Fatima Azfar's Projects
This project focuses on re-building the AlexNet architecture proposed in the paper: "ImageNet Classification with Deep Convolutional Neural Networks" by Alex Krizhevsky on a downscaled version of Tiny ImageNet dataset.
AstroAI is an interactive desktop application that provides personalized horoscopes and astrological insights using user-provided birth details and OpenAI's GPT-3.5-turbo model.
The Chinese Language Similarity project is a sophisticated Natural Language Processing (NLP) endeavor aimed at analyzing similarities in Chinese sentences and lyrics. Utilizing advanced techniques like TF-IDF, BERT embeddings, and a Bidirectional Temporal Siamese Network, it conducts thorough data preprocessing, vectorization, and semantic analysis
This project uses machine learning and neural networks to predict earthquake magnitudes, involving data preprocessing, feature selection, hyperparameter tuning, and model evaluation with regression metrics. It also employs XGBoost for enhanced predictions and includes data visualizations for performance analysis.
MERN Stack Web Application "EpiDetect" which uses a fine-tuned ResNet50 model for skin disease detection.
This repository contains all of my work done in 8 semesters of Bachelors in Data Science from FAST NUCES Lahore. I hope it helps my juniors one way or another!
Implementation of Federated Learning on a simple CNN architecture on CIFAR10 dataset.
A scraper built using beautiful soup and requests to get the data related to the features of products from a website.
Data Visualization Competition Winner | SOFTEC 24' | This project utilizes the Global Carbon Budget data, which tracks carbon dioxide emissions across different countries and sectors. Includes: Feature Engineering and Google Looker Dashboard
The Google Maps Scraper is a Python-based script designed to automate the process of extracting valuable information from Google Maps. Utilizing the Selenium WebDriver, it navigates Google Maps based on user-defined search keywords, collecting comprehensive details such as location names, operational statuses, physical addresses, URLs, etc.
The project leverages a combination of models and methods to predict missing letters in a word, enhancing the Hangman gameplay experience. The core methodologies include a Bidirectional LSTM (BiLSTM) model, a Directed Acyclic Word Graph (DAWG) structure, and a BERT model.
An AI-Based Medical Imaging Disease Prediction System
Classifying mitosis vs normal images using VGG-11, DenseNet, and custom CNN. Employing techniques like focal loss and hyperparameter optimization for improved accuracy.
This is a NLP project built using Python. The goal was to find patterns in the text of movie scripts and identify similarities and if we can successfully cluster the movies of a similar plot together and build a movie recommender trained on the movie script data.
This repository contains the complete implementation of 3 methods to build OCRs, Pytesseract, EasyOCR and Google Cloud Vision API.
Basic Implementation of Onion Routing
Pegasus Paraphraser is a text paraphrasing system built using the tuner007/pegasus_paraphrase model to generate simplified versions of input text by splitting it into sentences and leveraging an encoder-decoder architecture.
A Map Dashboard built in Power BI for Punjab Food Authority to supervise the operations taking place in the province. This project included Data Analysis, Data Preprocessing and Data Visualization.
This project implements a Retrieval-Augmented Generation (RAG) model that uses a directory containing text files as documents for information retrieval and generation. The model combines retrieval and generation capabilities to answer questions based on the provided documents.
ResNet-Optima is an advanced adaptation of the classic Residual Networks (ResNet) architecture, specifically designed to address two critical challenges in deep learning: generalization and computational efficiency, while maintaining the core principles of the original ResNet.
Developed a Sequence-to-Sequence (Seq2Seq) model with LSTM units for text summarization, utilizing the BBC News Summary dataset and implemented with an encoder-decoder architecture for effective information extraction and summarization.
This project implements a Text to Image Generator using a Conditional Generative Adversarial Network (GAN) for synthesizing floorplan images from textual descriptions. It includes features such as custom dataset handling, performance metrics like FID and IS, and configurable training options for optimization.
This project employs deep learning models to conduct sentiment analysis on a dataset of Urdu tweets, aiming to classify them into positive or negative sentiments. Using advanced neural network architectures like RNN, GRU, LSTM, and BiLSTM, and exploring the impact of different word embeddings.