Contributed to Transaction Model Risk Management and Review. Streamlined and automated key components of the Model Review Workflow.

Contributed to Traspilation Methods for efficient and fast mapping of Program Quantum Circuit on Hardware Topology Circuit via Graph-Cutting Algorithms.

Responsibilities included:

• Worked with Ritajit Majumdar, Anupama Ray and several other collaborators.
• Devised a novel graph cutting method for implementing a large program circuit to be run on limited qubits NISQ Hardware efficiently.

Contributed towards Neural Face Recognition Robustness Research Paper. Contributed towards Biometric Capacity and Fairness of Deep FR Systems.

Responsibilities include:

• Worked with Prof Richa Singh and Mayank Vatsa.
• Created a Multi-Threaded Fast benchmarking pipelines for robustness of Neural Face Recognition systems.
• Improved previously existing Fairness Benchmarks for Face Recognition Systems with new metrics and dataset creation.
• Contributed to ongoing research on Biometric Capacity of Diffusion and other Generative AI Models through Prompt Engineering.

Contributed toward Unsupervised Image Segmentation, Object Detection and Generative AI Research.

Responsibilities included:

• Worked with Professor Amitava Das in the Joint Embedding Group.
• Developed a Novel Attention-Diffusion Algorithm for unsupervised semantic segmentation and object detection using Diffusion models.
• Developed Large-Scale Data Generation Pipeline for Text-Based Object Discovery to tackle Extreme Object Detection and Classification.
• Leveraged the above system to generate richer visuo-textual joint embedding using contextual and self-supervised learning and RPE transformer.

Improved the existing system in production for annotating speech transcript data at scale and presented my work at IndoML23 Tutorial Session.

• Collaborated with a team of NLP Researchers on designing a Data Programming Pipeline using Active Learning and Weak Supervision.
• Designed a no-code Web Interface to label large Natural Language and Transcript datasets at scale.
• Improved existing enterprise labelling systems by incorporating Semantic Search, Hierarchical and Community Clustering strategies to label data at speed.
• Improved existing Active Learning approach by chaining Ensemble Models and Label Model for Single Label Classification Task.

Recieved The Best Project Award at the School of AI and Data Science during my Summer Internship.

• Worked on unraveling uncertainty as a major underlying latent variable for explaining emotional dynamics.
• Study revealed significant uncertainty representation difference in young and old subjects and identified the Brain Regions capturing this.
• Devised improved sequence models for multivariate BOLD time series predictions and entropy estimation for each Brain ROIs, for quantifying outcome uncertainty.
• Implemented modified dimension reduction and manifod learning techniques for Brain image data in order to capture neural uncertainty representation.

Best Report and Presentation of Data Analysis.

• Analysed the results of the surveys conducted on cyber abuse from various online and offline platforms.
• Modelled the response of people to their chances of being a victim through Decision Trees.
• Heirarchial Clustering revealed groups of people with similar traits who were cyber abused as well as groups prone to cyber abuse.
• Public sentiments about awareness and ways to curb cyber abuse were mined using natural language processing and sentiment classifiers.