Here's a snapshot of my projects on various topics of my interest in Data Science & Machine Learning.
If you'd like to know more project-specific details, the respective GitHub ( )
project repositories are linked as well.
Generative Deep Learning with Multi-modalities
This project takes creative inspiration from OpenAI's DALL·E family of AI models.
The built Neural Network generates visual elements containing contextual information in the user-provided text-prompt.
The model consists of two major components — a multimodal model combining language concepts with visual context (CLIP) & a generative transformer for generating image representations (VQGAN).
Predicting Bio-Molecular properties from Molecule Structure
Molecular structures can provide meaningful insights about their physical properties.
Graphs are a convenient way of representing bonds among atoms in a molecule.
This project explores Graph Deep Learning algorithms to extract features from Graph Representations of bio-molecular structures to predict their molecular properties with over 90% accuracy.
Creating Artificial Human Faces with Generative Deep Learning
Ever wondered if images of people online are actually of real people?
This project attempts to generate convincing images of non-existent individuals by using the CelebA dataset to explore Generative Deep Learning Algorithms.
This project was used as an opportunity to get familiar with various Generative Deep Learning Algorithms.
Navigation RL Agent to avoid Obstacles
This project explores different variations of the Deep-Q-Network (DQN) to train an RL Agent to navigate through a custom Environment.
The Agent's objective is to collect as many yellow bananas as possible while avoiding the blue bananas within a fixed number of Timesteps per Episode.
This project was a part of a set of assignments in the Udacity Deep RL Nanodegree Program.
AI Arcade
Play in real-time against Deep Reinforcement Learning Algorithms in short-duration micro-game environments!
Fast Food Image Classification
Who doesn't love fast food? Could we use images of different fast food items to programmatically segregate them?
This project explores building an Image Classifier to identify unique categories (101 classes) of fast food items.
The Image classifier was built using Tensorflow's food101 dataset
and has been deployed as a RESTful API on Heroku,
which can be used to classify user-uploaded images.
Classifying Cats & Dogs with Transfer Learning
This project explores image classification of Cats & Dogs using Kaggle's Dogs vs. Cats dataset.
3 Deep-Learning Models have been built for the classification task — a custom Deep CNN & 2 Transfer Learning (TL) models based on VGG-16 & Xception architectures.
In-depth model performance evaluation reveals the Xception TL model to be best performing model with 99.5% classification accuracy.
Image Description Generator with Deep Learning
A Deep Learning model for Image Annotation was built in TensorFlow.
The model architecture consists of pre-trained Inception V3 as a Convolutional Encoder used to generate embedded representations containing encoded features extracted from the images.
A RNN Decoder is used to translate the embedded information and generate text used as caption for the image.