Karan Variyambat Email & Phone Number

United States

La Jolla Shores, California, United States

Wildfire Smoke Detection• Working with Prof. Mai Nguyen on improving the performance of the multimodal SmokeyNet model for wildfire smoke detection. • Developing scripts for augmenting the existing dataset with new fire sequences, automating the data pre-processing pipeline. • Exploring architectural modifications leveraging ViViT, DINOv2, CNN and LSTM.

United States

• Enhanced the performance of a 4-chip FMCW radar by ~8%, developing calibration scripts in MATLAB for phase offset compensation.• Automated radar data acquisition and signal post-processing, reducing operational time by ~50%, creating Python, MATLAB and Lua scripts.• Prepared test case scenarios for machine learning models to enhance the perceptibility of imaging radars in automobiles.

La Jolla Shores, California, United States

Fall Detection Using Wearable Sensor Data • Developed a proof-of-concept machine learning model for fall detection, utilizing temporal readings of accelerometer and gyroscope. • Conducted data cleaning and feature engineering to aggregate the temporal readings and extract features, facilitating effective model training.• Employed cross-validation technique to address data imbalance and trained multiple models including Random Forest, XGBoost, and AdaBoost, achieving an average F1-score of 0.97.Ultrasound Image Registration • Implemented rigid image registration of ultrasound scans leveraging pystackreg, achieving an average NCC score of 0.94. • Performed qualitative analysis of registration results through image overlays and highlighted mismatch regions, enhancing the interpretability of registration output.

Pune, Maharashtra, India

• Demonstrated adept project management, leading the design and development of electrical subsystems for four land-based defense projects.• Improved the design and functionality of power distribution and control units for a strategic mobility platform project by meticulously researching and replacing redundant and obsolete components with optimal components, resulting in improved compactness and an enhanced user interface.• Developed functional logic and defined precise connections by preparing detailed schematics, wiring charts, and cable schedules using AutoCAD and Zuken E3.Series, achieving successful unit and system-level integration of 2 power distribution units for an air defense system. • Prepared a comprehensive technical proposal, conducting in-depth technical analysis of project requirements for an air defense system, helping receive the project contract. • Performed functional testing of units to ensure system reliability and resolved on-site issues during trials.

Mumbai, Maharashtra, India

• Created a balanced dataset by preprocessing about 1 million frames of deepfake videos from diverse sources viz., Face Forensics++, Google Deep Fake Dataset, and Deep Fake Detection Challenge Dataset (450GB+), ensuring exposure to a wide range of deepfake methods for model training.• Developed a model using CNN and leveraging Transfer Learning to detect deepfakes and achieved an F1 score of 0.91 on the test set.• Presented and published a paper titled "Employing Transfer-Learning based CNN Architectures to Enhance the Generalizability of Deepfake Detection" at the “2020 11th International Conference on Computing, Communication, and Networking Technologies (ICCCNT)” held in July 2020 at IIT Kharagpur.• Finished amongst the top 15% in the Deepfake Detection Challenge by Facebook hosted on Kaggle, utilizing the Nvidia DGX-1 AI Supercomputer for training and tuning the models.

Mumbai, Maharashtra, India

• Built deep learning-based CNN models and trained them on medical images to aid the detection of skin cancer, serving a dual purpose: • Identification of the skin lesion as benign or malignant with an accuracy of 0.89.• Probabilistic classification of the lesion into seven distinct categories and predicting the most probable one out of those seven with an accuracy of 0.82.

Master Of Science - Ms, Electrical And Computer Engineering

Majoring in Machine Learning and Data Science

Bachelor Of Technology - Btech, Electronics Engineering, 9.12/10 (Cgpa)