US-Belgium workshop: Atomic Switch Networks for Neuromorphic Reservoir Computing; Late Fall-2015/Early Spring 2016; University of Ghent-Belgium.

University Of California-Los Angeles, Los Angeles CA

Investigators

Abstract

NSF CNIC proposal #1444214 US-Belgium Workshop: Atomic Switch Networks for Neuromorphic Reservoir Computing Part 1: The human brain outperforms digital computers in a number of tasks such as image, motion tracking and sound recognition and decision making in complex and often noisy and error prone environments. Digital computers are by their nature poorly-suited for tasks such as autonomous control (navigation, robotics), pattern recognition (speech, vision) or prediction (weather, financial markets). A biologically inspired approach to computing, called Reservoir Computing (RC), on the other hand, has demonstrated the potential to perform complex tasks efficiently. To perform RC, a newly developed hardware platform called the Atomic Switch Network (ASN) uses nanotechnology to create billions of synthetic synapses wired up in a fashion similar to that of the neocortex in the human brain. The implementation of a functioning RC-ASN system requires the collaborative expertise from recognized world leaders in RC methods at Ghent University, Belgium and the UCLA team who have developed the ASN device. UCLA has proposed a participant-driven workshop involving invited lectures, hands-on tutorials with hardware and software and breakout discussions with the goal to accelerate realization of this new form of computers system. This workshop will provide international research opportunities to 5 US students and early career researchers, while also promoting team-building skills, student-driven collaboration, and cultural exchange. By combining concepts from nanoscience, neuroscience, and machine learning, this proposal seeks to leverage the collective expertise of all parties to advance this next-generation cognitive technology. The successful outcomes of this research will also benefit the BRAIN Initiative, which is a priority research area of the U.S. Part 2: Atomic Switch Networks (ASN) are a unique class of biologically inspired computing architectures designed to produce a complex, dynamical system through the collective interactions of functional nanoscale materials. These self-organized devices retain the intrinsic memory capacity of their component resistive switching elements while generating a class of emergent behaviors commonly associated with biological cognition. Their capacity for non-linear transformation of input information, which is processed and stored in a distributed fashion, generates patterns of dynamic spatiotemporal activity that can be used as the basis for a computational platform. Recent efforts to model, simulate, and measure the operational dynamics of ASNs toward hardware implementation of reservoir computing (RC), a burgeoning field that investigates the computational aptitude of complex biologically inspired systems to address problems in which data is constantly changing, incomplete, or subject to errors, indicate the necessity to establish a collaboration with experts in the field of machine learning. The combined expertise of proposed workshop participants will focus on a critical assessment of how to best utilize ASN devices to overcome current operational limits on real-time signal processing in the RC paradigm such as speed, network density, and scalability. Beyond lectures and discussion sections, tutorial workshops delivered by participants from the US and EU will be utilized to disseminate/demonstrate the current status of (1) modeling/simulation of ASNs, (2) physical implementations of ASNs, and (3) physical implementations of other hardware systems (memristors, optoelectronics, etc.). Targeted outcomes include identification of specific areas for near-term collaboration and follow-on funding within existing Core programs at the NSF. This new collaboration will provide a tremendous opportunity to explore the best-case scenario resulting from the world's leading RC research with a potentially groundbreaking platform for hardware-based RC to contribute to novel approaches in real-time information processing and computation.