Large-scale neural recordings contain high-dimensional structure that cannot be easily captured by existing data visualization methods. We therefore developed an embedding algorithm called Rastermap, which captures complex temporal and highly nonlinear relationships between neurons, and provides useful visualizations by assigning each neuron to a location in the embedding space. We applied Rastermap to a variety of datasets, including spontaneous neural activity, neural activity during a virtual reality task, widefield neural imaging data from a 2AFC task, and artificial neural activity from an agent playing atari games. We found within these datasets unique subpopulations of neurons encoding abstract elements of decision-making, the environment and behavioral states. To interrogate behavioral representations in the mouse brain, we developed a fast deep-learning model for tracking 13 distinct points on the mouse face recorded from arbitrary camera angles. The model was just as accurate as state-of-the-art pose estimation tools while being several times faster, making it a powerful tool for closed-loop behavioral experiments. Next, we aligned facial key points across mice in order to train a universal model to predict neural activity from behavior. The universal mouse model could predict neural activity as well as a model fit to a single mouse, showing that neural representations of behaviors are conserved across mice. The latent states extracted from the universal model contained interpretable mouse behaviors.
Paul Cisek (Professor, University of Montréal)
2PM BIOMED BLDG C44 and online: https://bristol-ac-uk.zoom.us/j/92144801092?pwd=YVNKeGFFRW1kb09iQnU0MkRIRzV5UT09
Meeting ID: 921 4480 1092 Passcode: 225286
Neural dynamics of embodied decisions
Decision-making is not a single thing, but rather a variety of processes that have been gradually elaborated and refined over the brain’s long evolutionary history. In the first part of my talk, I will briefly describe that history, emphasizing the kinds of decision processes that dominated animal behavior for hundreds of millions of years. In particular, I will focus on theories of how animals make real-time decisions between action opportunities available in a dynamically changing world. I will then summarize experimental studies suggesting that these kinds of action decisions unfold in parallel across cortical and subcortical circuits, laying the foundations for more advanced aspects of primate behavior.
29th April 2022
Karan Grewal (Research Scientist, Numenta)
4:15 PM BIOMED BLDG C44 and online: https://bristol-ac-uk.zoom.us/j/92144801092?pwd=YVNKeGFFRW1kb09iQnU0MkRIRzV5UT09
Meeting ID: 921 4480 1092 Passcode: 225286
Avoiding Catastrophe: Active Dendrites Enable Multi-Task Learning in Dynamic Environment
A key challenge for AI is to build systems that must adapt to changing task contexts and learn continuously. Although standard deep learning systems achieve state of the art results on static benchmarks, they often struggle in dynamic scenarios where the task changes over time, and this phenomenon is known as catastrophic forgetting. In this talk, I will discuss how biophysical properties of dendrites in the brain and local inhibitory systems enable networks to dynamically restrict and route information in a context-specific manner. Specifically, I will highlight the performance of a deep learning architecture that embodies properties of dendrites on two separate benchmarks requiring task-based adaptation: Meta-World (a multi-task reinforcement learning environment where a robotic agent must learn to solve a variety of manipulation tasks simultaneously) and permutedMNIST (a continual learning benchmark in which the model’s prediction task changes throughout training). Analysis on both benchmarks demonstrates the emergence of overlapping but distinct and sparse subnetworks, allowing the system to fluidly learn multiple tasks with minimal forgetting. This work sheds light on how biological properties of neurons can inform deep learning systems to address dynamic scenarios that are typically impossible for traditional ANNs to solve.
