Population Codes, Behavior, and Hierarchical Sparse Coding: an Unsupervised Learning Approach and its Connections to Artificial Neural Networks

Friday 31 May, 4pm at the Sherrington Library, DPAG Sherrington Building, Oxford

 

 

The Cortex Club is delighted to host A/ Professor Demba Ba from the Harvard University, who will be talking to us about his work on developing computational tools to identify neuronal populations during behaviour and deep sparse coding models to identify principles underlying hierarchical sensory processing in the brain. Please join us on May 31st, at the Sherrington Library, located in the Sherrington Building of the Department of Physiology, Anatomy and Genetics, Parks Road, Oxford.

A/ Professor Demba Ba has kindly agreed to meet students and staff individually. If you would like to arrange a meeting please contact Tai-Ying Lee at tai-ying.lee [at] dpag.ox.ac.uk.

 

Please also join us at the pub after the talk, to which everybody is welcome. Registration would be appreciated at: https://forms.gle/tuB9kEqwV5mapPKy8

 

Abstract

Two important problems in neuroscience are to understand 1) how populations of neurons encode stimuli and how this encoding is related to behavior and 2) how the brain represents sensory signals hierarchically. I have developed theoretical and computational unsupervised learning tools to answer these questions. In the first part of my talk, I will describe a statistical framework to identify sub-groups of neurons within a larger population that have similar response profiles. The framework clusters multiple rasters that exhibit nonlinear dynamics into an a-priori-unknown number of functional sub-groups that each comprises rasters with similar dynamics. I will show an application to clustering neuronal responses from the prefrontal cortex of mice in an experiment designed to characterize the neural underpinnings of the observational learning of fear. The method is able to identify “empathy” clusters of neurons, namely groups of neurons that allow an observer mouse to understand when a demonstrator demonstrator is in distress. In the second part of my talk, I will describe a deep generalization of the famous sparse coding model of Olhausen and Field. I will show a strong parallel between this deep sparse coding model and deep neural networks with ReLu nonlinearities, namely that a deep neural network architecture with ReLu nonlinearities arises from a finite sequence of cascaded sparse coding models, the outputs of which, except for the last element in the cascade, are sparse and unobservable. The benefits of the deep sparse coding model are two-fold. First, it gives answers based on theory to the question “what is the complexity of learning a deep ReLu auto-encoder?”. Second, it makes experimentally-testable predictions as to the principles that may underlie hierarchical sensory processing in the brain.

Understanding the brain’s model of the external world

Wednesday 22 May, 11am at the Large Lecture Theatre, Le Gros Clarke Building, Oxford

 

The Cortex Club is delighted to co-host, together with the Neurotheory Seminar Series, Prof Loren Frank from UCSF,  who will be talking to us about his work on how neuronal networks create a predictive model of the external world. Please join us on May 22nd, at the Large Lecture Theatre located in the Le Gros Clark Building of the Department of Physiology, Anatomy and Genetics, Parks Road, Oxford.

 

Prof. Loren Frank has kindly agreed to meet students and staff individually. If you would like to arrange a meeting please contact Tai-Ying Lee at tai-ying.lee [at] dpag.ox.ac.uk.

 

 

Abstract

The ability to create an accurate, predictive model of the world is one of the most remarkable attributes of the brain.  Our goal is to understand how activity and plasticity in neural circuits underlie both the ability to create this model and to use it to make decisions. In this talk I will focus on neural activity patterns that have the potential to play a central role in model-based decision-making.  This work began with the realization that the ability to generate and evaluate representations of hypothetical experience, whether of a counterfactual past or of a possible future, has profound adaptive value. How and when the brain might express these representations has not been clear, and I will describe work from my laboratory that has identified and characterized these representations as a surprisingly common motif in hippocampal spiking activity. We have also found that changes in firing rates in medial prefrontal cortex can be seen immediately before and during the expression of these hippocampal representations, consistent with the possibility that this activity marks the time of engagement and evaluation of mental simulations.

Next generation brain maps – concepts, challenges, collaboration

Friday 17 May, 14.00pm at the Sherrington Library, Sherrington Building, Oxford

 

The Cortex Club is delighted to host a Q&A session with Prof Katrin Amunts from the Institute of Neuroscience and Medicine, Forschungszentrum Julich, Germany, who will be giving the DPAG Head of Department Seminar in the Large Lecture Theatre at 13.00. Please join us after the lecture for the Q&A session on May 17th at the Sherrington Library located in the Sherrington Building, Department of Physiology, Anatomy and Genetics, Oxford.

