Wearable Sensing & Human Activity Recognition Who’s Wearing the Glasses? Behavioral biometrics to identify users by analyzing head gestures Context-Aware Dynamic Activity Recognition on Wearables Adaptive recognition system dynamically adjusting sampling rates, sensors, and feature complexity to recognize smoking-related and daily activities from wrist-worn motion sensors Biosignal-Based Stress Detection Detecting stress states from ECG, GSR, and heart rate signals using machine learning Feature Engineering for Wrist-Based Activity Recognition Analysis of motion, orientation, and rotation features extracted from wrist accelerometer data to classify complex activities Attention-Based Knowledge Distillation for HAR Lightweight human activity recognition (HAR) combining knowledge distillation and attention modules to improve performance on wearable sensor data Computer Vision & Image Modeling Image Stitching (Panoramic) from Scratch Homography estimation and panoramic reconstruction combining SVD, backward warping, and image blending (no built-in libraries like OpenCV) How Does a CNN Learn? In-depth analysis of image classification using convolutional networks, architecture variations, and training dynamics Sequence & Pattern Learning Next-Word Prediction from Scratch Sequence modeling of text data with a custom multi-layer perceptron built entirely from scratch (only NumPy) Expectation-Maximization for GMM from Scratch Unsupervised clustering and density estimation using a Gaussian Mixture Model (GMM) built entirely from scratch (only NumPy) Generative Modeling Variational Autoencoder for MNIST Digits Generative modeling of handwritten digits using a variational autoencoder (VAE) combining an LSTM encoder and CNN decoder
