PRNI 2016 - INVITED SPEAKERS
Title: How multi-voxel pattern analyses revolutionized human cognitive neuroscience
Abstract:
Multi-voxel pattern analyses (MVPA) represent a very significant upgrade of the toolkit of human cognitive neuroscience. Although caution is needed when the results are interpreted, MVPA have increased the level of detail with which we can probe the functional architecture of the human brain. In addition, MVPA provide a principled approach to not only investigate the localization of mental representations but also test hypotheses about the content and properties of these representations. I will discuss the power of MVPA in these two respects for understanding brain functioning in general as well as in clinical populations such as neurodevelopmental disorders.
Short bio:
Hans Op de Beeck (PhD, 2003) is a professor in cognitive and systems neuroscience at KU Leuven (Belgium). He received his training at KU Leuven and MIT. He has main-author publications in important neuroscience journals such as Nature Neuroscience, Nature Reviews Neuroscience, Trends in Cognitive Sciences, Current Biology, Journal of Neuroscience, and NeuroImage. He received many competitive awards and grants for his research, including a career development award from the Human Frontier Science Program and a starting grant from the European Research Council. His research integrates multivariate analyses in human, computational, and systems neuroscience, focusing upon topics such as visual cognition and neural plasticity.
Hans Op de Beeck
Title:
A common model of representational spaces in human cortex
Abstract:
The functional architecture of human cortex can be modeled as high-dimensional representational spaces in which patterns of brain activity are recast as vectors with basis functions that have tuning profiles and patterns of connectivity that are common across brains. Transformation matrices that rotate individual anatomical spaces into the common model space are derived with searchlight-based, whole cortex hyperalignment. Patterns of brain activity in individual brains are modeled as multiplexed topographic basis functions. This model provides a common structure that captures fine-grained distinctions among cortical patterns of response that are not modeled well by current brain atlases and, in addition, accounts for individual variability in features of coarse-scale functional topographies.
Short Bio:
Jim Haxby is the Evans Family Distinguished Professor and the Director of the Center for Cognitive Neuroscience at Dartmouth. He also is a professor in the Center for Mind/Brain Science (CIMeC) at the University of Trento. His work investigates how the brain perceives and understands faces and objects and the development and application of computational methods for ‘neural decoding’ – how to decode the thoughts and perceptions from patterns of brain activity. His most recent work has focused on the development of computational methods for building a common, high-dimensional model of representational spaces in human cortex.
James V. Haxby
Radoslaw Martin Cichy
Title:
Knowing what is happening where and when in the brain by quantitatively comparing representational similarities
Abstract:
Understanding visual cognition in the brain requires answering three questions: what is happening where and when in the human brain when we see? These seemingly simple questions pose a major methodological challenge for contemporary cognitive neuroscience: answering them in an integrated fashion requires the quantitative concertation of multivariate data from different brain measurement techniques, computational modeling and behavior. In this talk I will discuss approaches that aim to accomplish this integration, in particular emphasizing representational similarity analysis (RSA). In the first and major part, I will discuss how functional magnetic resonance imaging (fMRI) and magneto- and electroencephalography (MEEG) can be combined to yield a spatio-temporally integrated view of human brain activity. In the second part I will show how the RSA-based approach naturally extends to integration of computational models and behavioral data. I will highlight research revealing a relations between spatio-temporal brain dynamics and deep neural network models, as well as behavior (similarity ratings, reaction times in categorization). In the third part I will discuss potential future developments, pitfalls and directions.
Short Bio:
Radoslaw Martin Cichy heads the Visual Cognition group at Free University Berlin since May 2016. His research focuses on mapping and understanding the neural dynamics of visual object recognition, using MEEG, fMRI, and deep neural networks. Before starting his own group, RM Cichy worked at the Computer Science and Artificial Intelligence Laboratory at MIT in the group of Aude Oliva. He earned his B.S. degree in Cognitive Science from the Univeristy of Osnabrück and his Ph.D. from Humboldt University, working at the Bernstein Center for Computational Neuroscience Berlin with John-Dylan Haynes.
Alexandre Barachant
Title:
Information Geometry: A framework for manipulation and classification of M/EEG data
Abstract:
Information geometry considers probability distributions as points of a Riemannian manifold, and provides a natural metric through which it is possible to derive a set of geometric tools (distance, tangent, mean or median ...) allowing us to analyze, interpret and classify these points while keeping intact their information content. In this presentation, we will see how to apply this framework in the context of M/EEG data. To this end, epochs of M/EEG signals are represented by their multivariate normal distributions and treated as points of such manifold, allowing us to benefit from the multiple properties of the corresponding Riemannian metric. The theoretical link between this approach and sources separation methods will be shown, and examples of practical use of this framework will be presented.
Short Bio:
Alexandre Barachant is a French Researcher, expert in Brain computer interfacing and Biosignal analysis. He received his Ph.D. degree in signal processing in 2012 from the Grenoble University, France. During his Ph.D. thesis, he developed a robust and adaptive brain computer interface based on self-paced motor imagery. Between 2012 and 2013, he has been a post-doc fellow at the Centre National de la Recherche Scientifique (CNRS) in the GIPSA Laboratory, Grenoble, where he developed a calibration-less P300 brain computer interface for video-game control and assistive communication. In November 2014 he joined the Burke Medical Research Institute, New York, to study the effects of non-invasive brain stimulation for rehabilitation. Alexandre's work on Information geometry led him to win 3 international M/EEG data analysis competitions between 2014 and 2015. His research interests include statistical signal processing, Machine learning, Riemannian geometry and classification of neurophysiological recordings.
Zoltan Szabo
Title:
Hypothesis Testing with Kernels
Abstract:
In my talk, I am going to focus on two tasks: two-sample testing, and independence testing. In the first setting, we are given two sets of observations, and our goal is to check whether the two sets are statistically indistinguishable (this may be thought of as a generalization of a t-test). In the second setting, we have paired samples, and the goal is to determine whether they are independent (this generalizes the Pearson correlation test). Both tests are built around kernels, which are natural measures of similarity between objects, and have been developed for a very wide variety of domains. Thus, the tests apply to data as diverse as text, images, time series or graphs.
Short Bio:
Zoltan Szabo is a Research Associate at the Gatsby Unit, University College London (2013 - present). He holds a double PhD in Computer Science and Applied Mathematics from the Eotvos Lorand University (2009-2012; Budapest, Hungary). His primary research interests are information theory, statistical machine learning, empirical processes and kernel methods with applications in remote sensing (sustainability), distribution regression, structured sparsity, independent subspace analysis and its extensions, collaborative filtering.
