Kathakoli Sengupta

Master of Science in Artificial Intelligence

GPA: 3.96/4.00 | Boston,MA | Sep 2022 - May 2024

Boston University (Graduate School of Arts and Sciences)

Relevant Coursework: Tools for Data Science, Intro to Machine Learning, Artificial Intelligence, Image and Video Computing, Robot Learning and Vision for Navigation, Geometric Processing, Neural Networks and Fuzzy Control, NLP

Bachelor of Technology in Electronics and Communication Engineering

GPA: 9.33/10.00 | Vellore,India | Jul 2018 - May 2022

Vellore Institute of Technology,Vellore

Relevant Coursework: Neural Networks and Fuzzy Control, Robotics and Automation, Sensors and Instrumentation

Computer Vision Scientist

July 2024 - Ongoing

Computer Vision Intern

Dec 2023 - Jan 2024

Wicket LLC

• Improving Wicket's Intent Detector.
• Addressing generalization issues by implementing a spoof equalization loss to prevent overfitting to a particular type and introducing test-time style transformations for evaluation.
• Demonstrated a 3x performance boost by optimizing initial layer freezing and careful hyper-parameter tuning.
• Achieved performance improvements of 5.8% with a patch-based spoof detector trained using semi-supervised knowledge distillation and contrastive learning, effectively capturing facial features and textures.

Data Science Team Lead (Patent Equity Project)

Sep 2022 - Dec 2022

BU Spark!

• Collaborated with Spark to analyse USPTO official data for finding issued patent breakdown and correlation (with regression analysis) on grouping across agent type, gender etc for faster understanding of patent scenario.
• Concluded whether switching firms or subject matter impact attorney success rate grouped by gender with support graphs and success rate measures that is expected to significantly affect firm trend .

Data Science Intern

Sep 2021

The Sparks Foundation, India

• Analysed Global Terrorism Database comprehending patterns of terrorism like hot zone of attack, frequent tools for attacks etc with heatmaps, contour plots, barcharts and pairplots in matplotlib, seaborn etc.

Data Science Engineer Intern

Aug 2021 - Sep 2021

Pianalytix, India

• Developed the first prototype for their College Placement Prediction Model by designing and comparing KNN, Gaussian Naive Bayes, Support Vector Machine and Random Forest classifiers and visualizing each stream demand, age impact.

Research Assistant - Prof. Eshed Ohn Bar (Guide)

Mar 2023 - Ongoing

Human to Anything Lab (H2X Lab) - Boston University College of Engineering

• Introduced a 3D multi-modal navigation dataset for visually-impaired individuals with high-level and low-level text annotations by experts, a novel benchmark for future motion models.<\li>
• Introduced a motion prediction model featuring a transformer-based motion diversification module to generate diversity among subtle actions with RGB and textual context integration.<\li>
• Fine-tuned SOTA motion-language models like MotionGPT and trained RGB and text context-aware motion prediction frameworks to generalise for rare human conditions, such as lack of vision.<\li>
• Collaborated with RedHat to build a conditional routing module for dynamic task offloading that prioritizes controls provided by a local imitation navigation policy, thereby saving communication and computation energy, while offloading the task to cloud for complex scenarios.<\li>
• Proposed a multi-objective reward function designed to optimize energy efficiency while maintaining navigation performance in a custom-designed crowded navigation environment.<\li>
• Demonstrated SOTA performance across multiple CARLA-based navigation evaluation metrics and 35% Ecological Navigation Score boost over baselines.<\li>

Teaching Assistant - EC-518 Robot Learning

Sep 2023 - Dec 2023

Boston University College of Engineering

• Course Topics: Advanced Robotics, Deep Imitation Learning, Semantic Scene Understanding, Object Detection and Tracking, Reinforcement Learning, Sim2Real, Human Robot Interaction.

Research Assistant - Prof. Bang Bon Koo (Guide)

Oct 2022 - Aug 2023

Bio-Imaging and Informatics Lab (BIL Lab) - Boston University School of Medicine

• Designed a dual-channel ( intensity and ravens map) Global-Local Transformer (GLT) architecture for feature fusion resulting in improved accuracy for brain age prediction.<\li>
• Implemented Transfer Learning for limited dataset of Gulf-war Patients and used 3D segmentation masks for lobe-based analysis to determine the predominant one.<\li>

Machine Learning Intern - Dr. Rajarshi Gupta (Guide)

Sep 2021 - Nov 2021

University of Calcutta, Kolkata

• Segregated noisy ECG signals using a random forest classifier with features like Shannon- Entropy and Log Energy entropy.
• Increased the 9.2\% performance by balancing data and choosing proper feature space with Adaptive Synthetic Sampling (ADASYN) and Principal Component Analysis (PCA).

Data Science and Machine Learning Intern - Dr. Abhijit Majumdar (Guide)

Aug 2021 - Oct 2021

Indian Institute of Technology, Delhi

• Modelled a Random Forest Regressor with 92.88% accuracy score for determination of Ultraviolet Protection Factor Jute to measure fabric suitability.
• Designed a random forest classifier with 94.44% 3 fold cross-validation score for categorizing it that scaled fabric production rate and reduced time-consumption.