CNS 2021 | Invited-Symposium Sessions

 

#

TITLE

DATE & TIME

LOCATION

1 HOW PRIOR KNOWLEDGE SHAPES ENCODING OF NEW MEMORIES Sunday, March 14, 10:30AM - 12:30 PM (ET) Axon Room
2 IMPLICATIONS OF ANATOMICAL BRAIN NETWORK FEATURES FOR COGNITION Sunday, March 14, 10:30AM - 12:30 PM (ET) Dendrite Room
3 NEURAL NETWORKS IN COGNITIVE NEUROSCIENCE Tuesday, March 16, 10:30AM - 12:30 PM (ET) Axon Room
4 SYMPOSIUM IN HONOR OF ART SHIMAMURA Tuesday, March 16, 10:30AM - 12:30 PM (ET) Dendrite Room

Invited Symposium Session 1

HOW PRIOR KNOWLEDGE SHAPES ENCODING OF NEW MEMORIES

Sunday March 14, 2021, 10:30AM - 12:30PM (ET), Axon Room

Co-Chaired by Rik Henson and Andrea Greve, University of Cambridge

Speakers: Morris Moscovitch, Alison R. Preston, Qihong Lu, Andrea Greve

New memories are not written on a tabula rasa: rather, they are shaped by what we already know. In this symposium, the speakers will discuss how prior knowledge (such as a schema or situation model) influences the encoding of new episodic memories, drawing on a range of behavioral, neuroimaging and computational evidence.

TALK 1: THE INFLUENCE OF NEURAL CONTEXT AND REINSTATEMENT OF PRIOR KNOWLEDGE (SCHEMAS AND SEMANTIC CATEGORIES) ON ENCODING

Morris Moscovitch and Asaf Gilboa, University of Toronto

How does reinstatement of prior knowledge, in the form of schemas or semantic categories, influence encoding and subsequent memory? In the first set of studies, we used eye-tracking to examine how reinstating schemas of familiar locations (e.g., kitchen) influenced search and memory for target objects in younger and older adults. We found that schema-congruent targets were detected more quickly than incongruent targets, and this effect was enhanced in older adults. Eye-tracking revealed that these effects emerged early in search, and were stronger in older adults, suggesting that schema biasing of online processing increases with age. Both groups showed enhanced memory for the location, but not for identity, of congruent targets, with older adults showing over-reliance on schemas. In a second set of studies, participants reinstated (called to mind) event-based schemas and semantic categories prior to judging whether words belonged to one or the other. Subsequent memory was better for words related to schemas than to categories. We also recorded EEG from patients with ventro-medial prefrontal cortex (vmPFC) damage and matched controls during the task. While activity patterns overlapped considerably, reinstatement was most associated with pre-stimulus low-frequency desynchronization between vmPFC-posterior parietal for schemas, and between lateral temporal lobe -inferotemporal cortex for categories. Patients showed reduced desynchrony in both conditions, but similar connectivity patterns. Behaviourally, patients, particularly those with damage to the subcallosal vmPFC. showed deficits specific to schematic processing. We conclude that damage to vmPFC influences processing of both schemas and categories, but the underlying network-level mechanisms of this disruption differ.

