METHODS ARTICLE published: 20 January 2010 NEUROENGINEERING doi: 10.3389/neuro.16.017.2009 Cyber-workstation for computational neuroscience Jack DiGiovanna1*, Prapaporn Rattanatamrong 2, Ming Zhao 3, Babak Mahmoudi 4, Linda Hermer 5, Renato Figueiredo 2, Jose C. Principe 6, Jose Fortes 2 and Justin C. Sanchez 4 1 Neuroprosthetics Control Group, ETH Zurich, Switzerland 2 Advanced Computing & Information Systems Lab, University of Florida, Gainesville, FL, USA 3 School of Computing & Information Sciences, Florida International University, Miami, FL, USA 4 Neuroprosthetics Research Group, University of Florida, Gainesville, FL, USA 5 Department of Psychology, University of Florida, Gainesville, FL, USA 6 Computational NeuroEngineering Laboratory, University of Florida, Gainesville, FL, USA Edited by: A Cyber-Workstation (CW) to study in vivo, real-time interactions between computational Michele Giugliano, Ecole Polytechnique models and large-scale brain subsystems during behavioral experiments has been designed Federale De Lausanne, Switzerland; University of Antwerpen, Belgium; and implemented. The design philosophy seeks to directly link the in vivo neurophysiology University of Antwerp, Belgium laboratory with scalable computing resources to enable more sophisticated computational Reviewed by: neuroscience investigation. The architecture designed here allows scientists to develop new Alessandro E. P. Villa, Université Joseph models and integrate them with existing models (e.g. recursive least-squares regressor) by Fourier Grenoble, France specifying appropriate connections in a block-diagram. Then, adaptive middleware transparently Gediminas Luksys, Basel University, Switzerland implements these user specifi cations using the full power of remote grid-computing hardware. *Correspondence: In effect, the middleware deploys an on-demand and fl exible neuroscience research test-bed to Jack DiGiovanna, Neuroprosthesis provide the neurophysiology laboratory extensive computational power from an outside source. Control Group, Automatic Control Lab, The CW consolidates distributed software and hardware resources to support time-critical and/or Physikstrasse 3, Zürich 8057, resource-demanding computing during data collection from behaving animals. This power and Switzerland. e-mail: [email protected] fl exibility is important as experimental and theoretical neuroscience evolves based on insights gained from data-intensive experiments, new technologies and engineering methodologies. This paper describes briefl y the computational infrastructure and its most relevant components. Each component is discussed within a systematic process of setting up an in vivo, neuroscience experiment. Furthermore, a co-adaptive brain machine interface is implemented on the CW to illustrate how this integrated computational and experimental platform can be used to study systems neurophysiology and learning in a behavior task. We believe this implementation is also the fi rst remote execution and adaptation of a brain-machine interface. Keywords: cyber-workstation, distributed parallel processing, real-time computational neuroscience, brain-machine interface INTRODUCTION supercomputer (IBM Blue Gene) with 10,000 processors to model a In the last decade, there has been an explosion in the number of rat neocortical column (one processor per neuron). Despite current models and amount of physiological data that can be obtained from analytical and computational limitations, Henry Markram suggests either basic neuroscience or computational neuroscience experi- there will be suffi cient computational advances to model an entire ments (Trappenberg, 2002; Purves et al., 2004). The data from these human brain within 10 years. With a ‘complete’ brain model, behav- experiments is beginning to be leveraged to address the problem of ioral complexity could be investigated through simulated interac- reverse-engineering the brain (Markram, 2006). This process has been tions with the environment. An alternative to this approach is to identifi ed by the U.S. National Academy of Engineering as one of directly investigate behavioral complexity by modeling the relation- the 21st Century’s great challenges for engineering and neuroscience ship between behavior and brain subsystems (e.g. single neurons, because it has the potential to: (1) impact human health by providing neural ensembles) in vivo. This modeling may be achieved in a a guide for repairing diseased neural tissue and (2) create innovation typical neurophysiology laboratory given simple models. Notice in neuromorphic systems based on details of brain function. The chal- that both of these approaches only address one area of complexity lenge arises due to both anatomical and behavioral complexity. Human (anatomical or behavioral). brains contain approximately 1015 synapses connecting 1011 neurons Alternatively, it would be useful to provide increased compu- of different classes which are hierarchically self- organized across mul- tational capacity to facilitate real-time modeling of interactions tiple temporal and spatial scales (Abeles, 1991). Furthermore, the between multiple brain subsystems, learning, and behaviors. Such neuronal properties, structural networks, and functional networks interactions may provide insight into existing research questions can adapt based on experience and learning. (e.g. hierarchical vs. connectionist organization) by revealing func- A prominent project addressing anatomical complexity is tional relationships within the brain and between the brain and the Blue Brain Project which combines realistic neuron mod- external world. Recently, increased computation power has been els (Markram, 2006). Currently, experimental data is fed to a harnessed via a multi-core graphics processing unit for parallel Frontiers in Neuroengineering www.frontiersin.org January 2010 | Volume 2 | Article 17 | 1 DiGiovanna et al. Cyber-workstation for computational neuroscience computations (Wilson and Williams, 2009). Additionally, the ability introduces BMI and overviews the CABMI used to demonstrate for researchers to share (and improve) models has proven successful some CW capabilities. The fi nal section discusses implications of in projects like BCI2000 (Schalk et al., 2004) and Physiome (Asai the CW for computational neuroscience. et al., 2008). The Concierge platform at RIKEN provides central- ized experimental results and analysis for researchers around the FUNCTIONAL SYSTEM REQUIREMENTS world (Sakai et al., 2007). Here we aim to incorporate aspects of all The practical embodiment of our CW consolidated soft- those systems into a modular, user-friendly workstation for in vivo ware and hardware resources across three collaborating labs: experiments. We further believe that information technology has Neuroprosthetics Research Group (NRG), Advanced Computing an important role to play in encapsulating many details of the labo- and Information Systems (ACIS) Lab, and Computational ratory work – translating advanced computational neuroscience NeuroEngineering Laboratory (CNEL) at the University of methods into accessible tools for experimental neuroscience. The Florida. The CW enables neurophysiology researchers to setup, Cyber-Workstation (CW) developed here was designed with those carry out, monitor and review experiments locally that require philosophies in mind. powerful remote online and/or offl ine data processing. The initial The CW provides a computational infrastructure for real-time design of the CW was centered around the development of BMIs neural systems modeling and collection (and storage) of mul- where neural activity was used to directly control the position of tiple channels of neural and environmental data during behav- a robotic (prosthetic) arm. The CW was designed to improve the ior. Synchronous modeling and experimentation can provide a effi ciency and effi cacy of the typical workfl ow required by closed- window to functional aspects of neural systems that combine loop experiments. Specifi cally, several aspects of the workfl ow sensory, cognitive, and motor processing while interacting with can be automated, hidden or supported by a cyberinfrastructure dynamic environments. The CW allows a neuroscientist to expand that exposes, through a single point of access, only the needed in vivo paradigms by providing as much computational power functionality for a neuroscientist to conceptualize, conduct and as needed. reason about an experiment. Figure 1 depicts a high-level view The CW was developed around and will be demonstrated via a of what such an infrastructure might consist of. Neural signals Co-Adaptive Brain-Machine Interface (CABMI). Brain-Machine are sampled from behaving animals at NRG and sent across the Interfaces (BMIs) generally create a model of a specifi c brain sub- network for computing. At ACIS (>500 m away from NRG), a system’s (e.g. motor, parietal) interaction with the external world variety of computational models process these data on a pool (Sanchez and Principe, 2007) such that the BMI can decode user of servers; then results are aggregated and used to control robot intention to control a prosthetic. Co-adaptive BMIs go a step movement in real-time at NRG. The architecture of the CW further by engaging both the user and a computer model to learn is described in the next section and has general applicability to control a prosthetic’s movements based on interaction with beyond the specifi c environments of the ACIS, NRG and CNEL the environment (DiGiovanna et al.,