INC Seminar Series

Sponsor: Institute for Neural Computation

Affiliation: Gatsby Computational Neuroscience Unit University College London

Location:

San Diego Supercomputer Center, East Annex,
Level B1, South Wing
Located on Hopkins Drive, next to RIMAC, facing the Eucalyptus Grove

Time: 10:00–11:00

Title/Abstract:

"Hierarchical Bayesian Models of Language and Text"

In this talk I will present a new approach to modelling sequence data called the sequence memoizer. As opposed to most other sequence models, our model does not make any Markovian assumptions. Instead, we use a hierarchical Bayesian approach which enforces sharing of statistical strength across the different parts of the model. To make computations with the model efficient, and to better model the power-law statistics often observed in sequence data, we use a Bayesian nonparametric prior called the Pitman-Yor process as building blocks in the hierarchical model. We show state-of-the-art results on language modelling and text compression.

Sponsor: Institute for Neural Computation

Affiliation: Psychology Department, Cognitive Science Center, Computational Biomedical Imaging and Modeling Center, Rutgers University

Location:

San Diego Supercomputer Center, East Annex,
Level B1, South Wing, Room E129
Located on Hopkins Drive, next to RIMAC, facing the Eucalyptus Grove

Time: 1100 -1200

Host:Howard Poizner

Title/Abstract:

"Funneling attention in automated behavior"

In the first part of this talk I will report on some new findings on the role of motor cortex neuronal spike times on directional and temporal selectivity for point to point reaches. This work has been a collaborative effort with Jose Carmena from UC Berkeley and Jorge V. Jose from Indiana U. In the information transmission patterns of these cells we found an oscillatory angular selection rule useful to update different acceleration temporal scales. Such a signal from each motor cortical neuron can unambiguously inform when to start/stop the reach. In the second part of this talk I will report on some new findings from my sensory-motor integration lab at Rutgers U on the cross talk between complex automated full body movements and attention. This work has been in collaboration with undergraduate Psychology students that took my Sensation and Perception class in the Spring semester of 2009 and became interested in movement research. The common theme in both reports is attention modulation in automated actions.

Sponsor: Institute for Neural Computation

Affiliation: Columbia University College of Physicians and Surgeons

Location:

San Diego Supercomputer Center, East Annex,
Level B1, South Wing, Room E129
Located on Hopkins Drive, next to RIMAC, facing the Eucalyptus Grove

Time: 1330 -1430

Hosts: Howard Poizner/Terry Sejnowski

Title/Abstract:

"The Basal Ganglia and Implicit Choices in Motor Control"

Sponsor: Institute for Neural Computation

Affiliation: Research Imaging Institute University of Texas Health Science Center, San Antonio

Location:

San Diego Supercomputer Center, East Annex,
Level B1, South Wing, INC Main Area
Located on Hopkins Drive, next to RIMAC, facing the Eucalyptus Grove

Time: 1130 -1230

Host: Howard Poizner

Title/Abstract:

"Mapping processing time between cortical regions using single pulse transcranial magnetic stimulation"

This talk will focus on sensory contributions to motor control. Specifically, this work addresses sensory mechanisms involved in the control of the human voice. Subjects produce an "ah" at their typical pitch and their voice signal is fed back to the them through headphones. As subjects listen to their own voice, the pitch is unexpectedly shifted up or down. Subjects respond to this by immediately shifting the fundamental frequency of their voice production in the opposite direction of the pitch shift perturbation. We consider this response reflexive because subjects are no conscious of shifting their own pitch in response to the perturbation and the timing of the responses are very rapid. In this work we are using fMRI map the regions of the brain involved in the reflex and single pulse Transcranial Magnetic Stimulation to quantify processing time between cortical regions involved in the reflex. This work has direct application to understanding sensory motor integration in general and developing a firmer understanding of neurological disorders. The talk will focus on the role of M1 voice in the pitch shift reflex and the timing between auditory cortex and M1. Work supported by NIDCD R01 DC006243, Sensory Mechanisms of Voice Control (PI Charles Larson, Co-PI Donald A. Robin).

