To stay relevant and ahead of industry, INSOFE faculty actively involves itself in applied research, consulting and product development through INSOFE Labs. Current focus areas include pattern extraction, text analytics, image mining using deep learning and blockchain analytics. Currently, these are applied to retail, healthcare and manufacturing verticals.
Dr. Kishore Konda and Dr. Suryaprakash Kompalli, use both traditional computer vision techniques and deep learning techniques to solve a variety of computer vision problems. INSOFE has a specialist high-performance computing environment that enables the faculty members to conduct cutting-edge applied research. Their current work includes
Wound detection algorithms for measuring the size of the affected zones
Security and Video Surveillance
INSOFE is working on developing a video surveillance tool with enhanced smart capabilities. The tool identifies the violation and sends out instant automated response without requiring any manual interference. Since the current process involves manually going through the footage and checking for violations, it is not only a time-consuming process but also requires manual hours and effort. The smart video surveillance tool makes the entire process automated with an Embedded Machine Learning chip.
Dr. Praphul Chandra and Dr. Suryaprakash Kompalli are working on blockchain technology to develop smart contracts on highly secured blockchain data. They are exploring various applications in financial and supply chain sectors. They are developing a proprietary product that utilizes the above-mentioned technologies to solve problems in financial industry.
Dr. Anand Jayaraman, Dr. Manoj Duse, Dr. Gnana Bharathy and Dr. Parag Mantri work with structured Big Data sets and text and IOT to extract hidden patterns and to build mathematical models that provide sufficient clues about future. Their goal is to help practitioners make better decisions and create expert systems where experts are not available. Some of their current and recent works include
Ship from Store Optimization for a World’s leading Retailer
Helping a start-up provide comprehensive demand forecasting and loyalty program design for mom-pop stores
Helping a start-up develop early care alerting systems from clinical data. This is being conducted as a clinical trial for a global hospital
Helping one of the largest car manufacturers develop demand forecasting solution for over 200,000 components across 7 countries
Deep Learning is disrupting multiple disciplines such as Medical diagnosis, e-commerce, and security. It is the key technology behind several revolutionary applications like autonomous cars, Google’s image & voice recognition, Facebook tagging and Netflix & Amazon’s customer preference analysis.
INSOFE’s team of mentors and data scientists are involved in extensive research and product development activities which in-turn feed as inputs for the program curriculum. Students and interns at INSOFE acquire hands-on expertise while working on such interesting applications.
Meet ARIA, AI engine being developed by INSOFE students and interns under the mentorship of INSOFE faculty.
ARIA- Artificial to Real Intelligent Agent is INSOFE’s computerised personal assistant created by INSOFE interns and students. She is trained to perform multiple tasks using artificial intelligence.
ARIA's full potential is a matter of time but, for now, ARIA can create music, style images and can also detect mood.
Training your own AI agent to compose music for you? Isn’t it cool to create a program that can compose music on its own? Well it is, and that’s what ARIA does. ARIA uses Long Short-Term Memory cells (LSTM) in Recurring Neural Network (RNN) model to predict new music from a set of sequence that has been fed into it. RNN which is specialised for sequence inputs is fed with a lot of piano music node by node which in turn generates new piano music on its own.
A wide range of music can be generated through this model depending on the input data that has been fed into the model. So far, INSOFEans have been successful in generating piano music through ARIA. In near future, the model will be trained to generate music using more than one instrumental music.
Once we are able to train the model to generate music from a set of sequence data using multiple instrumental music and lyrics, the model can act like a real music composer which is capable of creating background scores, songs and cater to various needs of the entertainment industry.
Style transfer is an artistic rendition of images. ARIA is trained for style transfer wherein, it captures contours from original image and juxtaposes with the style of the style image and creates an artistic image out of the two. For this function, Convolutional Neural Network model is used to transfer a style from one image to the contours of another to generate a new image using two input images.
Have a look at the style transfer done by ARIA –
Ever wondered if it is possible for a computer to create a brand-new image without any human involvement? Yes, it is, this is what ARIA is capable of. ARIA is trained using Generative Adversarial Neural Network model wherein a lot of images are given as input and it creates a new image on its own based on those thousands of images that have been fed into the model. Which means, ARIA can create images of humans that do not exist as well!
Not in a good mood? ARIA will cheer you up! ARIA is trained through Convolutional Neural Network with thousands of images to detect the mood of each image, where it can read if the image is of a smiling face or a sad face and will cheer you up if you are not smiling. INSOFEans have used CNN model along with some basic computer vision that enables ARIA to detect mood.
Once you undergo PGP in Big Data Analytics and Optimisation at INSOFE, you will get a basic understanding of how these programmes work and you will be able to create these programs and applications on your own. A hang of these concepts lets you augment these functions and gives you an edge to create interesting new products that haven’t been invented yet.