Breakthrough clinched by joint research with German institutes
As the world grappleswith measures to stamp out the Corona virus, Indian Institute of Information Technology and Management-Kerala (IIITM-K), in a joint research project with German institutes, has made a breakthrough in developing-Analogue Integrated Circuits for implementing Generative Adversarial Network (GAN) that can possibly be used to analyse and interpret 2019-nCoV data for a possible solution to the global pandemic.
The development in this critical area has been achieved by joint research conducted in collaboration with Analogue Circuits and Image Sensors Lab, Seigen University and Fraunhofer, Germany, and Centre for Excellence in Artificial General Intelligence and Neuromorphic Systems (neuroAGI), IIITM-K.
Artificial Intelligence (AI) researchers in various countries, including India, are using this new age technology. AI researchers are using advanced neural networks such as GAN for discovering, analyzing, interpreting and visualizing COVID-19 virus data through computing and molecular structure discovery.
Generative Adversarial Networks (GAN) are neural networks that can be trained to produce or generate new content. These are popularly used for generating fake images, fake audio, fake text, and fake videos.
“We have been able to achieve a breakthrough by this complex and painstaking AI circuits research, with being able to accelerate and run GAN applications in low power devices” said Dr. A. P. James, Professor at School of Electronics of IIITM-K.
The paper, ‘Memristive GAN in Analog’, has been made available online from Friday (April 3) in Scientific Reports journal of Nature. In this collaborative work, the lead author is O. Krestinskaya while B. Choubay, Professor at Seigen University, is the co-author. The experts point out that the GANs have recently succeeded in generating new molecular structures that can bind with the 3C-like protease, which is one of the most important COVID-19 protein targets, and inhibit virus functioning. However, these GANs are notoriously difficult to train, with training usually being a time-consuming and computationally intensive process.
In a new study, energy and area-efficient analogue circuit implementation of such GAN architecture were demonstrated to generate realistic-looking images.
Researchers from Analogue Circuits and Image Sensors Lab, Seigen University and Fraunhofer, Germany and Centre for Excellence in Artificial General Intelligence and Neuromorphic Systems (neuroAGI), Indian Institute of Information Technology and Management (IIITM) – Kerala have developed analogue integrated circuits for implementing GAN with CMOS and nano-scale devices. Complementary Metal Oxide Semiconductor (CMOS) is a technology that is used to produce integrated circuits.
Authors of a research paper point out that the analog GAN with CMOS-memristor crossbar circuits can be implemented with the low on-chip area and be energy-efficient, and can accelerate the discovery of GAN-based applications. The analog domain circuits are capable of being integrated near to image sensors as a low power near sensor AI computing solution or be used as a co-processor for speeding up molecule structure discoveries for developing vaccines.