Sponsor: Institute for Neural Computation Seminar

Affiliation: Laboratory for Perceptual Dynamics, RIKEN Brain Science Institute

Location:

San Diego Supercomputer Center, East Annex,
Level B1, South Wing, INC Main Area
Located on Hopkins Drive, next to RIMAC, facing the Eucalyptus Grove

Time: 1330 -1430

Hosts: Sergei Gepshtein & Howard Poizner

Title/Abstract:

"Dynamics of EEG phase synchrony in the resting and working brain: quasi-stable intervals."

Cereb Cortex. 2010 Feb;20(2):365-82. Epub 2009 Jul 13.

Duration of coherence intervals in electrical brain activity in perceptual organization Nikolaev AR, Gepshtein S, Gong P, van Leeuwen C.

Laboratory for Perceptual Dynamics, RIKEN Brain Science Institute, Wako-shi 351-0198, Japan. nikolaev@brain.riken.jp

We investigated the relationship between visual experience and temporal intervals of synchronized brain activity. Using high-density scalp electroencephalography, we examined how synchronized activity depends on visual stimulus information and on individual observer sensitivity. In a perceptual grouping task, we varied the ambiguity of visual stimuli and estimated observer sensitivity to this variation. We found that durations of synchronized activity in the beta frequency band were associated with both stimulus ambiguity and sensitivity: the lower the stimulus ambiguity and the higher individual observer sensitivity the longer were the episodes of synchronized activity. Durations of synchronized activity intervals followed an extreme value distribution, indicating that they were limited by the slowest mechanism among the multiple neural mechanisms engaged in the perceptual task. Because the degree of stimulus ambiguity is (inversely) related to the amount of stimulus information, the durations of synchronous episodes reflect the amount of stimulus information processed in the task. We therefore interpreted our results as evidence that the alternating episodes of desynchronized and synchronized electrical brain activity reflect, respectively, the processing of information within local regions and the transfer of information across regions.

Sponsors: Joint Institute for Neural Computation - IEM Seminar

Affiliation: Distinguished McKnight University Professor Director, Center for Neuroengineering Director, NIH/NIBIB Training Program on Neuroimaging Department of Biomedical Engineering University of Minnesota

Location: Fung Auditorium, Powell-Focht Bioengineering Hall University of California San Diego MAP

Time: 1300 -1400

Title/Abstract:

Spatio-temporal Functional Neuroimaging of Brain Activity

Brain activity is distributed over the three-dimensional brain and evolves over time. Over the past decades, functional neuroimaging has emerged as an important interdisciplinary research area. This has been in particular promoted by the development of functional MRI and the significant advancement in electrophysiological neuroimaging using EEG/MEG. We will review our work in electrophysiological neuroimaging integrating EEG with structural MRI, and show its applications to aid presurgical planning in epilepsy patients. We will also review our work on multimodal functional neuroimaging integrating electrophysiological and hemodynamic measurements to significantly enhance the spatio-temporal resolution of imaging brain activity. Our recent work indicates that the event-related BOLD fMRI and electrophysiological data can be integrated in a principled way, leading to high-resolution spatio-temporal functional imaging of dynamic brain activation. We will also review the investigation of co-localization of hemodynamic and electrophysiological signals associated with motor imagery for brain-computer interface applications, using BOLD fMRI and electrophysiological neuroimaging.

Sponsors: Joint Institute for Neural Computation

Affiliation: Research Scientist V, Wadsworth Center, New York State Department of Health, Assoc. Prof., Dept. of Neurology, Albany Medical College, Assoc. Prof., Dept. of Biomed. Sci., State Univ. of New York at Albany, Adj. Assist. Prof., Dept. of Neurosurgery, Washington Univ. in St. Louis, Adj. Assoc. Prof., Dept. of Biomed. Eng., Rensselaer Polytechnic Institute

Location: Fung Auditorium, Powell-Focht Bioengineering Hall University of California San Diego MAP

Time: 1300 -1400

Host: Scott Makeig

Title/Abstract "Brain-Computer Interfacing Using Electrocorticography (ECoG)"

Brain-computer interfaces (BCIs) convert brain signals into outputs that communicate a user's intent. BCIs can be used by people to communicate and interact with their environment. However, practical applications of BCI technology are currently impeded by the limitations of the prevailing non-invasive and invasive recording methods. Electrocorticographic (ECoG) recordings from the cortical surface could be a powerful and practical alternative to these methods. ECoG has much higher fidelity than EEG and is likely to have greater long-term stability than intracortical recordings. In this talk, I will give an overview of the current state in BCI research. I will then demonstrate that ECoG can give detailed information about motor and language function that is in important ways comparable to that provided by intracortical recordings, and that it supports rapid acquisition of real-time BCI control in humans. Finally, I will show results from a comprehensive multi-center study that used ECoG and advanced signal processing methods to realize a novel functional mapping technology for invasive brain surgery.

