I am currently a member of the SmartThings Team at Samsung Research , Seoul, where my work involves: 1) developing a deep learning-based indoor positioning system and 2) vectorized floorplan reconstruction from partial/noisy LiDAR maps. I am also collaborating with Prof. Sanja Fidler at Vector Institute to improve object-centric representations by using image context in a self-supervised framework. I have also worked on ways to incorporate geometrical cues to learn dense representations for scene images.
I work at the intersection of scene understanding, self-supervised learning and robotics. Some of the problems that interest me are - learning actionable representations from scene images, understanding scenes at different levels of abstraction (from pixel level up to the image level) and the hierarchical relationships among them, and the challenges in working with 3D scenes. I am also interested in self-supervised approaches that jointly consider synthesis and discriminative tasks.
I graduated from Indian Institute of Technology Kanpur in May 2021, majoring in Electrical Engineering with minors in Machine Learning. I worked on natural language processing, optimization algorithms, stochastic modelling, and wireless networks during my undergraduate days.
For details, check my CV .
Btech in Electrical Engineering with Minors in Machine Learning, 2021
Indian Institute of Technology Kanpur
In this paper, we describe our system for Task 4 of SemEval 2020, which involves differentiating between natural language statements that conform to common sense and those that do not. The organizers propose three subtasks - first, selecting between two sentences, the one which is against common sense. Second, identifying the most crucial reason why a statement does not make sense. Third, generating novel reasons for explaining the against common sense statement. Out of the three subtasks, this paper reports the system description of subtask A and subtask B. This paper proposes a model based on transformer neural network architecture for addressing the subtasks. The novelty in work lies in the architecture design, which handles the logical implication of contradicting statements and simultaneous information extraction from both sentences. We use a parallel instance of transformers, which is responsible for a boost in the performance. We achieved an accuracy of 94.8% in subtask A and 89% in subtask B on the test set.
In this paper, we consider a realistic K tier heterogeneous cellular network (HetNet) consisting of two types of coexisting tiers - Poisson cluster process (PCP) based tiers with base-stations (BSs) and users (UE) distributed as PCP and are coupled, and Poisson point process (PPP) based tier with BSs and UEs distributed as mutually-independent PPPs. We derive the spectral efficiency (SE) for a typical user belonging to one of the tier under max-power association and open access. An important intermediate step involves deriving expressions for the Laplace transforms (LT) of total received power and interference which also opens new avenues for computation of other performance metrics. We also investigate the impact of clustering over the performance of both type of users via numerical simulation and provide important design insights.