Projects
CNN-based Price Prediction from Product Images (2019): Used Transfer Learning to extract visual features from Amazon product images and thereby, predicting their prices.
Pneumonia Detection in X-rays (2018): Implemented a ResNet architecture to detect the pneumonia affected areas in chest x-rays with 77.5% IoU [Report].
Acoustic-based Occupancy Detection (2018): Extracted audio features from collected data and applied Random Forest to achieve 97% accuracy.
Lung Cancer Detection in CT scans (2017): Used Transfer Learning to detect lung nodules in 3D CT scans from LIDC/IDRI dataset.
Undergrad Projects
Python web: Aleef (2016) : A search engine project for classifying and filtering violent religious contents in web.
Android: Online Newspaper App(2015): Google App Engine based mobile application that provides all latest news from most popular news website of Bangladesh.
Graphics: Ray Tracer (2015): Implementation of ray tracing simulation in computer graphics with options for image rendering, reflection, and refraction for plane, sphere, cylinder, triangle shapes.
Android: Al-Quran(2015): Al-Quran in both Bangla and English Translation.
Android: Nearest ATM Finder (2014): Smartphone application to find the nearest ATM booths and distance from users current location (Android)
PHP: Online Course Portal (2014): A online course management system for both teachers and students (PHP)
OS: Basic Operating System (2014): An implementation of simple operating system based on Nachos Framework with multi-programming, threading, memory management with caching and virtual memory management.
Networking: Routing Protocol (2014): Implementation of DVRP (Distance Vector Routing Protocol) with bit stuffing, hashing for error detection
UML: Information System Design (2014): An Information system designed for official management activity of RAJUK (UML)
LEX & YACC: Simple Compiler (2014): A simple compiler for Pascal. (LEX, YACC)
JAVA: Web Browser (2012): A simple web browser including basic Html parser (JAVA)
C: DX Ball (2011): A simple version of DX Ball game using iGraphics Library (C)
Python: Simple Search Engine (2011): Implemented a simple indexing for search-engine.
Machine Learning Courses:
[Feb '16] Machine Learning by Andrew Ng., Coursera (Certificate, Grade: 95.7%)
Implemented a two-layer neural network with backpropagation to detect handwritten digits and an email spam filter using SVM with Gaussian Kernel. Along with these supervised learning models, I also implemented K-means algorithm and dimensionality reduction technique with PCA.
[Sep '17] Neural Networks and Deep Learning by Andrew Ng., Coursera (Certificate, Grade: 98.6%)
Implemented a cat classifier using both logistic regression and L-layered deep neural network model.
[Sep '17] Improving Deep Neural Networks by Andrew Ng., Coursera (Certificate, Grade: 96.0%)
Implemented L2-regularization, dropout, Adam optimization techniques from scratch and built a sign language detection model using Tensorflow.
[Oct '17] Structuring Machine Learning Projects by Andrew Ng., Coursera (Certificate, Grade: 96.7%)
Learned the basics of Train/dev/test set distributions, high bias/variance problem, Transfer Learning, Multi-task Learning and End-to-end Deep Learning.
[Oct '17] CS224n: Natural Language Processing with Deep Learning by Chris Manning and Richard Socher.
Implemented Word2Vec Skip-gram Model (with negative sampling) from scratch, and built a Linear Classifier Model and a neural-network-based Dependency Parser for grammatical structure analysis of a sentence.
[Jan '18] Convolutional Neural Networks by Andrew Ng., Coursera (Certificate Grade: 98.4%)
Implemented CNN model and ResNet blocks from scratch, YOLO algorithm for car detection, Face detection & verification and Neural Style Transfer (NST) techniques.
[Feb '18] Sequence Models by Andrew Ng., Coursera
Implemented RNN model with LSTM cell using Numpy and built a Character-level Text Generation Model using RNN.
Personal Projects
- Crawled Prothom-alo, a popular bangla news site, to create a dataset of ~45,000 news articles in bangla language. Then, pre-processed the data and implemented a phrase detection model. I trained a Word2Vec skip-gram model using the dataset to create word embeddings of size 128.
- I implemented a CNN model including one Convolution Layer(with filter size 2,3 and 4), Max-pooling and Dropout. I trained the model with IMDB movie review dataset to classify the reviews as positive or negative.