22nd April 2022
H Freyja Ólafsdóttir (Assistant Professor of Neurophysiology, Donders Institute for Brain, Cognition and Behaviour)
1PM BIOMED BLDG C44 and online: https://bristol-ac-uk.zoom.us/j/92144801092?pwd=YVNKeGFFRW1kb09iQnU0MkRIRzV5UT09
Meeting ID: 921 4480 1092 Passcode: 225286
Hippocampal-entorhinal circuits for spatial memory
The hippocampus is important for spatial and episodic memory. Place cells – the principal cell of the hippocampus – represent information about an animal’s spatial location. Yet, during sleep and rest place cells spontaneously recapitulate (‘replay’) past trajectories. Replay has been hypothesised to serve a variety of functions in memory. In my talk I will describe recent work I carried out which showed replay may support a dual function: underpinning both spatial planning as well as the consolidation of new memories. Namely, we found during rest periods place and grid cells, from the deep medial entorhinal cortex (dMEC, the principal cortical output region of the hippocampus), replayed coherently. Importantly, putative dMEC replay lagged place cell replay by ~11ms; suggesting the replay coordination may reflect consolidation. Moreover, in a separate study we found replay occurring just before movement to or upon arrival at a reward site preferentially depicted locations and trajectories consistent with the animals’ current task demands; perhaps indicative of spatial planning. However, we also found replay could dynamically ‘switch’ between a planning and consolidation mode, in relation to engagement with task demands, and we found planning-like replay predicted the accuracy of imminent spatial decision. Finally, I will discuss unpublished work showing how the formation of hippocampal-dMEC cell assemblies during encoding periods may underlie hippocampal-dMEC replay coordination and on-going work where we employ an ontogenetic approach to elucidating the neural circuit mechanisms of spatial memory.
8th April 2022
Mark D Humphries (Professor of Computational Neuroscience, University of Nottingham)
1PM BIOMED BLDG C44 and online: https://bristol-ac-uk.zoom.us/j/92144801092?pwd=YVNKeGFFRW1kb09iQnU0MkRIRzV5UT09
Meeting ID: 921 4480 1092 Passcode: 225286
Simultaneous and separable latent encoding of arm movement direction and kinematics in motor cortex
Little is known about if and how multiple features of movement are simultaneously encoded by population activity in motor cortex. Using neural activity from dorsal premotor cortex (PMd) and motor cortex (M1) as monkeys performed a sequential arm movement task, in this talk I will show that the direction and kinematics of arm movements are simultaneously but separably encoded in the low-dimensional trajectories of population activity. Trajectories of population activity encoded the direction of arm movement, with the distances between neural trajectories proportional to the difference in angle between the directions they encoded. By contrast, different durations of arm movements in the same direction were encoded by the how long the neural trajectory took to traverse. A recurrent neural network (RNN) model of our results suggested the direction and duration could be independently controlled by respectively rotating the inputs to motor cortex and scaling the effective neuron time constant within motor cortex. Our results propose a mechanism for the simultaneous yet independent control of multiple arm movement features by motor cortex.
1st April 2022
Youssef Mohamed (Doctoral Student, KTH Royal Institute of Technology)
1PM BIOMED BLDG C44 and online: https://bristol-ac-uk.zoom.us/j/92144801092?pwd=YVNKeGFFRW1kb09iQnU0MkRIRzV5UT09
Meeting ID: 921 4480 1092 Passcode: 225286
Human-Robot Interaction and Social Robotics
How robots perceive humans is as important as how humans perceive robots. Hence, developing systems that are able to understand and rationalize our actions, is the first step to creating more socially aware robots. Nonetheless, doing so can be “complicated” as human actions can sometimes be illogical but patterns can be detected and inferences can be made using AI. In this seminar we will be discussing the state-of-the-art approaches of detecting human intentions and internal states and how those systems are able to do so. Furthermore, we will also bring up some of the moral and societal concerns that those systems invoke.
