Close
Biotechgate
| |

Home Page

Action required: Please refresh your browser

We have recently implemented some changes that require a hard refresh of your browser: Please hold down the CTRL-key and press the F5 key.
After a successful hard refresh, this message should not appear anymore.

More details about this topic are available here »

King Abdullah University of Science and Technology Sensors Lab Announces Breakthroughs Published on Brain Inspired Computing
By: PR Newswire Association LLC. - 12 Aug 2015Back to overview list

THUWAL, Kingdom of Saudi Arabia, Aug. 12, 2015 /PRNewswire-iReach/ -- Quality research in the area of artificial intelligence is always in high demand in many diverse circles.  Adding to this pool recently is a team led by Professor Khaled Nabil Salama of King Abdullah University of Science and Technology, Saudi Arabia recently had their research and approach on "Brain Inspired Computing: Memristors Empower Spiking Neurons with Stochasticity" published in the June 2015 issue of the well-respected IEEE Journal on Emerging and Selected Topics in Circuits and Systems on Solid-state Memristive Devices and Systems (DOI 10.1109/JETCAS.2015.2435512) and presented at the 7th International IEEE EMBS Neural Engineering Conference (NER'15), Montpellier, France (DOI 10.1109/NER.2015.7146633).

Photo - http://photos.prnewswire.com/prnh/20150812/257873

The research published explores the difficulty building rational agents to solve different AI search, optimization, and inference problems, and especially in  high-level cognitive tasks, such as natural language understanding, which do not have precise formalizations into concrete optimization or inference problems.  It then goes on the provide potential solutions to this problem using the brain as a model and the technology of memristors and stochasticity as a potential solution. "Real world problems are extremely hard to solve in a brute-force manner," commented Professor Salama.   "This is often the case when there is no algorithm available that can guarantee finding the global optimum for a problem."

Mauran Al-Shedivat, lead author of the work continued, "In other words, we do not know what exact problem we should formally pose and solve to enable a machine to understand human languages as humans do.  This implies that even if we had a computer of unlimited capacity, we still would not know how to build machines of human-level intelligence." Despite these difficulties, the team's research and paper continues on to explore one mechanism, stochastic neural behavior, that the biological system, the brain, uses to perform cognitive tasks and acting as a clear inspiration for AI research that would like to duplicate the brain's abilities in this area.

"The two general questions we need to ask here are: (i) what kind of models can describe a neural learning system built of stochastic components? and (ii) how can we implement such systems efficiently?," remarked Gert Cauwenberghs, co-author of the work, professor of bioengineering and co-director of the Institute for Neural Computation at the University of California, San Diego. "Recent theoretical advances highlighted that networks of stochastically spiking neurons can implement energy-based models for probabilistic inference and deep learning that offer powerful performance on demanding AI tasks. Here, we show how to implement such neural models very efficiently in a brain-like manner."

The authors proposed a circuit-level implementation of an inherently stochastic spiking neuron that follows a specific behavioral model. The proposed neural circuit relies on the stochasticity of the memristor—a nano-scale multi-state device that can switch between its states and change its resistive properties in a non-deterministic fashion. "Our model was capable of exhibiting deterministic behavior along with sharp, spontaneous state jumps with a probability that depended on the applied voltage," explains Rawan Naous, a co-author of the work. 

An implementation of a neuromorphic neural soma that integrated the synaptic input and triggered spike generation using memristive switching is also proposed in the research paper. The spiking behavior of the memristive neuron was akin to the stochastic firing intensity of the SRM. Finally, a probabilistic winner-take-all network was implemented and demonstrated with simulations that a simple stochastic neuron can be used for building spiking neural networks for probabilistic learning. 

The team concludes in their research paper that this constitutes an important step toward scalable and efficient probabilistic neuromorphic platforms, capable of solving cognitive perception tasks.

For more information on the King Abdullah University of Science and Technology Sensors Lab be sure to visit http://sensors.kaust.edu.sa.

Media Contact: Khaled Nabil Salama, King Abdullah University of Science and Technology, +9668084420, khaled.salama@kaust.edu.sa

News distributed by PR Newswire iReach: https://ireach.prnewswire.com

SOURCE King Abdullah University of Science and Technology

Copyright 2015 PR Newswire Association LLC. Back to overview list
to the top ↑