We learn from our alumni in this interaction series, often technically, sometimes semi-technically.

Ujjal Kr Dutta SatSure Analytics

Dr Ujjal Kr Dutta is a Manager, Data Science at SatSure Analytics, where he leads efforts in satellite remote sensing, machine and deep learning, to deliver actionable insights across various sectors, such as agriculture, infrastructure, finance, etc. Prior to this, he has worked as a Staff Research Scientist-Manager at Dolby Research Labs, and as a Lead Data Scientist at Myntra (Flipkart-Walmart Group). He has held positions / collaborated with institutions such as University College London, National University of Singapore, MBZUAI, Monash University, IIT Madras, and IIT Guwahati. He obtained his PhD from the Department of Computer Science and Engineering at IIT Madras, where he worked with Dr Chandra Sekhar C, along with Dr Mehrtash Harandi from Monash University/ Data61-CSIRO.

His research interests lie primarily in applied machine learning, while spanning across representation, semi-/self-supervised learning, domain generalization and incremental learning, graph/subspace clustering, differential geometry, Physics-Informed Neural Networks, model compression, object detection, image segmentation, sequential models, etc. He has 17+ publications across AAAI, NeurIPSw, ECCVw, ICASSP, IEEE TAI, etc, and also reviews regularly for the mainstream conferences (CVPR, ECCV, etc) and journals (IEEE TNNLS, IEEE TGRS, etc). He is also a recipient of the Myntra AIM HIGH Award, IBM Best PhD Thesis Award, and IITM Institute Research Award. In addition to his academic and professional achievements, he is an accomplished athlete, with various medals in Olympic-style Weightlifting, Powerlifting, including an Inter-IIT silver medal, and four-times Strongman at IIT Madras.

Self-supervised learning has emerged as a powerful paradigm for learning rich representations from data without the need for manual annotations. In this talk, we embark on an odyssey through various domains, showcasing how self-supervision can be leveraged to tackle challenging problems in computer vision and remote sensing.

We begin our journey by exploring self-supervised learning on natural scene images, demonstrating how pretext tasks can learn transferable features. We then delve into the realm of e-commerce, where we utilize self-supervision to learn representations from product images, enabling effective product retrieval and recommendation. Next, we venture into the domain of remote sensing, where we showcase the application of self-supervised learning to satellite and aerial imagery.

We present pretext tasks tailored for remote sensing data, such as multi-modal fusion and change detection, and demonstrate their effectiveness in learning representations that capture the unique characteristics of remote sensing imagery. Throughout our odyssey, we highlight the importance of designing appropriate pretext tasks for each domain and discuss strategies for adapting self-supervised learning techniques to specific challenges.

Organizers