25th March 2022
Dr Alexandra Keinath (Postdoctoral Fellow, McGill University)
2PM BIOMED BLDG C44 and online: https://bristol-ac-uk.zoom.us/j/92144801092?pwd=YVNKeGFFRW1kb09iQnU0MkRIRzV5UT09
Meeting ID: 921 4480 1092 Passcode: 225286
Dynamic maps of a dynamic world: how hippocampal and entorhinal representations cope with an ever-changing experience of an unstable world
Extensive research has revealed that the hippocampus and entorhinal cortex maintain a rich representation of space through the coordinated activity of place cells, grid cells, and other spatial cell types. Frequently described as a ‘cognitive map’ or a ‘hippocampal map’, these maps are thought to support episodic memory through their instantiation and retrieval. Though often a useful and intuitive metaphor, a map typically evokes a static representation of the external world. However, the world itself as well as our experience of it are intrinsically dynamic. Here I will present three projects where we address how hippocampal and entorhinal representations adapt to, incorporate, and overcome these dynamics. In the first project, I will describe how boundaries dynamically anchor entorhinal grid cells and human spatial memory alike when the shape of a familiar environment is changed. In the second project, I will describe how the hippocampus maintains a representation of the recent past even in the absence of disambiguating sensory and explicit task demands, a representation which causally depends on intrinsic hippocampal circuitry. In the third project, I will describe how the hippocampus preserves a stable representation of context despite ongoing representational changes across a timescale of weeks. Together, these projects yield new insight into the dynamic and adaptive nature of our hippocampal and entorhinal representations and set the stage for exciting future work building on these techniques and paradigms.
18th March 2022
Dr Petra Fischer (Lecturer, University of Bristol)
1PM BIOMED BLDG C44
Coordination of neural activity during action choice
Life is a series of action choices. One can ignore external events or react to them. I will discuss two projects that show cortical gamma synchronization during rapid evaluation of external events and adjustment of an ongoing action. Our behavioural task is simple, but can be used as a powerful tool to understand how neural activity is coordinated within cortico-subcortical circuits to produce fundamental components of flexible behaviour. I will discuss the implications of brief periods of neural synchronization for coordination of neural activity and ideas for future projects.
11th March 2022
Darinka Trübutschek (Research associate, the Max Planck Institute for Empirical Aesthetics)
1PM Meeting ID: 921 4480 1092 Passcode: 225286
https://bristol-ac-uk.zoom.us/j/92144801092?pwd=YVNKeGFFRW1kb09iQnU0MkRIRzV5UT09
Decoding our thoughts: Tracking the contents of non-conscious working memory
Our daily and intellectual lives depend on our ability to hold information in mind for immediate use. Despite a rich research history, cracking the neuro-cognitive code of such working memory remains one of the most important challenges for neuroscience to date. According to the pre-dominant theoretical stance, maintaining information in working memory requires conscious, effortful activity sustained over the entire delay period.
However, this might reflect only the tip-of-the-iceberg. Recent studies have shed doubt on these long-standing assumptions showing either that (1) even subliminal stimuli may be stored for several seconds in non-conscious working memory and that (2) memories might also be retained in the absence of accompanying neural activity in activity-silent working memory by means of slowly decaying synaptic changes.
During this talk, I will present evidence from a series of behavioral and magnetoencephalography experiments that help to reconcile these diverging frameworks. Specifically, while the short-term maintenance of information may be entirely decoupled from both conscious experience and persistent neural activity, the hallmark feature of working memory – the ability to manipulate information – requires both. As such, these findings provide strong evidence against a genuinely non-conscious ‘working’ memory, and instead point towards the existence of an activity-silent short-term memory. Moreover, they offer a theoretical framework of how both active and/or conscious and activity-silent and/or non-conscious brain processes may interact to support working memory.
25th February 2022
Dr Charlotte Horne (Senior Research Associate, University of Bristol)
1PM BIOMED BLDG C44
The role of cognitive control in treatment-resistant schizophrenia and the relationship with sleep loss
In part 1 of the talk, I will present recent work completed during my post-doc at King’s College London. Approximately one third of patients with schizophrenia fail to respond to antipsychotic medication – termed ‘treatment resistance’ – and the underlying mechanisms remain unclear. Here, I present data suggesting poor cognitive control may be a mechanism underlying treatment resistant schizophrenia. We investigated effective connectivity within a network of interacting regions responsible for cognitive control, sensory and reward processing using Dynamic Causal Modelling (DCM) whilst patients performed an fMRI reward learning task. Treatment-resistant patients had reduced top-down connectivity compared to treatment-responsive patients which may be underpinned by a glutamatergic abnormality (as measured using MR spectroscopy). Our findings suggest that treatment resistance may represent a subtype of schizophrenia with a distinct underlying mechanism that is not targeted by current medication.