Please register for the Q&A session at: https://forms.gle/oVofw8CJn9YKosdF7
Free sandwich lunch provided!

 

 

Abstract

The human brain is a highly complex system, with different levels of spatial organisation. E.g., on a macroscopic level, the brain shows a highly variable folding pattern, while nerve cells on a microscopical level are arranged in layers and columns in a regionally specific way. Cytoarchitecture is a concept that itself encompasses different aspects of brain organization – the different cell types have distinct morphology, molecular, genetic and connectional fingerprints. Axons form complex networks at the level of microcircuits or large cognitive system. To capture the cellular and axonal architecture and to study the role of a specific brain region to function or behaviour requires to analyse the brain in 3D with microscopical resolution. Deep-learning offers new tools to 3D reconstruct images of histological
sections at the microscopical scale, and convolutional neuronal networks support to automatize brain mapping. Considering the size of the brain with its nearly 86 billion nerve cells, HPC-based workflows play an increasing
role for developing high-resolution brain models, to tame brain complexity. To develop such tools is key in the Human Brain Project. It is building a European research infrastructure for brain research, to collaborate towards a better understanding of the human brain.

Circadian Clock Genes and the Importance of Timing in Aging and Longevity

Thursday 16 May, 12.30pm at the Seminar Room A, Level 6 West Wing, John Radcliffe Hospital, Oxford

 

The Cortex Club is excited to host a Q&A session with Prof Joseph Takahasi from University of Texas Southwestern Medical Center, who will be giving the Thomas Willis Lecture.  Please join us on May 16th, in Seminar Room A, Level 6, located at the West Wing of the John Radcliffe Hospital, Oxford.

To attend the Q&A session please register at: https://forms.gle/hBc2udGJ9j5vcWo87

If you would like to sign up for the Thomas Willis Lecture at 16.30pm, please register at: https://www.eventbrite.co.uk/e/thomas-willis-day-2019-tickets-59453580259

 

 

Joseph S. Takahashi, Ph.D. is the Loyd B. Sands Distinguished Chair in Neuroscience, an Investigator in the Howard Hughes Medical Institute, and Chair of the Department of Neuroscience at the University of Texas Southwestern
Medical Center in Dallas. Dr. Takahashi joined the UT Southwestern faculty in 2009, and prior to that was a faculty member in the Department of Neurobiology at Northwestern University from 1983 to 2009. Dr. Takahashi utilizes forward genetics and positional cloning in the mouse as a tool for discovery of genes underlying neurobiology and behavior, and his discovery of the mouse and human clock genes led to a description of a conserved circadian clock
mechanism in animals. He is the author of more than 290 scientific publications and the recipient of many awards including the Honma International Prize in Biological Rhythms Research in 1986, W. Alden Spencer Award in Neuroscience from Columbia University in 2001, and Outstanding Scientific Achievement Award from the Sleep Research Society in 2012. He was elected a Fellow of the American Academy of Arts and Sciences in 2000, a Member of the National Academy of Sciences in 2003, and a Member of the National Academy of Medicine in 2014.

Deep learning tools for pose estimation in the laboratory

Monday 13 May, 4pm at the Lecture Theatre, Le Gros Clark Building, Oxford

 

The Cortex Club is excited to present Dr Mackenzie Mathis, Rowland Junior Fellow at Harvard University, who will be talking to us about her work on a new tools to study deep learning using pose estimation. Please join us on May 13th, at the Lecture Theatre, located in the Le Gros Clark Building of the Department of Physiology, Anatomy and Genetics, Parks Road, Oxford.

Dr Mackenzie Mathis has kindly agreed to meet students and staff individually. If you would like to arrange a meeting please contact Tai-Ying Lee at tai-ying.lee [at] dpag.ox.ac.uk.

Please also join us at the pub after the talk, to which everybody is welcome.

 

 

Abstract

Deep learning tools for pose estimation in the laboratory

Quantifying behavior is crucial for many applications in neuroscience. Videography provides easy methods for the observation and recording of animal behavior in diverse settings, yet extracting particular aspects of a behavior for further analysis can be highly time consuming. I will discuss an efficient method for markerless pose estimation based on transfer learning with deep neural networks that achieves excellent results (i.e. you can match human labeling accuracy) with minimal training data (typically 50-200 frames). In this work, we demonstrate the versatility of this framework by tracking various body parts in multiple species across a broad collection of behaviors.