency alpha/beta (~8-30 Hz) rhythms act as control signals that gate access to and clear out working memory. In other words, a rhythmic dance between brain rhythms may underlie your ability to control your own thoughts.

TALK 3: Recruitment of Domain-General Inhibitory Control Supports Suppression of Encoding of Episodic Memories

Ryan J. Hubbard & Lili Sahakyan, University of Illinois Urbana-Champaign

Not every experience is something we want to remember, and the ability to stop encoding of information into memory can be adaptive. Research using the directed forgetting method suggests that humans can strategically forget information, but the mechanisms underlying this process remain debated. The selective rehearsal account claims that forgetting is a passive process, whereas cognitive neuroscientific work indicates active engagement of the prefrontal cortex following cues to forget, suggesting recruitment of inhibitory processes to suppress encoding. In this talk, I will present work that directly tested the role of inhibition in encoding suppression with a cross-task design, relating the behavioral and EEG data from participants completing a Stop Signal task – a task specifically testing inhibitory processing abilities – to a novel item-method directed forgetting task with both encoding suppression (Forget) and thought substitution (Imagine) cues. Behaviorally, Forget and Imagine cues produced similar rates of forgetting, but through separable neural processes, with Forget cues eliciting frontal oscillatory power changes that were predictive of future forgetting. Importantly, Stop Signal task performance (SSRTs) was correlated with successful forgetting, and brain-behavior analysis demonstrated that engagement of right-frontal beta activity following motoric stopping was related to successful forgetting. Finally, classifiers trained on neural signals discriminating successful and unsuccessful motoric stopping could also classify successful and unsuccessful forgetting following Forget cues. Finally, I will speculate on what the inhibitory system is “acting upon” to suppress encoding - namely, the binding of item information to the ongoing context, rather than the representation of the item itself.

TALK 4: A Common Brain Mechanism for Stopping Unwanted Actions and Memories

Dace Apšvalka, MRC Cognition and Brain Sciences Unit, University of Cambridge

Stopping unwanted memories is essential to mental health and well-being, and so is stopping unwanted physical actions. Could there be a common mechanism that allows us to stop both unwanted memories and actions? The research on memory stopping and action stopping has primarily been independent of each other. There is a general agreement that the right prefrontal cortex (PFC) is critically involved in inhibitory control. Memory control research has highlighted the importance of the right dorsolateral PFC controlling the hippocampus. The action control research has highlighted the importance of the right ventrolateral PFC controlling the primary motor cortex (M1). Are the two PFC regions indeed controlling only a specific domain, or could they be involved in controlling both memories and actions? We integrated the separate bodies of research and examined the potential domain-general mechanisms of inhibitory control of memories and actions in the same individuals. Moreover, we investigated how a common mechanism could govern memory and action stopping if they are mediated by distinct neural systems (hippocampus and M1, respectively).

Invited Symposium Session 4

Paths to Increased Brain-Behavior Reproducibility

Tuesday, March 28, 2023, 10:00AM - 12:00PM (PT), Grand Ballroom B/C

Chair: Nico Dosenbach, Washington University in St. Louis
Speakers: Scott Marek, Stephanie Noble, Thomas Yeo, Russ Poldrack

Linking brain MRI metrics to human behavioral variables has been a main endeavor of cognitive neuroscience. MRI has greatly elevated our understanding of the human brain through many well-replicated studies mapping abilities to specific structures and functions. Just as for psychology, genomics and medicine, concerns have been raised about the reproducibility of some brain-behavior associations, due to methodological variability, data mining for significant results, overfitting, confirmation and publication biases and inadequate statistical power. This symposium will first highlight the critically important differences across brain-behavior association types (i.e., task fMRI vs. Brain-Wide Association Studies [BWAS]). Next, the presenters will discuss approaches and methods for improving the reproducibility of correlations between task fMRI and resting-state functional connectivity (RSFC) metrics and behaviors.

TALK 1: Brain-Wide Association Studies: A Year in Review

Scott Marek, Department of Radiology, Washington University of School of Medicine

Recent work from our group called for sample sizes into the thousands for reproducibility in brain-wide association studies (BWAS). In this talk, I will discuss the mechanism by which small samples have undermined BWAS and the need for large samples for reproducibility and generalizability. I will provide further clarification on the critical differences between BWAS and non-BWAS MRI, with a focus on leveraging the strengths of each approach to inform one another. We suggest the choice in study design (BWAS vs. non-BWAS) should align with a priori defined study goals to propel functional MRI towards practical and clinical utility. As an early career neuroimaging researcher with relatively fewer monetary resources than senior scientists, I will conclude my talk with a practical discussion on how I envision balancing relatively limited resources with the desire to perform robust and reproducible research.

TALK 2: Considerations for Improving Measurement Reliability and Validity in fMRI

Stephanie Noble, Department of Radiology and Biomedical Imaging, Yale University

Measurement reliability (i.e., stability of a measure) and validity (i.e., innate correspondence of a measure with a target) are desirable and complementary facets for understanding the practical utility of a measure. The reliability and validity of fMRI is just beginning to be fully empirically characterized with the advent of larger-than-ever datasets and efficient computational methods. However, recent reports have pointed to limitations in both reliability and validity for typical fMRI studies, with particularly heightened attention to low test-retest reliability and power in the past several years. Reflecting on my recent work, I will discuss practical choices fMRI researchers can make to improve test-retest reliability and power. In particular, I will highlight how researchers can improve both simultaneously by 1) moving beyond the level of individual voxels and edges to large-scale and multivariate analytical methods, and 2) using bigger and deeper data. Finally, I will highlight the importance of making choices that improve reliability so long as they do not diminish validity.

TALK 3: Insights From Large-Scale Datasets for Optimizing Study Design and Boosting Prediction Accuracy

Thomas Yeo, Centre for Sleep & Cognition, National University of Singapore

Studies have shown that more participants and greater scan time improve prediction of behavioral traits from resting-state functional connectivity. Here we ask the following question: given a fixed scan budget, should we scan more participants (for shorter duration), or less participants (for longer duration) to maximize behavioral prediction accuracy? Surprisingly, we find that total scan duration (number of participants * scan time per participant) explains prediction accuracy of cognitive performance remarkably well (with some caveats). This relationship generalizes across cognitive scores and datasets, and might be potentially useful for future study design. In a second line of work, we propose a meta-matching framework to translate predictive models trained from a large dataset to predict new phenotypes in a small dataset. We demonstrate that meta-matching can greatly boost the prediction of new phenotypes in small independent datasets in many (although not all) scenarios.

TALK 4: Improving Robustness and iInterpretability in Brain-Behavior Modeling

Russ Poldrack, Department of Psychology, Stanford University

When we identify a relationship across individuals between task fMRI activation and a behavioral measure or group variable, we generally wish to interpret this as reflecting differences between the neural computations performed across those individuals. However, this interpretation suffers from a potential confound: Whenever response times differ across conditions, it is possible that differential activity could simply reflect differences in the duration of neural engagement, rather than differences in the kind of neural processing being engaged. The relationship between time on task and activation has been acknowledged for more than a decade, yet most task fMRI studies still do not address this potential confound. I will outline an approach for the effective modeling of response times in task fMRI data, showing that it can provide much more interpretable maps of task activation by differentiating task-specific effects from more general response time effects. I will also outline how response time differences between subjects can leak into brain-behavior correlation analyses at the group level, endangering the interpretability of observed brain-behavior correlations.






MARCH 25–28

Latest from Twitter