In part 2 of the talk, I will (informally) present some of my current interests around the causal role of sleep loss on the development of mental illness (e.g. depressive and psychotic symptoms). In particular the effects of sleep/circadian disruption on cognitive control and what makes some people more vulnerable to these effects. I hope to create more of an informal discussion around these ideas as I am really keen to hear more about similar research taking place in Bristol and to collaborate.
18th February 2022
Karan Grewal (Research Scientist, Numenta)
4PM Meeting ID: 921 4480 1092 Passcode: 225286
https://bristol-ac-uk.zoom.us/j/92144801092?pwd=YVNKeGFFRW1kb09iQnU0MkRIRzV5UT09
Avoiding Catastrophe: Active Dendrites Enable Multi-Task Learning in Dynamic Environment
A key challenge for AI is to build systems that must adapt to changing task contexts and learn continuously. Although standard deep learning systems achieve state of the art results on static benchmarks, they often struggle in dynamic scenarios where the task changes over time, and this phenomenon is known as catastrophic forgetting. In this talk, I will discuss how biophysical properties of dendrites in the brain and local inhibitory systems enable networks to dynamically restrict and route information in a context-specific manner. Specifically, I will highlight the performance of a deep learning architecture that embodies properties of dendrites on two separate benchmarks requiring task-based adaptation: Meta-World (a multi-task reinforcement learning environment where a robotic agent must learn to solve a variety of manipulation tasks simultaneously) and permutedMNIST (a continual learning benchmark in which the model’s prediction task changes throughout training). Analysis on both benchmarks demonstrates the emergence of overlapping but distinct and sparse subnetworks, allowing the system to fluidly learn multiple tasks with minimal forgetting. This work sheds light on how biological properties of neurons can inform deep learning systems to address dynamic scenarios that are typically impossible for traditional ANNs to solve.
11th February 2022
Dr Lillian J Brady (Postdoctoral Research Fellow, Vanderbilt University)
3PM Meeting ID: 937 0313 1980 Passcode: 422725
https://bristol-ac-uk.zoom.us/j/93703131980?pwd=MFBNbDU4bFcxeTRRL3NxdlU2SVZ4UT09
Sex differences in cholinergic regulation of dopamine release through nicotinic receptors mediate sexually dimorphic behavior
Dopamine release dynamics in the mesolimbic dopamine pathway, which connects the ventral tegmental area (VTA) to the nucleus accumbens (NAc), are an essential component of the process that controls motivation and reward-seeking behavior in Substance Use Disorder. In the NAc specifically, tonic and phasic dopamine release is known to play a critical role in converting information about environmental reward-predictive cues to anticipated rewarding outcomes and is heavily modulated by the activity of cholinergic (ChAT) interneurons signaling through nicotinic acetylcholine receptors (nAChRs). Using operant conditioning and fast-scan cyclic voltammetry with pharmacology we defined sex differences in ChAT regulation of dopamine release underlying sex-specific motivational strategies for non-drug rewards. We find critical differences in cholinergic regulation of dopamine terminals that underlies distinct differences in behavioral strategies between males and females.