TALK 2: HIPPOCAMPAL-MEDIAL PREFRONTAL INTERACTIONS GUIDE HOW EXISTING MEMORIES BIAS NEW LEARNING

Alison R. Preston, The University of Texas at Austin

Prominent theories posit that cognitive maps are formed by integrating knowledge across distinct experiences. An important advantage of cognitive maps is their flexibility; information learned in one context can be generalized to new situations, biasing how new content is processed. Here, we provide a representational account of this flexibility by quantifying how cognitive maps of space bias decision making in non-spatial tasks. Participants learned the locations of novel objects in distinct virtual environments during active navigation. After learning, we found that hippocampus and medial prefrontal cortex (mPFC) representations became both more similar for objects that shared an environment and more separable for objects experienced in different environments. We then tested how hippocampus-mPFC cognitive maps influenced behavior in a non-spatial task. Twenty-four hours after spatial learning, participants were scanned while performing an incidental sequence task using the same objects from spatial learning. Unbeknownst to participants, objects were presented in sequential triplets comprising three objects from either the same or different environments. We found that individuals with more coherent cognitive maps were slower to make object preference judgments at triplet boundaries. Moreover, we found that neural response at triplet boundaries was altered in several ways. Reinstatement of the virtual contexts associated with objects decreased at boundaries along with hippocampal-mPFC connectivity. At the same time, mPFC responses increased after a boundary, reflecting increased uncertainty about the next potential item in a sequence. Together, these finding reveal how hippocampal-mPFC cognitive maps drive neural prediction and behavior in non-spatial contexts, a hallmark of their flexibility.

TALK 3: MODELING WHEN EPISODIC ENCODING SHOULD TAKE PLACE TO SUPPORT EVENT PREDICTION

Qihong Lu, Princeton University

When should episodic encoding take place to support event understanding? Traditional, well-controlled memory experiments usually provide explicit instruction about when to encode, but it is less obvious when to do this in naturalistic settings. To address this question, we trained a memory-augmented neural network to predict upcoming events in an environment where situations (sets of parameters governing transitions between events) sometimes reoccurred; the model was allowed to learn a policy for when to retrieve episodic memories in the service of event prediction. With regard to encoding policy, we found that model variants that selectively store episodic memories at the end of an event (but not mid-event) show better subsequent event prediction performance and less incorrect recall, because encoding at event boundaries produces a more complete representation of the situation that is less confusable with other situations. This modeling result provides a normative explanation for recent human neuroimaging findings showing that the hippocampus is preferentially engaged at event boundaries during naturalistic movie viewing. We also show how this policy allows the model to integrate across multiple, discontinuous segments of a narrative to build an increasingly complete situation representation.

TALK 4: SCHEMA AND PREDICTION ERROR IN EPISODIC MEMORY ENCODING

Andrea Greve, University of Cambridge, UK

Episodic memories are rarely created in isolation but often encoded in relation to one’s contextual expectations that originate from our prior knowledge, i.e. schemas. It is widely established that episodic memory is facilitated when novel information is congruent with our schema. Recent work, however, also proposes superior memory for novel information that conflicts with pre-existing schema, i.e. causing a prediction error. In this talk, we focus on the role of schema expectancy and violations in episodic learning. We outline a theoretical framework, ‘Schema-Linked Interactions between Medial prefrontal and Medial temporal region’ (SLIMM), which predicts that episodic memory is a “U-shaped” function of expectancy, and that the two ends of this U-shape are functionally dissociable, since they depend on different neural systems. We provide support for these predictions using both simple schemas (rules) learned during an experiment, and using pre-experimental schemas for more naturalistic stimuli within an immersive virtual reality environment.

Invited Symposium Session 2

IMPLICATIONS OF ANATOMICAL BRAIN NETWORK FEATURES FOR COGNITION

Sunday March 14, 2021, 10:30AM - 12:30PM (ET), Dendrite Room

Chair by Danielle Bassett, University of Pennsylvania & Santa Fe Institute

Speakers: Masud Husain, Sofie Valk, Maria Giulia Preti, Danielle S. Bassett

In this symposium, we will canvas a growing line of scientific inquiry that seeks to understand how anatomical brain networks account for cognitive processes in humans. Such networks consist of white matter tracts whose patterning is estimated from diffusion imaging data using state-of-the-art tractography techniques. Collectively, the work demonstrates that the architecture of white matter networks provides important explanations for higher-order cognitive processes in the domains of social cognition, complex reasoning, and executive function, as well as for motivational and cognitive dysfunction.