Sponsor: Institute for Neural Computation and Temporal Dynamics of Learning Center Seminar

Affiliation: Director, Research Imaging Center, University of Texas Health Sciences Center

Location:

San Diego Supercomputer Center (Directions)

Time: 1400 - 1515

Host: Howard Poizner

Title/Abstract:

"Modeling Treatment Effects with Functional NeuroImaging"

Sponsor: Joint Calit2 - Institute for Neural Computation - Biological Physics Seminar

Affiliation:

Location:

Calit2 Auditorium, Atkinson Hall, UC San Diego

Time: 1100-1200

Host:

Title/Abstract:

"Resilient machines"

For the past 15 years, computer graphics has been coming of age, moving from the lab to a commodity on the desktop, notebook, and now the mobile phone. Much of graphics can be described as the rendition of three-dimensional models as two-dimensional images. Its inverse problem, computer vision, concerns the reconstruction of three-dimensional environments from two-dimensional images, and -- like most inverse problems -- is much harder. We are now in a transitional period, with vision techniques beginning to break out of the well-controlled environment of the lab, with its calibrated cameras and powerful workstations, and into the real world of cheap digital cameras and pocket-sized computers. With the combined capabilities of vision and graphics, we finally have the tools to realize the mirror world envisioned by authors who popularized the idea of 'cyberspace'. After surveying computer vision techniques and seminal results, we'll review the algorithms underlying Photosynth, a tool allowing one to reconstruct 3D from digital photography, and chart its convergence with mapping and Virtual Earth.

Sponsor: Institute for Neural Computation / Embodied Cognition Lab and Temporal Dynamics of Learning Center Seminar

Affiliation: Center of Intelligent Systems and their Applications, University of Edinburgh

Location: Cognitive Science Building, UCSD, Room 280

Time: 1100-1200

Host: Rafael Nunez

Title/Abstract:

"Modeling the development of mathematical ideas"

In Proofs and Refutations, Lakatos presented an account of the origin and development of mathematical ideas in which he argued strongly against the "deductivist" style in which mathematics is presented as an ever-increasing set of universal, absolute, certain truths which exist independently of humans. Instead, he emphasised the social nature of mathematical progress, describing how different mathematicians may have different interpretations of a conjecture, examples or counterexamples of it, and beliefs regarding its value or theoremhood. We believe that Lakatos is the philosophical counterpart to Lakoff and Nunez's cognitive account of mathematics, which emphasises bodily experience of a physical world as grounding mathematical ideas. Aspects of Lakatos's theory were implimented as an agent architecture which builds on an automated theory formation system (Simon Colton's HR program), and are currently aiming to provide a computational representation of some of Lakoff and Nunez's ideas. In this talk we will describe our system and discuss ideas for modelling embodied mathematics.

URL: http://homepages.inf.ed.ac.uk/apease/research/index.html

Sponsor: Institute for Neural Computation / Temporal Dynamics of Learning Center Seminar

Affiliation: IBM Watson Research Center

Location: Computer Science Department, UCSD, Room 1202

Time: 1100-1200 pm

Hosts: Javier Movellan and Garrison Cottrell

Title/Abstract:

“Factorial Hidden Markov Models, Super-Human Speech Separation, and Word Confusability”

