Mol2net Computational Models of the Brain

Mol2net Computational Models of the Brain

Mol2Net, 2015, 1(Section E), pages 1-10, Short Communication 1 http://sciforum.net/conference/mol2net-1 SciForum Mol2Net Computational Models of the Brain L.A. Pastur-Romay1, F. Cedrón1, A. Pazos12 & A. B. Porto-Pazos12* 1 Department of Information and Communications Technologies, University of A Coruña, A Coruña, 15071, Spain; E-Mails: [email protected], [email protected] 2 Instituto de Investigación Biomédica de A Coruña (INIBIC), Complexo Hospitalario Universitario de A Coruña (CHUAC), A Coruña, 15006, Spain; E-Mail: [email protected]; * Author to whom correspondence should be addressed; E-Mail: [email protected]; Tel.: +34-881-011-380; Fax: +34-981-167160. Published: 4 December 2015 Abstract: Different research projects around the world are trying to emulate the human brain. They employ diverse types of computational models: digital models, analog models and hybrid models. This communication includes a summary of some main projects, as well as future trends in this subject. It is focused on various works that look for advanced progress in Neuroscience and still others which seek new discoveries in Computer Science (neuromorphic hardware, machine learning techniques). In addition, given the proven importance of glial cells in information processing, the importance of considering astrocytes into the brain computational models is pointed out. Keywords: brain emulation, neuromorphic chip, neuron-astrocyte computational models, brain computational models 1. Introduction The first computational brain models were algorithms based on neural networks to process created with the goal of reproducing this the information, and Computational extraordinary organ, in order to understand and Neuroscience which seeks to create realistic mimic the way the information is processed, as models of the brain. In the seventies the field of well as its energy efficiency [1-9]. From these Brain Machine Interface (BMI) also emerged, works, basically two scientific disciplines whose purpose was to create systems that emerge: the connectionism branch of Artificial connected the brain directly to an external Intelligence, which is aimed at developing device. At the same time, a branch of Mol2Net, 2015, 1(Section E), pages 1-10, Short Communication 2 http://sciforum.net/conference/mol2net-1 Neuroscience, known as Neuroprosthetics, was the neuron models were very simple and the formed, which sought to build artificial devices simulation was x1542 times slower than in real to replace the functions of nervous systems time [12]. However, it should be pointed out that which are damaged in patients. At the end of the until now in most computational brain models eighties, Carver Mead [10, 11] proposed the the capacity to process the information from the concept of Neuromorphic Engineer to describe other half of the brain, containing 84 billion glial the use of Very Large Scale Integration (VLSI) cells [13], has not been taken in consideration. systems which contained analog circuits to According to the Neural Doctrine, neurons are mimic the neurons. the only cells in the nervous system involved in All these scientific disciplines have tried to information processing, and the glial cells only model the brain in one way or another. Over the play a support role. But over the past two past century, many experts in these fields have decades this theory has started to be seriously predicted that in 10 or 20 years a computational debated. Some discoveries have demonstrated system comparable to the human brain would be the capacity of the glial cells to participate in built. But all these predictions had failed because information processing [14-17]. In this of the technological limitations and the communication, some works focused on underestimation of the brain capacity. Although implementing artificial astrocytes in the brain IBM ran the first simulation with approximately models are referenced. the same number of neurons as the human brain, Figure 1. Brain Model classification. 2. Models Classification under development, such as: digital models, analog models and hybrid models. Classifications of brain models can be This classification is shown in Figure 1. performed from different perspectives. In this • Digital models: they compute information communication, different computer models that using the binary system to simulate and have been classified from the point of view of parallelize the behavior of the brain cells. From signal processing by hardware are currently the software models, the realistic computer Mol2Net, 2015, 1(Section E), pages 1-10, Short Communication 3 http://sciforum.net/conference/mol2net-1 models are first considered, which are those computer architectures based on brain shaping the internal structure of the cells (ion