Hi, I'm working at Fos4X, as a Data Scientist. A Computer Science graduate from IIIT Delhi; one of the reputed research-oriented institutes in India, I’m primarily interested in automating things to make human lives easier. I enjoy working with data to make data-driven decisions.
Working as a data scientist at fos4X (rotor-blade sensing in wind-turbines) in Analytical Software and Model(ASM) team to provide smart monitoring solutions to the customer.
Worked at the R&D segment of the fourth largest mobile network operator in the world. I joined as a Trainee in Data Science Team, and worked on a high-impact project; Segmentation of over 300 Million customers in order to perform experiments on relevant customer group, to make data-informed strategies and improve our customer journey. I also contributed to Network Site Planning, to predict the next most suitable location for cell tower/site.
Interned as an AI Engineer at Digital Product School (a program by UnternehmerTUM) as a part of a cross-functional, international, culturally-diverse team wherein we learned to apply Agile development principles (and design thinking) to build our Minimum Viable Product(MVP).
Our product lets foreign tourists experience a city based on their interests through personalized tours guided by an autonomous (Level 4) car. We also generated useful insights from user interviews & product testing.
Company sponsors: Audi, BMW and ADAC. Reference: Dr. Afsaneh Asaei (AI head of UnternehmerTUM)
Predict time series data on national level and map it to all the states. Achieved state-of-the-art results. (Journal in writing)
Predicted change in contribution and productivity of STEM workers in the financial sector using NLP models. Reference: Dr.Christos
Worked with Prof. Prithwiraj Choudhury from Harvard Business School on the analysis & predictions of salary variations w.r.t. country and companies.
Teaching Assistant for a class of 250+ students. I was responsible for:
Making questions for Quizzes and Assignments.
Checking the answer sheets (Assignments and Exams).
Taking tutorials once a week for practice.
Assisting the visually impaired student.
Worked on three major research projects:
Attribute Classification in Low Resolution: Inspired by real-life datasets involving as small as 32*32 face image resolution, I used Multi-Task Convolutional Neural Networks to improve accuracy of the state-of-the-art models in Attribute Classification.
Morphing using GANs: Created a new architecture for Generative Adversarial Network to take in two face image inputs instead of noise vector and generate a morphed image of the two.
Fooling Presentation Attack Detection Algorithms: Work described in the publications section.
Our proposed research uses adversarial perturbations to attack the Presentation Attack Detection algorithms at feature level via transformation of features from one class (attack class) to another (real class). The feature tampering network utilizes convolutional autoencoder to learn the perturbations. We increased the EER from 20.1% to 55.7% with our attack.
Explained the impact of image pre-processing using image transformations on Presentation Attack Detection algorithms wherein I increase the error of the state-of-the-art PAD algorithms by 15-30% for all datasets.