Capsule Network for MNIST Classification

Implement CapsNet from Geoffry Hinton's paper Dynamic Routing Between Capsules.

Model the hierarchical relationships inside of a neural network with 3D graphics (Dynamic routing)

Utilise libraries and frameworks like Tensorflow, tqdm and numpy to organise code and enhance maintainability.

Check pointing, training of weights and storage in logs/models.

Feature Selection and Data Visualization with Matplotlib.

Work on MNIST dataset for easy comparison with SOTA models.

Generating Matching Shoe Bags from Shoe Images using DiscoGAN

Implement research paper entitled “Towards the Automatic Anime Characters Creation with Generative Adversarial Networks”.

Model Customised-Face2Anime-GAN to generate new content and for anime generation with Python

Generation of complex random variables with Inverse transform method

Design generator to create fake data and discriminator to classify its output as real.

Learn to alter the network's weights to reduce the error or loss of the output

Load and extract data and pre-trained VGG weights

Crime Analysis and Prediction using Machine Learning

In this project, we will build a machine learning module, the model works on the concept of time series forecasting and clustering.

The aim is to find spatial and temporal criminal hotspots and also forecasting of crime using a set of real-world datasets of crimes.

In addition, we will predict what type of crime might occur next in a specific location within a particular time.

After successful running of the module the analysis and the forecasting results are shown through graphs and plots.

We intend to provide an analysis study by combining our findings of a particular crimes dataset with its demographics information.

Problem Statement, Scope and Objectives, Dataset Importing

Plotting and analysis of different crime, Algorithms, Train the Model

Evaluation of the Model, Parameter Tuning, Making Predictions, Forecasting results

Automatic Speech Recognition (ASR)

Build an Automatic speech recognition system (ASR) based on a mix of acoustic model and language model.

Ensure to determine the phonetic units in the language.

Model the word sequences simultaneously for Indian accent English (IN-EN) speakers.

Training and testing of the model could be done by making a custom dataset of audio recordings.

Final evaluation of the ASR is done on the basis of Word Error Rate.