May 2022 – Dec 2022 · 7 mos · Seattle, Washington, United States · Remote

• 1) Explored the utility of various text-to-image generative models for text-based segmentation of images.
• 2) Demonstrated the use of latent diffusion models (LDMs) pretrained on large-scale internet data for text-based segmentation.
• 3) Proposed novel ways to utilize features from the internal stages of the LDM to improve the segmentation performance by nearly 6% on real images and 20% on AI-generated images.

May 2020 – Dec 2020 · 7 mos · Campbell, California, United States · Remote

• 1) Worked with fellow researchers on various audio-visual and self-supervised learning techniques.
• 2) Prototyped novel learning algorithms for various audio and video synthesis approaches in large-scale production systems.
• 3) Integrated solutions in cross-language technology stack consisting of Python, C++, and CUDA.

Aug 2017 – Dec 2023 · 6 yrs 4 mos · College Park, MD

• 1) Text-based segmentation of real and AI-generated images by leveraging text-to-image latent diffusion generative model pretrained on large-scale internet data.
• 2) Self-training methods such as knowledge-distillation targeted for monocular depth estimation.
• 3) Domain Adaptation between synthetic and real datasets for applications such as Monocular Depth Estimation of outdoor scenes and Face Normal Estimation.
• 4) Violence detection in videos using a Bidirectional ConvLSTM network.
• 5) Guided Inpainting using Generative Adversarial Networks that can enable the use of different car images as guides to edit cars in street view scenes.

Aug 2016 – Dec 2017 · 1 yr 4 mos · College Park, MD

• As a Research Assistant in the Department of Geographical Sciences, I contributed to the following areas:
• 1) Crop yield prediction using the EPIC model that utilizes data from the soil, site, remote sensing, daily and monthly weather for different regions in the US.
• 2) Cropland data layer prediction based on the history of crops using a Bidirectional ConvLSTM network.
• 3) Development of a crop phenology algorithm that accurately predicts the timings of different stages of crop growth ranging from Greenup to Senescence.
• This role requires immense organizational skills, dedication and time management and not to mention a deep understanding and knowledge in High-Performance Computing systems considering the huge amounts of data processing that happens often. The work taught me how to organize and plan a research process to achieve a desired outcome in the specific timeframe and moreover the set of skills I picked up ranges from understanding the requirements of a professor to writing potential papers for publications.

May 2014 – Jul 2014 · 2 mos · Noida, Uttar Pradesh, India

• Analysis of Simultaneous Switching Noise and Cross Talk effect in PBA+SLA :-
• Studied performance of DDR (PBA) and PCI (SLA) integrated on PCB (Cadence Sigrity Tool)
• Depicted the SSN effect by DDR on PCI through eye diagram at receiver of PCI
• Analyzed the SSN and crosstalk effect by varying parameters viz.,the Strobe Pair of PCI bus,
• length of these strobes and speed of the DDR bus
• Achievement :- The presence of SSN was confirmed through a decrease in eye height of PCI, in DDR ON case compared with DDR OFF case