TALK 1: BRAIN NETWORKS IN MOTIVATIONAL AND COGNITIVE DYSFUNCTION

Dr. Masud Husain, Oxford University

How do brain networks underpin human behaviour? An important way to investigate this issue has emerged through the investigation of white matter connections between brain regions. Evidence in both healthy people and individuals with diseases that affect the white matter have begun to reveal how changes in structural network integrity might relate to changes in cognitive function and motivated behaviour. In a UK Biobank sample of over 22,000 people, both increasing age and cardiovascular risk factors such as raised blood pressure, were significantly correlated with altered frontoparietal white matter integrity, which in turn was associated with significant reductions in executive function. The results further show that raised systolic blood pressure might have a direct impact on cognitive function across the sample population, aged 44-73. Cardiovascular risk factors can in the long term lead to a common condition known as small vessel cerebrovascular disease (SVD) which impacts on white matter structure, traditionally indexed by white matter hyperintensities on MR imaging. SVD is associated with cognitive impairment and motivational deficits manifest in the syndrome of apathy. Study of such patients has revealed that alterations in the integrity of specific white matter tracts linking medial frontal regions and the basal ganglia to these regions is associated with apathy. These examples show how alterations in white matter integrity can lead to specific cognitive and motivational deficits. They also point to how imaging of brain network connections might contribute to monitoring brain health or tracking the impact of interventions to prevent its decline.

TALK 2: SHAPING BRAIN STRUCTURE, HOW EVOLUTION AND COGNITION ORGANIZE THE BRAIN

Dr. Sofie Valk, Max Planck Institute for Human Cognitive and Brain Sciences

Current global crises, leading to increased social disconnection and the need for social cohesion, highlights the importance of social and cognitive skills to enhance human cooperation on a community level, as well as to overcome loneliness and social isolation at the level of the individual. The topology of the cerebral cortex has been proposed to be an important prerequisite for human social interaction and cognition. Here I will present two novel studies assessing how evolution, genes and social behavior shape large-scale brain organization in healthy adults. In the first study we assessed the genetic basis of large-scale cortical organisation of thickness covariance using twin-models (Human Connectome Project) as well as cross-species comparisons (PRIME-DE) in combination with non-linear dimension reduction techniques. We found two organizational patterns traversing posterior-to-anterior and inferior-to-superior axes in both humans and macaques reflecting functional and evolutionary patterning. In a second study we evaluated the structural and functional plasticity of large-scale brain organization following 3-months of training (i) mindfulness-based attention and interoception, (ii) socio-affective skills, and (iii) socio-cognitive skills respectively (ReSource study). Contrasting the effect of each training module, we found diverging patterns of functional network reorganization. Training of attention-mindfulness resulted in segregation of cortical networks, whereas socio-cognitive training resulted in their integration. Socio-affective training had stabilizing effects on functional organization. Together these studies how evolution, genes, and social skills shape the cerebral cortex.

TALK 3: EXPLORING BRAIN FUNCTION-STRUCTURE COUPLING DURING RESTING-STATE AND TASKS

Dr. Maria Giulia Preti, University of Geneva

How brain functional activity is shaped by the underlying brain structure is a complex question still lacking a clear answer. Advanced and non-invasive MRI techniques –such as functional and diffusion MRI– provides us with unique information about both functional and structural connectivity, respectively. However, the degree to which brain structure limits brain function is hard to quantify. In this context, we introduce a structural-decoupling index, defined at each brain area, as a novel measure quantifying the structure-function relationship; i.e., indicating the degree to which the functional signal detaches from the anatomical backbone underneath. The strength of function-structure coupling shows to spatially vary throughout the brain with a specific gradient going from areas related to lower-level functions (sensory, motor) to regions corresponding to higher-level ones (e.g., memory, reward, emotions). In particular, the activity in primary sensory regions appears more strongly coupled with brain structure, while higher-level regions, such as portions of the executive control network, the amygdala and language areas, show a functional activity more independent from the structure. The existence of this macroscale gradient of function-structure coupling, showed here for the first time, appears in line with evidence from other modalities, reporting a similar hierarchy for functional connectivity, temporal hierarchy and microstructural properties. Further, the structural-decoupling index appears to characterize well both different functional tasks and different individuals.

