Projects
A collection of my work in machine learning, physics, and software engineering.
Deep Learning Computer Vision at Siemens R&D
Developing deep-learning computer vision models to identify and interpret electrical elements and symbols for smart grids. Building state-of-the-art Computer Vision models and training them on Azure ML studio using Compute Clusters.
Transformer-based News Summarization
Advanced NLP project using BART transformer for news summarization. Achieved loss reduction from 1.5276 to 0.1102, with high ROUGE scores (rouge1: 0.7753, rouge2: 0.6970). Integrated with Weights & Biases and Hugging Face Hub.
Graph Neural Networks Classification
GCN implementation for link prediction on Cora dataset, achieving 87.89% test accuracy. Optimized train-validation-test splits and implemented early stopping with cross-entropy loss evaluation.
Physics Informed Neural Networks
PINN implementation in PyTorch for 1D harmonic oscillators, combining data fidelity and physical law compliance in the loss function. Includes analytical solution integration and training visualizations.
Prophet Time Series Forecasting
Stock forecasting system using Facebook's Prophet library, featuring trend analysis, seasonality decomposition, and changepoint detection. Includes MAE evaluation and Plotly visualizations.
Pedestrian Detection with HOG
Advanced pedestrian detection system using OpenCV's HOG. Implemented Sobel operators for gradient calculations and visualization techniques for feature extraction.
Real-time Face Detection
Real-time face detection and recognition system using OpenCV and face_recognition. Features pickle-based facial encoding storage and live video processing.
Web Traffic Forecasting
ML-based web traffic prediction achieving 81.07% accuracy using Logistic Regression and SVM. Includes advanced preprocessing with Mahalanobis distance and achieved 0.87 ROC AUC score.
Particle Swarm Optimization
Implementation of PSO for optimizing non-convex functions, using swarm intelligence to find global minima in complex landscapes.
ML Algorithms from Scratch
Comprehensive implementation of ML algorithms including AdaBoost, Decision Trees, KNN, Regression, Naive Bayes, PCA, Random Forest, and SVM. Complete with validation and testing.
Quantum KNN Classifier
Quantum implementation of KNN using Qiskit, featuring amplitude encoding and custom Quantum SWAP Test module. Includes multiple quantum registers and classical result processing.
Quantum Fourier Transform
QFT implementation using PennyLane with RMSProp optimization in a two-qubit system. Includes 3D visualization of quantum state convergence.
Simulated Annealing for TSP
Optimized solution for the Traveling Salesman Problem using Simulated Annealing, achieving route cost of 28 units after 50,000 iterations. Features stochastic perturbations and Metropolis criteria.