26th November 2021
Dr Emma Cahill (Lecturer, University of Bristol)
1PM BIOMED BLDG C44
BLA, BLA, BLA… what about the BNST!?: Role of the amygdala and extended amygdala circuits for the detection of threat cues in rats
Learning which threats to avoid is key for survival. We can model how a rodent learns about threats using pavlovian associative conditioning tasks, where the rodent learns that a specific conditioned stimulus was predictive of an aversive event. The imminence of a threat is assessed by either the threat cue physical proximity as close/distant, or by a psychological prediction of its likelihood as a recognisable predictable/unpredictable event. There is a wealth of data regarding the neurochemical mechanisms that take place within the amygdala nuclei, basolateral (BLA) and Central (CeN) for this associative learning to occur. However, the so-called extended amygdala, the bed nuclei of the stria terminalis (BNST), seem to play a role specifically in the detection of ambiguous threat cues, more akin to anxiety-like responses. In this talk, I will present some recent data on the activation of the amygdala and extended amygdala after modifications of the threat cue. In one set of experiments, the predictability of the CS-US relationship was modified by changing the reinforcement contingency during training. In a second set of experiments, the salience of the CS was modified at test to reduce its detectability. There remain many open questions regarding under what conditions the specific subnuclei of the BNST become recruited and how they regulate interconnected amygdala circuits to influence responding to threats.
19th November 2021
Professor Anne Collins (Assistant Professor, University of California, Berkeley)
3PM Meeting ID: 937 0313 1980 Passcode: 422725
https://bristol-ac-uk.zoom.us/j/93703131980?pwd=MFBNbDU4bFcxeTRRL3NxdlU2SVZ4UT09
Executive contributions to reinforcement learning computations in humans
The study of the neural processes that support reinforcement learning has been greatly successful. It has characterized a simple brain network (including cortico-basal ganglia loops and dopaminergic signalling) that enables animals to learn to make valuable choices, using valanced outcomes. However, increasing evidence shows that the story is more complex in humans, where additional processes also contribute importantly to learning. In this talk, I will show three examples of how prefrontal-dependent executive processes are essential to reinforcement learning in humans, operating both in parallel to the brain’s reinforcement learning network, as well as feeding this network information.
12th November 2021
Dr Abhishek Banerjee (Senior Lecturer, University of Newcastle)
1PM BIOMED BLDG C42
Prefrontal reprogramming of sensory cortex: Cellular and computational principles
Animals adapt their behaviour in response to variable changes in reward reinforcement. Value-based decision-making involves multiple cognitive maps across distributed brain areas. It is less clear which brain regions are essential and how changes in neural responses flexibly re-maps guiding adaptive behaviour. In this talk, I will highlight behavioural-neural interactions between orbitofrontal and somatosensory circuits that implement flexible decision-making. I will present further evidence of how some of these functions are disrupted in autism spectrum disorders, arguing for a new conceptual framework based on computational psychiatry to understand cognitive pathophysiology in neurological disorders.
22nd October 2021
Dr Michele Veldsman (Research Scientist, University of Oxford)
1PM GEOG BLDG G.11N SR1
MRI Markers of vascular cognitive impairment
Cerebrovascular risk factors increase the likelihood of dementia. High cerebrovascular burden leads to vascular dementia, accounting for around 20% of dementia cases. Less well appreciated, is that up to 70% of patients with Alzheimer’s disease (AD) also have cerebrovascular disease pathology at post-mortem. The impact of mixed pathologies is likely greatly underestimated. Studies of neurodegenerative dementias rarely control cerebrovascular burden, beyond age and obvious magnetic resonance imaging (MRI) markers like white matter hyperintensities (WMHs). In this talk, I will show work investigating MRI markers of cerebrovascular burden in healthy ageing in 22 000 people from the UK Biobank. I will demonstrate new methods for the estimation of the spatial distribution of cerebrovascular risk-related WMHs and their impact on cognition. I will also present work looking at the importance of microstructural integrity of normal appearing white matter and integrity of grey matter in distributed brain networks for the preservation of cognitive function in healthy ageing and after ischaemic stroke. Together, I will build up a picture of the important MRI markers of cerebrovascular burden that may act as transdiagnostic markers of cognitive impairment.