In modeling complex phenomena such as speech, it is natural to consider interactions between different models. In this talk we consider two different types of interaction that share a common computational structure. One stems from the problem of single-channel speech separation, in which two or more speech signals have to be extracted and recognized from an acoustic mixture. Another arises in the problem of estimating word confusability, the probability of mistaking one spoken word for another. Both of these problems, it turns out, can be reduced to inference in factorial hidden Markov models. In the speech separation problem, we have models of the grammar and acoustics of different speakers, which are combined according to an acoustic interaction model. Unfortunately, exact inference in the resulting factorial hidden Markov model is exponentially costly in the size of the grammar and the number of sources. To surmount this problem we propose a linear-time loopy belief propagation algorithm. We show that it achieves super-human speech separation in the PASCAL speech-separation challenge, in a fraction of the time required for exact factorial HMM inference. In the word confusability problem, a point of departure is to compute distances between hidden Markov models representing the acoustic features of words. Commonly used measures such as the Kullback-Liebler and Bhattacharyya divergences, however, are analytically intractable for hidden Markov models. We therefore introduce variational bounds to reduce the computation to closed-form divergences between components. In the case of Bhattacharyya divergence the resulting approximation has the same structure as a two-dimensional factorial HMM. We apply the resulting forward algorithm to a simple word confusability task, and discuss future directions.

Biosketch:
John R. Hershey received the BS degree (1992) in cognitive science from the University of California, Los Angeles and the PhD degree (2004) in cognitive science from the University of California, San Diego, where he was a founding member of the Machine Perception Laboratory. His doctoral thesis explored the use of generative graphical models for speech enhancement, face tracking, and combinations of the two. During his time at UCSD, he interned at Mitsubishi Electric Research Labs in Cambridge, Massachusetts in 2001, and in the Machine Learning and Applied Statistics Group at Microsoft Research in Seattle in 2003. In 2004, he spent a year as a visiting researcher in the Speech Group at Microsoft Research. Since 2005, he has been at IBM T. J. Watson Research Center in New York, where he is a research staff member in the Speech Algorithms and Engines group.

Sponsor: Institute for Neural Computation / Swartz Center Seminar

Affiliation: Wake Forest University

Location: Cognitive Science Building, UCSD, Room 180

Time: 1400-1500

Hosts:

Title/Abstract:

“Biomechanical Modeling of Upper Limb Movements”

The biomechanical problem of estimating skeletal dynamics and muscle force trajectories from body motion capture measurements of positions of markers placed on the skin is formally an underdetermined mathematical inverse problem. Solutions to this problem begin with a forward model of the anatomy, connectivity, and action of the musculo-skeletal system. Dr. Holzbaur has worked with the Scott Delp group at Stanford on the development of open source software for biomechanical modeling, and has contributed a detailed upper limb model. She will speak on the uses of biomechanical modeling for better defining the behavior that the brain controls, and in medical and rehabilitation research and treatment, and on the OpenSim computing environment.

Sponsor: Institute for Neural Computation Seminar

Affiliation: Center of Intelligent Systems and their Applications, University of Edinburgh

Location: Cognitive Science Building, UCSD, Room 280

Time: 1400 - 1500

Hosts:

Title/Abstract:

"Human Interactions in intelligent immersive spaces"

Sponsor: Institute for Neural Computation and the Department of Bioengineering, UCSD

Affiliation: Michigan State University

Location: Powell-Focht Bioengineering Hal, University of California, San Diego

When: 1100-1200

Host: Gert Cauwenberghs

Title/Abstract:

"Sub-microwatt Sensors for Structural Health Monitoring of Biomechanical Implants"

Long-term fatigue is one of the major causes for premature mechanical failure in biomechanical implants (e.g. hip or knee implants). Self-powered sensors that autonomously monitor the statistics of the loading cycles experienced by the implant can serve as an important diagnostic tool for predicting onset of fatigue. This talk will explore the potential of power-harvesting silicon micro-sensors for monitoring structural health of biomechanical implants. The key challenge lies in the fact that the total electrical power that can be harvested using localized strain variations in-vivo is less than 1 microwatt which exceeds the power-budget of many state-of-the-art sensors. I will present a novel long-term, batteryless mechanical usage monitoring technique that combines piezoelectric transduction with hot-electron injection on CMOS floating-gate transistors. The method is energy scalable and exploits the operational primitives in the device physics of the injection process for sensing, computation and storage. The principle can be extended for computing different usage statistics which include level crossings and impact rates, both of which are important for fatigue prediction. I will present experimental results from fabricated prototypes demonstrating the operation of the sensor beyond 10,000,000 cycles and a current consumption less than 160nA.

Sponsor: Institute for Neural Computation

Affiliation: University of Sydney (Australia)

Location: Cognitive Science Building, UCSD, Room 180

Time: 1100-1200

Host: Terrence Sejnowski

Title/Abstract:

"Quantitative Modeling of Multiscale Brain Activity"