functioning. channels, organelles, etc.) allowing the study of • Analog models: they consist of their functions/operations. The generation of neuromorphic hardware elements where action potentials, activation of neurons, and information is processed with analog signals, that synapse creation are simulated by mathematical is, they do not operate with binary values, as equations implemented in the software, with information is processed with continuous values. specifically-designed tools. In addition, the This allows computation to be more efficient, so connectionist models are taken into account, that analog computation could be used in which, given a known behavior is expected to be applications where energy efficiency is very achieved, such as a classification, object important. recognition in images, regression, etc., allow • Hybrid models: they have been classified searching for a structure of artificial process as such those assembled using hardware elements (neurons and/or astrocytes) that give composed of both analog and digital sufficient rise to such behavior. With regard to components. These models seek to make the digital hardware models, they propose new most of each type of computer. Num. Type of Simulated Projects Name Institution Objectives Duration Refs. neurons neurons synapses Human Brain European Hogdking & 106 5x108 1, 2, 3, 4 2005 18 Project Union Huxley Univ. Leaky integrate- Software SPAUN 2.5x106 1012 1 2012 19 Waterloo and-fire Point neuron models, leaky Univ. 5 7 SpiNNaker 2.5x10 integrate-and- 8x10 1, 2, 3, 4 2005 20 Manchester Digital Models Digital fire, Izhikevich's models Hardware Improved leaky SyNAPSE IBM 1011 integrate-and- 1014 2, 3 2008 21 fire. Adaptive European exponential 2011- BrainScales 4x106 109 22 Union integrate and 1, 2, 3 2015 fire neurons Analog Integrate-and- Univ. Models fire with two HiAER-IFAT California at 250.000 5x106 1, 2, 3, 4 2004 23-27 compartments San Diego for neuron Univ. Hogdking & NeuroDyn 4 12 1 2004 23-27 California at Huxley. 384 Mol2Net, 2015, 1(Section E), pages 1-10, Short Communication 4 http://sciforum.net/conference/mol2net-1 San Diego parameters and 24 channels. Quadratic integrate-and- fire somatic compartment + Stanford Dendritic Neurogrid 106 109 1, 3, 4 2007 28 Hybrid University compartment Models model with 4 Hogdking & Huxley channels BRAIN Qualcomm not public not public not public 1, 2, 3 2013 29 Initiative Table 1. Overview of key features of relevant projects. Objectives (1. Computational Neuroscience; 2. Artificial Intelligence; 3. Neuromorphic chips; 4. Build devices to help disable people). 3. Characteristics of the models Europeans try to increase scientific knowledge about the brain. However, the major American Table 1 shows an overview of key features of projects are rather focused on carrying out a main projects around the world that model the revolution in the computer industry, laying the brain. In this table they are grouped according to foundation for future computer systems. the classification referred to: Number of neurons: the simulation with the Project name: it usually contains words like largest number of neurons was made by the ‘neuron’, ‘spike’ or ‘brain’. SyNAPSE project in 2012 with 5.4x1011 Institution: it is observed that most neurons, a quantity even higher than a human institutions are universities, but there are some brain, which is around 8.6x1010 [13]. It should be projects developed in companies, such as IBM noted that this simulation is not expected to be (SyNAPSE) or QUALCOMM (BRAIN realistic and uses very simplified neuronal Initiative). Most modeling works are coordinated models. Furthermore, the simulation runs 1542 by groups of the prestigious American times slower than real time and 1.5 million universities, like Stanford (Neurogrid). There are BlueGene/Q cores [30] were necessary. also projects coordinated in prestigious European Types of Neurons: there are many types of universities like University of Lausanne (HBP) neuronal models with different levels of realism or University of Manchester (SpiNNaker). The and complexity. These implementations can be projects developed in European universities are either software or hardware-based. When it mainly supported through the European Union, comes to software connectionist models, while in the case of US projects, funding comes artificial neurons are simple processing elements from DARPA and NIH (National Institutes of which operate following sigmoid or threshold Health). The most important difference between mathematical functions [31], although there are European and American projects is that progressively more software models using built- Mol2Net, 2015, 1(Section E), pages 1-10, Short Communication 5 http://sciforum.net/conference/mol2net-1 in

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    10 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us