PG Level Advanced Certification Course in AI and MLOps

Domain: Food / Wellness industry

Techniques: CNN based Image segmentation

Overview and Problem Statement:
Worldwide, obesity has nearly tripled since 1975. In 2016, more than 1.9 billion adults, 18 years and older, were overweight (WHO sources). In such a situation, documenting dietary calorie intake is crucial to manage weight loss. Food image segmentation is a critical and indispensable task for developing health-related applications such as automated estimation of food calories and nutrients as a means for dietary monitoring. One of the challenges in this area is the improvement of accuracy in dietary assessment by food image analysis. However, how to derive the food information (e.g., food type and portion size) from food images effectively is a challenging task and an open research problem. In this project, a model that can segment the food components present in an input food image will be built and an application that can predict the food class and the food portions from it will be designed.

Domain: R&D (Computer Vision)

Techniques: Generative Adversarial Network

Overview and Problem Statement:
Estimating a high resolution (HR) image from its low-resolution (LR) counterpart is referred to as super-resolution (SR). Super-resolution is a task concerned with upscaling images from low-resolution sizes such as 90 x 90, into high-resolution sizes such as 360 x 360 (upscaling factor of 4x). Recovering the finer texture details of an image while achieving super resolution at large upscaling factors has received great attention in the computer vision research community. Super-resolution holds great importance in recovering photorealistic details lost to resolution effects and can be applied to a number of real life tasks such as CCTV image enhancement for identification of criminals, restoration of old family photos, medical imaging data enhancement, turning a smartphone camera to an SLR, autonomous vehicle vision enhancement etc. The goal of this study is to acquire high resolution images from low resolution images by training a super-resolution generative adversarial network (SRGAN) on the CelebA Dataset, that contains images of celebrities.

Domain: Online Knowledge Platform

Techniques: NLP, Machine Learning

Overview and Problem Statement:
Quora is a platform to gain and share knowledge on any topic. It allows people to ask questions and connect with people who contribute unique insights and quality answers. This equips people to learn from each other and to understand topics in diverse subjects. Over 100 million users visit Quora every month, and a lot of them inadvertently ask replicate questions, worded differently by different users. Multiple questions with the same intent can cause seekers to spend more time finding the best answer to their question. This also leads to an inefficient system for writers, as they spend time on answering multiple versions of the same question. The goal of this project is to use NLP, ML algorithms and create an application to classify whether Quora question pairs are duplicate or not. This will be instrumental in providing high quality answers and an improved experience for Quora writers and seekers.

Domain: Healthcare – Cancer diagnostic

Techniques: Classical Machine Learning, Deep Learning

Overview and Problem Statement:
Breast cancer is the most common cancer among women worldwide, accounting for 25 percent of all cancer cases and has affected 2.1 million people in 2015. Breast cancer can be categorized into (i) Benign (noncancerous, non life-threatening) tumors and (ii) Malignant (Cancerous, rapidly invading nearby tissue) tumors. Identification of the benign and malignant conditions is normally achieved by imaging of the targeted area of the body using mammography, computed tomography, MRI, ultrasound etc. which aids in diagnosis by physicians. Identifying benign or malignant tumors through machine learning techniques can provide dramatic improvement in diagnostic accuracy. For this project, the Breast Cancer Wisconsin dataset consisting of fine-needle aspirate (FNA) digitized image feature measurements will be used. The final goal is to use the data to train machine learning classifiers (eg. KNN, SVM, Decision tree, Random Forest, Ensemble Learning) and deep neural networks (CNN), and accurately identify benign vs malignant breast cancer with high accuracy.

Domain: Surveillance & Security

Techniques: CNN, RNN

Overview and Problem Statement:
Applications such as surveillance, video retrieval and human-computer interaction require methods for recognizing human actions in various scenarios. In the area of robotics, the tasks of autonomous navigation or social interaction could also take advantage of the knowledge extracted from live video recordings. Typical scenarios include scenes with cluttered, moving backgrounds, nonstationary camera, scale variations, individual variations in appearance and cloth of people, changes in light and view point and so forth. All of these conditions introduce challenging problems that can be addressed using deep learning (computer vision) models. The given dataset consists of labeled videos of 6 human actions (walking, jogging, running, boxing, hand waving and hand clapping) performed several times by 25 subjects. The goal of the project is to classify the actions performed in the video frames.

Domain: Information Security Systems

Techniques: ML Engineering

Overview and Problem Statement:
One of the core components of information security systems is the anomalous user behavior detection, that aids in the detection of intrusion, insider threat and authentication system break. Anomalous behavior based alarm to the system administrator can be combined with other information to determine whether it constitutes an unauthorized or malicious use of a resource. The goal of this project is to develop an effective anomalous user behavior detection system by applying the Isolation Forest algorithm to an enterprise dataset.

Domain: Energy

Techniques: ML Engineering

Overview and Problem Statement:
Maintaining optimal energy efficiency is one of the key areas of focus in residential buildings, offices and factories. By considering features like energy prices, equipment maintenance, labor costs and inventory, ML algorithms can schedule the most appropriate time to perform energy-intensive activities. This can enable enterprises to maximize cost savings by running the right processes at the right time. The goal of this project is to develop a machine learning based application to track energy usage patterns over time so as to reduce the amount of wasted energy and thus perform cost savings.