Harshiv Chandra

SPARK Summer Research Intern

Indian Institute of Technology, Roorkee, IIT Roorkee

Roorkee

06.2022 - 08.2022

Deep Learning based forecasting, Data Science
Selected for a summer internship at IIT Roorkee under the SPARK program from amongst an international candidate pool
Designed and implemented an optimized Long Short-Term Memory (LSTM) model, using preprocessed Synthetic Aperture Radar image data from satellites for flood zone forecasting in Southeast Asia
PROJECTS
Lung Image Quality Assessment
Collaborated with faculty at the University of York and BITS Pilani Dubai to create an automated pipeline to evaluate input lung image datasets on deep learning models
The developed pipeline consists of an Auto tokenizer layer, a CNN classifier & an ensemble segmentation model with U-Net, UNet++, FCN and SegNet
Whitepaper submitted to Journal of Biomedical and Health Informatics for publication
Technologies used: Python with PyTorch and Pandas libraries
Stock Market forecasting using Deep Learning Models
Evaluated deep learning and machine learning algorithms (LSTMs, CNNs, LSTM+CNN, SVR) for stock price prediction
Integrated Yahoo Finance API for real-time stock data retrieval and analysis
Technologies used: Python with Scikit-learn, Pandas, and BeautifulSoup libraries; and the Yahoo Finance API
Bone Fracture Segmentation
Designed and developed a U-Net model to segment bone fracture zones in X-Ray imagery
Generated labels for segmentation using an automated k-means clustering model on the FracAtlas dataset
Technologies used: Python with Scikit-Learn, Pandas and TensorFlow libraries
Circuit Mapping on Memristor CrossBar Array
Designed a mapping algorithm to convert a majority-inverter netlist onto a memristor crossbar array
Developed functions using the MockTurtle logic synthesis library from EPFL
Technologies, C++ with MockTurtle Library
Mental Health Prediction using Images
Created a deep learning framework to score images for signs of mental stress
Developed a model with an Encoder-Decoder architecture, sentiment analysis, and classifiers (Random Forest, SVM)
Technologies used: Python with Scikit-learn, HuggingFace, and NumPy libraries.