The Silicon Review ^Asia

The new promising technology developed by IIT researchers can be useful to bring neural networks in portable devices. Researchers from IIT Hyderabad have created a new method which can help to install ANNs (high volume artificial neural networks) on portable devices and smartphones.

The mechanisms of the artificial neural networkwork like the human nervous system which is made up of problem-solvingconnected elements.

But they require a lot of storage space and huge computations. Therefore, if these networks are compressed to adjust to available space on devices can hamper the output quality. The new method could solve the problem.

Currently, the researchers have patented their design and few industries are using it.

As per a member from the research team, the new method can be applied in the cores of mobile platforms. The neural networks are made up of numerical files and are stored in binary form in devices. The new method assists by assembling binary files into fragments of a definite fixed length. The process helped the researchers to produce repetitions of BIT patterns of certain frequencies. The team have attained a compression of about 64% and a minimum single model decompression time of 0.33 seconds.

The researchers have showcased their collective efforts towards the new development at the International Symposium on Circuits and Systems in Italy. The Science and Engineering Research Board partiallyfunded the research work.