Neural Computation Workshop Artificial Intelligence (2011)

Neural Computation Workshop Artificial Intelligence (2011) Dartmouth College Lectures (with subtitles) Please note: This is very sophisticated stuff - don't leech this if you're not ready for this. Don't waste the community's bandwidth. Neural Computation: Population Coding of High-Level Representations The Center for Cognitive Neuroscience at Dartmouth and the Neukom Institute hosted the workshop, "Neural Computation: Population Coding of High-Level Representations". The workshop, August 18 and 19, 2011 consisted of talks by about a dozen speakers with an emphasis on extensive discussion after each talk. The theme of the workshop was computational approaches to modeling how complex stimuli are encoded in population responses and how to decode brain activity to identify the information content that is represented in population responses. The talks emphasized work in visual neuroscience but were not limited to visual representation. In addition to the high-profile speakers, each speaker was invited to bring a junior member from their lab – grad student, post-doc, or junior faculty. A poster session to highlight the work of the junior colleagues was included as an effective way to enrich discussion and enable one-on-one discussions with senior investigators. Videos of the speakers (in alphabetical order): 01: Charles Cadieu, PhD, Redwood Center for Neuroscience, UC Berkeley Learning Intermediate-Level Representations of Form and Motion from Natural Movies 02: Ed Connor, Prof. of Neuroscience; Director, The Zanvyl Krieger Mind/Brain Institute, Johns Hopkins Neural coding of object structure in the ventral visual pathway 03: Jim DiCarlo, MD, PhD, Assoc. Prof. of Neuroscience; McGovern Institute for Brain Research, MIT Untangling object recognition: Which neuronal population codes can explain human object recognition performance? 04: (video not available) Jack Gallant, Prof. of Psychology and Helen Wills Neuroscience Institute; Director, The Gallant Lab, UC Berkeley Using voxel-wise encoding models to discover brain representations and to decode brain activity 04-2: (replacement for missing video 04) Brain representations of information in natural movies-HD Jack Gallant, Helen Wills Neuroscience Institute, University of California, Berkeley Abstract: How is human visual system organized into functional areas, and what information is represented in each area? The best current technology for exploring these issues is functional MRI, but thus far fMRI has only provided coarse and patchy information about the functional organization of the human visual system. We have developed a new approach for collecting and modeling fMRI data that reveals the functional organization of the human visual system in much greater detail than was possible previously. Our approach is based on estimation of quantitative voxel-wise encoding models from fMRI responses evoked by natural movies. Projection of these encoding models onto flattened maps of individual human brains reveals a highly detailed and systematic representation of structural and semantic information distributed across wide swaths of visual and non-visual cortex. These patterns are consistent with the coarse parcellations provided by previous techniques, but provide much more detailed information and extend well beyond areas identified in earlier studies. Furthermore, examination of voxel-wise encoding models reveals what specific information is represented within each visual area, and suggests how the visual system exploits simple spatial and temporal features in order to construct semantic representations of objects and scenes. Finally, one additional benefit of our approach is that estimated encoding models can be easily converted into decoding models. These decoding models recover both the structural and semantic information in natural movies, even from slow hemodynamic signals. 05: Jim Haxby, Evans Family Distinguished Professor; Director, Center for Cognitive Neuroscience; Prof. of Psychological and Brain Sciences, Dartmouth Building common, high-dimensional models of neural representational spaces 06: Nikolaus Kriegeskorte, Faculty, Cognition and Brain Sciences Unit, University of Cambridge Representational similarity analysis of visual-object population codes 07: Tom Mitchell, Prof. of Computer Science; Chair, Machine Learning Dept., Carnegie Mellon What neural activity encodes about stimuli, where and when 08: Alice O’Toole, Prof. Cognition and Neuroscience, UT Dallas Understanding neural representation of facial identity, race, and viewpoint: Constraining the neural with the perceptual 09: Tomaso Poggio, Eugene McDermott Professor in the Brain Sciences and Human Behavior; Director of the Center for Biological and Computational Learning, MIT The computational magic of the ventral stream: Towards a theory 10: Eero Simoncelli , Investigator, Howard Hughes Medical Institute; Prof. of Neural Science, Mathematics and Psychology, NYU Metamers of the ventral stream text: Encoding and decoding with voxel-wise models - Gallant.pdf source: http://neukom.dartmouth.edu/programs/neuralcomp-workshop11.html tags: AI, brain, mind, neural, artificial intelligence, computing, hardware, software, posthuman