Deep learning brought a lot of advancements in artificial intelligence, but have limitations in terms of learning from small data, handling uncertainty and model selection. We develop Bayesian deep learning models which could overcome these limitations using Bayesian learning idea. We work on developing efficient inference techniques for Bayesian deep learning models. On the applied side, we work on developing Bayesian deep learning models for problem arising in autonomous driving vehicles.
We work on bringing Bayesian principles to Large Language Models (LLMs), allowing them to be more safe and robust by bringing uncertainty modeling capability in them.
Research focuses on designing continuous depth deep learning models inspired from differential equations. Treating the computation of intermediate feature representation in deep learning models as a solution to differential equations has tremendous advantage in terms of model selection and reduced number of parameters. This field is in an early stage with lot of promises, we focus on developing various differential equation based neural networks for machine learning problems.
Humans learn continuously over time, adapting and learning new tasks. However, deep learning models exhibit catastrophic forgetting when trained on a sequence of tasks. We work on developing deep learning models, which can overcome catastrophic forgetting through generative replay and Bayesian learning techniques such as variational continual learning. We work on continual learning problems in both vision and natural language processing.
We also work on causal learning teachniques which allows the model to reason and become more robust.
Research focuses on developing generative models such as generative adversarial networks , variational autoencoders and normalizing flows , which can model the distribution of data and generate samples from them. We work on overcoming the limitations of the existing generative models and develop them for problems in vision and language.
Bayesian non-parametric models allow one to learn rich and flexible models due to their non-parametric nature and allow the model complexity to be determined by the data. This helps to overcome the problem of model selection to a great extent and helps to model uncertainty better. Current work focusses on developing Bayesian non-parametric models such as Gaussian processes and efficient inference algorithms for them for various learning problems. We develop deep learning models such as deep Gaussian processes, and develop efficient inference algorithms for them. They can generalize from small data and can automatically determine model architecture to a great extent.
Online social networks provide a platform for sharing information at a massive scale. In order to leverage this data, we work on various problems arising in social media like Twitter and location based social networks such as information diffusion, user activity modeling , semantic annotation etc. We develop models based on point processes such as Hawkes process and graph neural networks to address these problems. We have also worked on problems like rumour detection, disaster management, and topic detection and tracking in social media. We also work on analyzing and solving problems arising in community question answering sites such as stack exchange.
With the widespread use of AI across various domains, we use machine learning and deep learning models to solve various problems related to astrophysics such as Redshift prediction, and Galaxy morphology classification.
We also work on poblems from mechanical engineering where we develop physics-informed neural networks for efficiently solving flow equations such as wave equations and Navier stokes equations.
The project seeks to assist urban local bodies and government agencies in monitoring, analyzing, and planning the sustainable growth of urban infrastructure and resources using state-of-the-art AI solutions and services. While there is increased availability of data due ro digitization efforts, there is a lack of corresponding user-friendly AI tools for infrastructure/resource monitoring and analytics. We fill this gap by develop AI based solutions and technologies for several problems under urban governenance such as Mobility and transportation, air pollution, waste management etc.
We can use machine learning to solve various real-world problems. One such problem is traffic data management. We use various machine learning and deep learning algorithms to understand and model traffic congestion and hence manage it efficiently. We also work on analyzing and developing models for solving problems arising in several other domains such as Telemetry data, ESG data, and time-series data.
AGC funded project on scientific Large Language Models
Sony research grants for Continual learning in deep neural networks.
SERB core research grant for continual learning for vision and language processing.
Unrestricted research grant from Accenture Technology for learning from alternative data sources.
Nvidia sponsored project on Bayesian deep learning for computer vision and autonomous driving.
DST funding for the project Towards Developing Next-generation Deep Learning
DST funding for the project Machine Learning for Astrophysical Data Analysis
JICA CKP funding for Bayesian Deep Learning in collaboration with Dr. Emtiyaz Khan from RIKEN Centre for Advanced Intelligence Project, Tokyo.
DST SERB funding for rumour detection in social media
Smart Cities for Emerging Countries Based on Sensing, Network, and Big Data Analysis of Multimodal Regional Transport System (M2Smart) (Funded by JICA / JST SATREPS, 2017-2022)
Projects under the Technology Innovation Hub for Autonomous Navigation (TIHAAN)