Selected Projects

Age Regression from Brain MRI

Predicting brain age given MRI brain data and comparing it to the chronological age of the patient is useful for detecting neurodegeneration and preventing cognitive decline. Using data from 652 healthy subjects we implemented three methods for age regression. The first approach uses the ratios of overall brain volume and the volumes of different brain tissues such as grey matter (GM), white matter (WM) and cerebrospinal fluid (CSF). The next two approaches use grey matter maps for PCA-based regression and CNN-based regression, respectively. The final testing shows that the CNN-based model performs better than the other considered models.

Deep Reinforcement Learning Agent with Randomly Generated Mazes

Implemented a deep Q network using PyTorch to train an agent that is capable of tackling any random maze with a set of unknown obstacles and unknown goal locations.

Sentence-level Quality Estimation for English-German Machine Translation

The aim of the project is to exploit Natural Language Processing techniques to find the most suitable way of evaluating machine translation quality. The focus was on English-German corpora, where we showed that the use of a hierarchy of BiLSTMs along with adaptive max-pooling layers resulted in the best performance compared to:
* The use of pre-trained sentence-level embedding
* Using GloVe word embedding along with traditional machine learning regressors

Mini Neural Networks Library Using Python

In this project, I have contributed to the following:
* Implementing a mini-library for neural network
* Using the implemented library to solve a classification task in insurance pricing

Image-Based Robot Learning for Manipulation

In this work, I surveyed the different ways robot learning can be conducted. I focused on comparing deep learning approaches with the normal pipelined machine learning solutions. I, Further, experimented and evaluated the performance of two different deep learning solutions with imitation learning, using Panda robot arm for the task of reaching a target object in the scene. My experiments show that key-points based CNNs solution (with the special spatial softmax layer) outperforms the direct CNNs one (normal CNN with flattened layer) with a success rate of 80.40%.