TALK 4: ANATOMICAL NETWORK CONSTRAINTS UPON (AND SUPPORT FOR) COGNITIVE CONTROL

Dr. Danielle S. Bassett, J. Peter Skirkanich Professor, University of Pennsylvania & External Professor, Santa Fe Institute

The human brain is a complex organ characterized by heterogeneous patterns of interconnections. Non-invasive imaging techniques now allow for these patterns to be carefully and comprehensively mapped in individual humans, paving the way for a better understanding of how wiring supports cognitive processes. While a large body of work now focuses on descriptive statistics to characterize these wiring patterns, a critical open question lies in how the organization of these networks constrains the potential repertoire of brain dynamics.  In this talk, I will describe an approach for understanding how perturbations to brain dynamics propagate through complex wiring patterns, driving the brain into new states of activity. Drawing on a range of disciplinary tools – from graph theory to network control theory and optimization – I will identify control points in brain networks and characterize trajectories of brain activity states following perturbation to those points. Finally, I will describe how these computational tools and approaches can be used to better understand the brain's intrinsic control mechanisms and their alterations in psychiatric conditions.

Invited Symposium Session 3

NEURAL NETWORKS IN COGNITIVE NEUROSCIENCE

Tuesday March 16, 2021, 10:30AM - 12:30PM (ET), Axon Room

Co-Chaired by Christopher Summerfield, University of Oxford & Google DeepMind & Grace Lindsay, University College London

Speakers: Grace Lindsay, Robert Yang, Talia Konkle and Chris Summerfield

Deep learning models have powered recent progress in AI research, and neuroscientists are increasingly looking (once again) to neural networks as computational theories of perception and cognition. In this symposium, we consider recent work in which neural networks have shed light on the mechanisms that underlie higher cognition, including attention and working memory, task-level control, numerical cognition, and abstract reasoning.

TALK 1: EXPLORING THE TOP-DOWN SIGNALS NEEDED FOR VISUAL ATTENTION

Grace Lindsay, University College London

Countless behavioral experiments have documented the beneficial effect that selective visual attention has on performance of difficult visual tasks. Through concurrent neural recordings, the ways in which attention alters neural activity have been found as well. Can these observed neural changes account for the observed behavioral changes? In this talk I will show how this question can be explored using deep convolutional neural networks. Specifically, I will show how implementing the observed neural changes in this model lead to performance changes that mimic those seen in subjects. I will also show how a different form of neural changes calculated to be optimal for task performance can also create these performance enhancements. I will then propose an experiment to determine if the top-down signals that control attention implement these optimal changes.

TALK 2: HOW TO STUDY COGNITION WITH RECURRENT NEURAL NETWORKS

Robert Yang, Massachusetts Institute of Technology

The use of artificial neural networks, including recurrent neural networks (RNNs), to study cognition has a long history in cognitive science. Advance in deep learning over the past decade has brought forth powerful new tools, making it substantially easier to use RNNs to address neuroscience and cognitive science questions. In comparison to traditional computational models in neuroscience, RNNs can have substantial advantage at explaining complex behavior and complex neural activity patterns. They rapidly generate mechanistic hypotheses of cognitive computation. RNNs further provide a natural way to flexibly combine bottom-up biological knowledge with top-down computational goals into computational models. I will demonstrate the unique utility of RNNs for studying cognition with a number of examples. In particular, I will show that RNNs allow us to study how different cognitive tasks may be instantiated in a single neural circuit. Finally, I will discuss the pitfalls of most existing RNN models, and how we can move forward in the next decade to build better RNN models of brains and mind.

TALK 3: HOW DO NEURAL NETWORKS LEARN OBJECT CATEGORIES?

Talia Konkle, Harvard University

The human brain learns a useful representation of the visual world—supporting capacities like object categorization, face recognition, reading, numerical reasoning, and more.  Using object-trained deep convolutional neural networks, we have found that many visuo-cognitive signatures are emergent in the feature spaces of these object-trained networks, from face-selective brain responses to behavioral judgments of letter identity and numerosity, without requiring specialized representational mechanisms per se.  Further, we have found that even explicit category-level representational pressure is not a necessary prerequisite, as networks trained with a simpler learning objective—discriminating all views of the world from all other views—can learn similar emergent brain-like visual representational formats as category-supervised networks. Taken together, these empirical-computational results provide evidence that support a theory-shift away from object categories and specialized mechanisms towards generic tasks and more primitive feature spaces that can support a surprisingly rich set of visual and cognitive operations.

