Satarupa D.

Key Skills

Software

Other

Top Subject Matter

TinyML computer vision

knowledge distillation

curriculum learning

Top Data Types

Image

Top Task Types

Fine-tuning

Classification

Freelancer Overview

M.Tech Thesis: Confidence-Aware Curriculum Knowledge Distillation for Resource-Constrained TinyML. Brings 2+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Other. Education includes Master of Technology, National Institute of Technology, Durgapur (2026) and Bachelor of Technology, St. Thomas College of Engineering and Technology (2024). AI-training focus includes data types such as Image and labeling workflows including Fine-tuning and Classification.

Labeling Experience

M.Tech Thesis: Confidence-Aware Curriculum Knowledge Distillation for Resource-Constrained TinyML

OtherImageFine-tuning

Developed a confidence-aware knowledge distillation pipeline to train lightweight TinyML computer vision student models. Used Monte Carlo Dropout to drive an adaptive curriculum learning strategy and dynamically adjust distillation weighting during training. Optimized student model performance under strict memory and computational constraints with repeated experimentation across training runs. • Implemented confidence-aware curriculum learning using Monte Carlo Dropout • Designed dynamic loss weighting tied to confidence signals and model constraints • Trained and validated lightweight CNN students distilled from a high-capacity teacher • Measured accuracy improvements for resource-constrained TinyML deployment

2025 - Present

Project: Confidence Aware Curriculum Learning-Based Knowledge Distillation | TinyML

OtherImageFine-tuning

Built lightweight CNN TinyML models by distilling a high-capacity teacher into memory-efficient student networks. Implemented confidence-aware curriculum learning using Monte-Carlo Dropout to control training progression. Designed dynamic loss weighting that accounts for prediction confidence and model size, improving training effectiveness under embedded constraints. • Distilled teacher knowledge into student networks for TinyML • Applied confidence-aware curriculum learning via Monte-Carlo Dropout • Created dynamic loss weighting based on model size and confidence • Performed dataset separability analysis using feature embeddings and evaluated confusion behavior

2025 - 2026

Hydrophobicity Classification of Composite Insulators - Computer Vision, TinyML

ImageClassification

No description provided.

2025 - 2025

Project: Hydrophobicity Classification of Composite Insulators | Computer Vision, TinyML

OtherImageClassification

Conducted a case study of the CL-KD framework for insulator hydrophobicity classification across HC1–HC7. Used Edge Impulse for transfer learning with MobileNetV2 and EfficientNet as feature backbones. Applied pruning and knowledge distillation to optimize models for real-time TinyML deployment on edge devices. • Trained image classifiers for hydrophobicity categories HC1–HC7 • Used Edge Impulse transfer learning with MobileNetV2 and EfficientNet • Optimized models using pruning • Improved compactness/performance using knowledge distillation for edge deployment

2025 - 2025

Education

Work History