Telluride impressions

In July 2012, as in every summer of the past 17 years, the small town of Telluride, Colorado, in the heart of the Rocky Mountains, became the center of the worldwide neuromorphic engineering community for three weeks. The Telluride Neuromorphic Cognition Engineering Workshop (http://www.neuromorphs.net), which is co-organized by the University of Maryland, Johns Hopkins University and INI, has been an incredible success story since its beginnings. It has brought together world class researchers in neuromorphic engineering, neuroscience, robotics, theory, sensors, chip design, cognitive science, and many other disciplines to learn and exchange ideas in talks and tutorials. But, most importantly the goal of the workshop is to engage both junior and senior researchers in hands-on projects for three intensive weeks, which in the past has very often lead to new fruitful collaborations, or has provided the foundation for joint publications.

More than 80 researchers from all over the world (9 from INI) attended this year's workshop. The workshop was centered around 4 topic areas, which were selected by the organizers from a variety of proposed projects. Each topic area had 2-3 responsible organizers and a number of invited experts that gave tutorials and presentations related to the project, brought necessary hardware equipment, and supported the students during their projects.

The topic area on “Learning and Computational Intelligence in Neuromorphic Cognitive Systems” was organized by Giacomo Indiveri from INI and Gert Cauwenberghs from UCSD. The goal of this group was to address the problem of implementing efficient and robust “cognition” on neuromorphic hardware systems through learning, by combining models from machine learning and computational neuroscience, and studying how such models can be implemented in the different available hardware platforms from INI, UCSD, the SpiNNaker system from the University of Manchester, and convolution chips from the University of Sevilla. Many of the projects also implemented models for processing neuromorphic sensory data, e.g. from the Dynamic Vision Sensor developed in Zurich. Among the results were hierarchical (HMAX) vision models implemented in hardware, hardware reservoir computing models, spatio-temporal pattern detectors, applications of Dynamic Field Theory, and new learning approaches for event-based vision.

Kwabena Boahen from Stanford and Chris Eliasmith from the University of Waterloo organized the topic area on “Integrating Perception, Cognition, and Action in Neuromorphic Hardware and Software”, which was the continuation of a successful project from the previous year. Their goal was to build cognitive devices that interact with the real-world, using the Nengo software simulator from Waterloo to set up functional neural networks, as well as Stanford's Neurogrid and Manchester's SpiNNaker hardware for efficient real-world implementation. The projects developed several software models and various robots with basic cognitive behaviors, that showed a promising work-flow from Nengo models to hardware to an implementation that runs on real behaving robots.

“Human attention in the machine”, possibly the most ambitious topic area in Telluride history, was organized by Shihab Shamma (Univ. of Maryland), Malcolm Slaney (Microsoft), and Barbara Shinn-Cunningham (Boston Univ.). The goal was to study human auditory attention, by decoding EEG signals in real-time to determine which of two simultaneous auditory streams a person was paying attention to. Previously this had been done with invasive or more expensive methods, but this was the first demonstration that such a task can be done with EEG. The group recorded and analyzed data with an advanced EEG system, and built an impressive live demo, in which the brain signals were used to identify the attended speaker in combined 360° auditory-visual camera input.

The fourth topic area “Social Neuroscience and Robotic Pet Project” was organized by Sergi Bermudez i Badia (Madeira ITI) and Ulysses Bernadet (Vancouver) – both INI alumni. This project introduced a new topic, social neuroscience and robotics, by building models of social interactions and implementing them on a variety of robots, such as the humanoid iCub robot. The projects realized emotion and gesture recognition, an EEG interface for controlling facial expressions of the iCub, and even a flirting robot.

In addition to the topic areas, a series of talks on computational neuroscience was organized by Terry Sejnowski (UCSD) in the second week, and featured presentations by Andrea Chiba, Barry Richmond, Javier Movellan, Steve Zucker, Mark Churchland, John Allman, Tanya Sharpee, and Mike Stryker. The topics covered diverse fields, ranging from biological vision, memory, motor systems, or brain dynamics to neurogenomics, microconnectomics, or computer vision. Together with presentations by the topic area leaders and the invited faculty, and tutorials by Rajit Manohar and Jennifer Hasler on Asynchronous circuits and Field-programmable Analog Arrays (FPAA) respectively, Telluride provided a lot of opportunities for students and senior researchers to learn about the current state-of-the-art in various fields.

Telluride 2012 has clearly shown that neuromorphic systems are becoming more cognitive. Learning is playing an increasingly important role, and this year we have seen quite impressive projects where neuromorphic systems have interacted with the real world in intelligent ways. Using neuromorphic sensors like the DVS has become easier, and the community is constantly improving its repertoire of tools for processing event-based input data in efficient ways. This is of course also facilitated by new hardware for implementing circuits of spiking neurons, and the associated software tools. All the projects had a truly interdisciplinary spirit, which has always been characteristic for Telluride, and projects like “Social Neuroscience” and “Human Attention” have opened doors to new research fields. In particular, this year we have seen great contributions and very fruitful collaboration with the EEG community, which was also reflected by the “Best New Neuromorph” award going to EEG expert Ed Lalor from Trinity College Dublin. So the future of the field looks bright, and we expect this trend to continue, so even more interdisciplinary projects can be expected in the coming years.

A workshop like this is of course not only an important scientific event, but also presents a great opportunity to showcase Neuromorphic Engineering to the general public. Importantly, the “Neuromorphs” every year are being welcomed very warmly by the town of Telluride, and we are trying to give back to this community. Traditionally the Neuromorphs have always been participating in the annual 4th of July Parade, and this year was no exception, with the “Social Robots” finishing second in our award category, thanks to great creativity and efforts by the students. Public lecture events by Ralph Etienne-Cummings and Shihab Shamma, as well as open-doors days for visiting school-kids have provided great opportunities for the public to learn about this exciting new field. In addition, videos have been recorded throughout the workshop, and will be available soon to present the unique Telluride workshop atmosphere to the world, and the Wiki at http://www.neuromorphs.net provides a source of information about the workshop itself and the results of the projects.

The Telluride Neuromorphic Cognition Engineering Workshop is directly financially supported by grants from the National Science Foundation and the Air Force Office of Scientific Research, as well as the European Commission and the organizing universities.

-Michael Pfeiffer