About
Our research group, Lab 1055 (find out why it is called so at https://lab1055.github.io! follow us on X at https://twitter.com/lab1055), works at the intersection of the theory and application of machine learning - with a focus on computer vision and more recently, multimodal models. With a strong interest in the mathematical fundamentals and a passion for real-world application, our group aims on being at the forefront of the field, by carrying out impactful research in the areas of deep/machine learning, computer vision and multimodal learning, guided by application contexts derived from real-world use.
Keywords: Deep/Machine Learning, Computer Vision, Multimodal AI Models, Reasoning/Explainable AI
Our problems of interest in recent times have focused on addressing foundational and applied problems in multimodal foundation models, including their reasoning and safety. We are also deeply interested in envisioning newer directions in AI research that could have downstream impact on large-scale models.
For many years until now, we focused on learning reliable and robust AI/ML systems in ever-evolving environments, where we addressed problems such as:
Explainable and robust machine/deep learning: Problems on explainable AI (more recently, focused on ante-hoc inherently interpretable methods in lieu of post-hoc explainability), use of causality in machine learning, model robustness, compositionality in deep learning models
Organic lifelong learning: Learning continuously in evolving environments with whatever data and labels are available at hand, including settings such as continual learning, open-world learning, domain adaptation/generalization, zero/few-shot learning, active learning, and even the amalgamation of these settings that could organically arise in the real world. We continue to be interested in how these settings apply to large-scale foundation models of today.
Some of the application domains where we have applied our research in the past include (not as actively these days):
Agriculture: Plant phenotyping using computer vision
Drone-based vision: Detection of objects from drone imagery, as well as low-resolution imagery
Autonomous navigation: adding levels of autonomy to driving vehicles in developing countries, focusing on India
Human behavior understanding: Detection of emotions, human poses, gestures, etc of the human body using images and videos
This interview featured in the IEEE Signal Processing Newsletter (Feb 2021 issue) also describes our (slightly older) research efforts.
Completed/Past Projects (Selected)
Exploring Connections between Adversarial Robustness and Explainability (Google Research Scholar Award, Microsoft Research Postdoctoral Research Grant)
Learning with Limited Labeled Data: Solving the Next Generation of Machine Learning Problems (DST-JSPS Indo-Japan Collaborative Research program)
Learning with Weak Supervision for Autonomous Vehicles (Funded by Intel and SERB IMPRINT program)
Explainable Deep Learning (Funded by Adobe)
Deep Generative Models: Going Beyond Supervised Learning (Funded by Intel)
Towards Next-Generation Deep Learning: Faster, Smaller, Easier (Funded by DST/SERB ICPS program)
Object Detection in Drone Images (Funded by MeitY, MoE)
Explainable Machine Learning (Funded by MHRD, DST and Honeywell through the MHRD UAY program)
Deep Learning for Agriculture (A DST-JST SICORP Collaborative Project with Univ of Tokyo, IIT-B, IIIT-H, and PJTSAU)
Non-convex Optimization and Deep Learning (Funded by Intel PhD Fellowship and SERB MATRICS program)
Object Detection in Unconstrained Settings and Low-resolution Thermal Images (Funded by DRDO and IBM)
My even-older-past research focused on the use of machine learning and computer vision in assistive technology applications. Please see this link for some of these directions.
Funding
We are grateful to the following organizations whose support sustains our research.
 |