Embodied neuromorphic real-world architectures of perception, cognition and action

Members: anna mura, Erik Billing, Brandon Kelly, Boris Duran, Luis Camunas, Jorg Conradt, Daniel Neil, Daniel Mendat, Diogo Pata, Estela Bicho, Greg Cohen, Guillaume Garreau, Giovanni Maffei, Garrick Orchard, Jasmine Berry, Jamal Molin, Jonathan Tapson, Kayode Sanni, Marcela Mendoza, Manu Rastogi, Mathis Richter, Andrew Mundy, Mark Wang, Ernst Niebur, Michael Pfeiffer, Paul Verschure, Sadique Sheik, Stephen Deiss, Sergio Davies, Shih-Chii Liu, shashikant koul, Tim C. Pearce, Timmer Horiuchi, Tony Lewis, Thomas Murray, Tobi Delbruck, Thomas Trappenberg, Terry Stewart, Vikram Ramanarayanan, Xavier Lagorce, Yulia Sandamirskaya

Organizers:: Paul Verschure (U Pompeu Fabra, - ICREA, Barcelona) Andreas Andreou (Johns Hopkins University)

Despite great progress in our understanding of the components of bodies and brains, the system level principles of their integration remain elusive. This is a fundamental challenge because while the structures of the brains and bodies are comprised of many components whose performance strongly relies on large-scale integration ranging from multi-sensor fusion, decision making and action execution to conscious experience. The functionality of the manifold of components of bodies and brains must be consistent with constraints imposed by the physical structures that embody these integrative architectures. This workshop will address the challenge of understanding and building embodied neuromorphic real-world architectures of perception, cognition and action through both presentations, discussion and concrete experimentation.

From an architecture and task point of view the starting point we will focus on is the robot based and biologically grounded Distributed Adaptive Control (DAC) theory of mind and brain ( “2003 verschure nature_425_620.pdf”;” 2012 Verschure DAC overview BICA.pdf”; see Attachments). DAC and the associated simulator IQR was already introduced in Telluride in 1996!. DAC is unique in that it has been explored using robots and mechatronic systems (including Ada) in a range of tasks, links between symbolic and sub/non-symbolic approaches, has been mapped to a number of key brain systems and given rise to novel neurorehabilitation technologies. During the workshop we want to address all of these aspects and more. DAC proposes that the brain is organized as a three-layered control structure with tight coupling within and between these layers. DAC distinguishes the Soma and the Reactive, Adaptive and Contextual layers. Across these layers a columnar organization exists that deals with the processing of states of the World or exteroception, the Self or interoception and action that mediates between the two. A core question thus becomes how the interaction between and across these layers and columns is organized.

From the perspective of physical structures that embody brain-like architectures, the starting point will be the framework developed over a period of years and summarized in: A. S. Cassidy, J. Georgiou, and A. G. Andreou, “Design of silicon brains in the nano-CMOS era: spiking neurons, learning synapses and neural architecture optimization,” Neural Networks, vol. 45, pp. 4–26, Jun. 2013. Our approach to the design of spiking neurons and STDP learning circuits relies on parallel computational structures where neurons are abstracted as digital arithmetic logic units and communication processors. Using this approach, we have developed arrays of silicon neurons that scale to millions of neurons in a single state- of-the-art Field Programmable Gate Array (FPGA). We demonstrate the validity of the design methodology through the implementation of cortical development in a circuit of spiking neurons, STDP synapses, and neural architecture optimization.

Over the three weeks of the workshop we will address a number of questions and challenges that pertain to the mapping of DAC to both the brain and artefacts. Conversely, we will ask how brain and machines inform the advancement or falsification of brain theories such as DAC and their derived models. By starting from this concrete example we aim to identify the key organization principles of brains and neuromorphic technologies. The details of the programme will also depend on the inputs on the participants.

The project will evolve around both theoretical and conceptual explorations and concrete hands-on experimentation using robot based neuromorphic architectures for learning, navigation and decision-making that will be provided by the organizers. We will kick start the project work with tutorials on the IQR simulation environment and DAC.

Relevant tutorials can be obtained from the Convergent Science Network  http://csnetwork.eu:  http://csnetwork.eu/request-form-iqr-simulator-for-large-scale-neural-systems The DAC tutorial can be downloaded from the Workshop file server.