EDLUT (Event-Driven simulator based on Look-Up-Tables)
The EDLUT (Event-Driven simulator based on Look-Up-Tables) is an advanced tool that allows splitting neural network simulations in two stages:
- Cell behavior characterization. In this first stage, each cell is simulated reiteratively with different input stimuli and from different cell states. This allows scanning the cell behavior which is compiled into look-up-tables. Usually the cell model dynamics are defined with several differential equations that define the cell behavior. Therefore at this stage the tool uses advance numerical calculation methods (such as Runge-Kutta model) to provide estimations of the cell states after receiving a specific stimulus. This represents a massive simulation load for each cell type but once it is done and the results are stored in well structured tables we can avoid numerical calculation during network simulations. Furthermore, simulations can be done adopting an event-driven simulation scheme which accelerates significantly the simulation speed.
- Network simulation towards system behavior characterization. At this stage, we run multiple network simulations with different weight configuration or even including Spike-Time Dependent (STD) learning rules. This stage does not require numerical calculation; the neural states in different times along the simulation are retrieved from the look-up-tables. This allows massive simulations only possible with this kind of advance simulation tool.
More detailed information about EDLUT simulation process can be found in the EDLUT Brief Description or the EDLUT presentation. If you are interested in scientific results using EDLUT, they can be found in Related Publications.
Research Goals
EDLUT is a simulation engine for biologically plausible neural networks. It is a tool for studying the computational principles of neural systems and eventually contributing to reveal how different functionalities of the Brain and Central Nervous System are based on cell and topology properties.
We adopt the attitude of engineers: “I understand how it works when a build it”.
Investigating and creating models of nervous subsystems requires more than the simulation engine itself (EDLUT). It also needs cell models, network models, functional working hypothesis, etc. For this purpose, it is necessary an interdisciplinary research effort with contributions of neurophysiology groups, biological computing, cognitive systems, biology modelers, efficient computing, etc. In SpikeFORCE and SENSOPAC several interdisciplinary cross-enriching collaborations are taking place for building biologically plausible models of neural subsystems such as the cerebellum, Inferior Olive, Cuneate Nucleuous, etc.
Development Team and Collaborators
The original EDLUT has been developed at the University of Granada (Dept. of Computer Architecture and Technology). The main developers are R. R. Carrillo, J. Garrido and E. Ros. The research coordination was performed by E. Ros.
Now EDLUT has been released as Open Source facilitated by the OSL “Oficina de Software Libre” through the advice of J.J. Merelo of the University of Granada. This means that any other development effort can be done by any other member or the research community.
we have collaborated and are collaborating with different research groups such as University of Pavia (Egidio D’Angelo and Sergio Solinas), University of Pierre and Marie Curie at Paris (Angelo Arleo and Jean Baptiste Passot), University of Erasmus (Chris de Zeeuw and Jornt de Gruijl), University of Lund (Henrik Jörntell and Carl Fredrik Ekerot), SICS (Martin Nilson) and other researchers such as Boris Barbour (CNRS), Olivier Coenen and Mike Arnold.
Nevertheless, the final goal of understanting the computational principles of the Central Nervous System and how they are related with cell and topological properties is a medium and long term target which requires a continuous and international effort. Therefore any further collaboration is welcome.
Currently, we are improving performances and usability of EDLUT, so if you think that you could help in this issue, or you only have doubts about how to use EDLUT in your own simulations, don't hesitate to send an email to Jesús Garrido or Eduardo Ros.
Current Status
Currently we have abstracted different cell models such a Granule cell, Golgi cell, Purkinje cell, Hodgkin and Huxley model, etc. Some of them are represented as simple integrate and fire models and some include inherent dynamics (such as active ion conductances) that allow studying how specific cell properties impact system functionality.
We have done system models and networks of several hundred thousand cells (a simplified cerebellum model) on a conventional computer and we have done several hundred million simulations of 5 Kneurons to characterize sub-network dynamics.
Scientific results can be found in the related scientific papers and others currently under revision.
As future work we plan to interface the EDLUT simulation engine with other simulation tools widely used, such as NEURON or GENESIS. Furthermore, we are also interfacing EDLUT with LSAM (Large Scale Analog Model) under development at SICS (Martin Nilson) in the framework of SENSOPAC.
Therefore the important message is that EDLUT allows efficient simulations of medium and large scale networks on conventional computers thanks to the event-driven simulation strategy based on Look-up-tables (of pre-compiled neuron models) which avoid intensive calculations during network simulations.
Further research on this issue and related topics will require intensive collaborations at international and interdisciplinary levels. EDLUT is open software; therefore it can be adapted or further developed by different research institutes. The EDLUT research core group at University of Granada (Eduardo Ros, Richard Carrillo and Jesus Garrido) are open to collaborations along this line.
Acknowledgment
The development of the EDLUT platform has been supported by two EU grants, SpikeFORCE (IST-2001-35271) and SENSOPAC (IST-028056). SENSOPAC stands for Sensorymotor structuring of Perception and Action for emerging cognition.
