Sumukha Manjunath
Development
NC, United States
Skills
Machine Learning (ML)
About
Sumukha Manjunath's skills align with IT R&D Professionals (Information and Communication Technology). Sumukha also has skills associated with Programmers (Information and Communication Technology). Sumukha Manjunath has 5 years of work experience.
View more
Work Experience
Machine Learning Engineer Intern
Raven Industries
May 2023 - April 2024
- Improved YOLO-NAS model performance by 20% through increased data diversity by training an image2image diffusion model and engineering a Generative AI-driven data augmentation pipeline. Reduced model unit testing and evaluation pipeline design time by 90% and streamlined CI/CD of models to production by developing a model agnostic pipeline design and orchestration library using DAG data structure and PyTorch. Contributed to the filing of a patent for "Attention Based Feature for object oriented granular neighbor search" as the primary inventor Accelerated data curation time by 80% by automating it through containerized deployment of Python-based image recommendation system using AWS S3, Lambda, DynamoDB, and Facebook AI Similarity Search index (FAISS). Achieved a 99% reduction in memory consumption for image storage by training a Vision Transformer for image representation as 1D NumPy array using self-supervised distributed training with self-distillation and masked-image-modelling.
Graduate Student Researcher
Interpretable Visual Modeling, Computing and Learning Lab
January 2023 - May 2023
- Reviewed and analyzed literature on large foundation models, notably Contrastive Language Image Pretraining (CLIP), exploring methodologies for fine-tuning them for downstream applications. Devised and tested a fine-tuning approach for the pre-trained CLIP model with a focus on preserving out-of-distribution (OOD) performance.
Senior Data Scientist
Mphasis
May 2021 - June 2022
- Expedited patient diagnosis by 60% by leading the engineering of Convolutional Neural Networks based diagnostic pipelines using Python, TensorFlow, Azure Services, FastAPI and Docker for Fundus and OCT systems.
Senior Software Engineer
Robert Bosch
August 2018 - May 2021
- Increased driver safety by improving road sign detection by 40% through Faster-RCNN integration to ADAS feature. Improved obstacle and free space detection by 20% by integrating a Fully Convolutional Network (FCN) trained using TensorFlow to an ADAS component for automated lane changes. Led a team of 3 engineers in the development and deployment of a blueberry farm's real-time harvest estimation tool, achieving 50% accuracy enhancement through a custom Unet model, optimized with pruning and quantization
Education