TALK 4: NEURAL STRUCTURE ALIGNMENT IN HUMANS AND NEURAL NETWORKS

Chris Summerfield, University of Oxford & Google DeepMind

A prerequisite for intelligent behaviour is to understand how stimuli are related and to generalise this knowledge across contexts. Generalisation can be advantageous when relational patterns are shared by physically dissimilar stimuli, but it is unclear how this is achieved at the neural or computational levels. I will describe 2 studies in which humans and neural networks learned to perform comparisons among numbers (study 1) or transitively ordered stimuli (study 2) in a way that benefits from generalising the core concept of “more” or “less” between contexts. Using multivariate analysis of human brain signals from both EEG and fMRI, as well as analysis of neural network hidden unit activity, we observed that both systems developed parallel “magnitude lines” for each context, and that these transitively ordered state spaces were aligned in a way that explicitly facilitated generalisation of relational concepts (more and less). In human BOLD signals, this activity was prominently observed in the parietal cortex and other dorsal stream structures. We also observed that multivariate patterns of human brain activity changed rapidly during “knowledge assembly”, as learning about a single associative link led to a rapid rearrangement of the relational structure.  Together, these studies suggest a computational mechanism by which we encode and generalise relational knowledge about magnitude, expressed in human brain activity.

Invited Symposium Session 4

IN HONOR OF ART SHIMAMURA

Tuesday March 16, 2021, 10:30AM - 12:30PM (ET), Dendrite Room

Introductions by Rich Ivry, UC Berkeley

Speakers: Juliana Baldo, Charan Ranganath, Mike Anderson, Bill Prinzmetal

This symposium will honor the memory of Art Shimamura who died on October 6, 2020. Art was a professor in the Department of Psychology at the University of California, Berkeley, and a founding member of the Cognitive Neuroscience Society. He was a talented scientist and an award-winning teacher, extraordinarily generous, down-to-earth, funny, and creative. Art had an inspirational influence on his students, colleagues, and friends, and his highly cited research had a profound impact on our understanding of memory, cognition, and the brain. As a Guggenheim Fellow, he explored the relationships between art and cognitive neuroscience in his book, Experiencing Art: In the Brain of the Beholder. He also published other books, including Get SMART! Five Steps Toward a Healthy Brain and A Walk Around O’ahu: My Personal Pilgrimage. In this symposium, a representative set of speakers will talk about Art and his work — one representing his influence on undergraduate students, one representing his influence on graduate students, one representing his influence on postdoctoral trainees, and one representing his influence on scientific colleagues.

TALK 1: RENAISSANCE MAN: REMINISCENCES FROM A STUDENT OF ART

Juliana Baldo, UC Davis

When was the last time you watched a movie? What movie did you see? Who was with you? These seemingly simple questions pose quite a difficult task for patients with frontal lobe injury, as well as many of the rest of us, especially as we age. These were the kinds of questions that my Mentor and friend, Arthur Shimamura, liked to ask. He was intrigued by the complexity of human memory and its close relatives, attention and metacognition, and he wanted to furthermore determine which parts of the brain gave rise to these abilities. Art came to believe that portions of lateral prefrontal cortex were critical for these abilities, as evidenced by the wide array of subtle cognitive difficulties observed in patients with focal frontal lesions. Unlike many of us who tend to stay in a single lane studying memory, language, attention, etc., Art was always considering how these different cognitive domains criss-crossed and interacted to produce complex human behaviors. This approach greatly influenced my work, as I collaborated with him on studies looking at the interaction of attention and memory in frontal lobe patients. In those studies, we found that lateral prefrontal cortex plays a role in memory in part due to its role in focusing attention and supporting strategic retrieval of remote information. This cross-domain approach also influenced my subsequent work, in which I have shown that the disruption of language centers in aphasia results in a variety of non-linguistic cognitive deficits, including impaired reasoning and problem solving.

TALK 2: WORKING WITH MEMORY: A JOURNEY THAT STARTED IN THE SHIMLAB

Charan Ranganath, UC Davis

Art Shimamura once wrote, “...no brain region works in isolation. My personal mantra is: it's a whole-brain issue, stupid!”. Although most memory researchers would agree with this principle, in practice, most theories on the neuroscience of episodic memory focus largely or entirely on the hippocampus. Art thought of memory and the brain in a more holistic sense. He studied many topics, but remained consistent in his curiosity about how different cognitive processes come together when we engage with the world. I was an undergraduate in Art’s lab when he did groundbreaking research demonstrating the critical role of the prefrontal cortex in determining what we remember and why we forget. As Art made the transition to neuroimaging, he focused on how regions in the parietal cortex might play a role in putting together, or “binding,” information in memory. Later, he advanced the idea that we can better understand cognition by studying how we perceive and understand movies and works of art. In my talk, I will describe our recent research on memory for naturalistic events, and I will highlight how this research has continued in the directions that began in Art’s lab. I will describe recent efforts to build a computational framework to understand how we remember naturalistic events depicted in films and stories. This research highlights the importance of understanding how distributed cortical networks interact with the hippocampus as we perceive, understand, and remember complex events.

TALK 3: INHIBITORY CONTROL OVER MEMORY BY THE PREFRONTAL CORTEX

Mike Anderson, Cambridge

The ability to overcome internal distraction created by interfering memories and thoughts is a key component in the effective use of our memories.   Art was an early leader in cognitive neuroscience who recognized the pivotal role that the prefrontal cortex plays in mitigating this interference, and in showing that damage to this structure could yield profound difficulties in retrieval.   One mechanism by which the prefrontal cortex might achieve this function is inhibitory control, a hypothetical mechanism by which control processes could down-regulate the activity of representations in posterior cortex that interfered with our ongoing goals.   Inhibitory control was of special interest to Art in part because, as he often amusingly declared, he himself had very little inhibitory control.   When I was a post-doctoral fellow in Art’s laboratory at Berkeley, Art provided inspiration and nurturing of my own interest in inhibitory control over memory, and his wise and warm influence was instrumental in helping me pursue the work on this subject that I continue even to this day, decades later.  In my talk, I will describe work we have done on the role of inhibitory control in regulating memory retrieval illustrating that it is a direct outgrowth of Art’s early ideas. I will describe the work that we have done with multimodal imaging and neuropsychology to demonstrate the causal role of the right dorsolateral prefrontal cortex in suppressing memory retrieval and document the key role that these processes play in regulating both intrusive thoughts and unwanted behaviors.

TALK 4: THE MYSTERY SPOT AND VISUAL ILLUSIONS

Bill Prinzmetal, UC Berkeley

I had the good fortune to work with Art Shimamura on several projects. One project started as a family outing to a roadside attraction called the Mystery Spot (MS). The MS is similar to dozens of attractions consisting of a shack tilted and pitched about 18°, resulting in illusions such as water running uphill and people appearing to stand at gravity-defying angles. These illusions can be described as the misperception of orientation as in the rod-and-frame effect. A phenomenon at the MS that interested us was that comparing two people the same height, the one standing nearest the low side of the roof would appear taller. It occurred to us that the misperception of orientation could cause the misperception of height. At the MS, we measured observers’ perception of horizontal and their misperception of height. Subjects’ misperception of orientation predicted their misperception of height. Art built a small-scale MS in a shoebox with toy soldiers as people. We discovered that the miniature MS stimuli contained the Ponzo illusion: The toy soldiers were like the test lines, and the “shack,” the oblique lines. These findings led us to suspect that many seemingly disparate illusions (e.g., Ponzo, Zöllner, Wündt-Herring) could be related to the misperception of orientation.  I will describe a series of experiments inspired from Art’s miniature MS to compare other accounts of the Ponzo illusion to our orientation-perception theory; and also experiments showing that the Zollner, Poggendorff, Ponzo illusions and tilt-induction effect behaved like the rod-and-frame effect.

 

CNS2022-Logo_FNL4web

APRIL 23–26 • 2022

Latest from Twitter