Projects

Synesthesia is defined as a perceptual phenomenon when two unrelated senses become linked, for example seeing shapes when hearing sounds or colours when thinking of numbers etc. The following project is based on the spoken digits dataset, a collection of recordings with human-spoken digits. By leveraging the power of computer-vision techniques and the Mel spectrogram (in short an image representation of the frequencies as seen in the image) we can teach also teach machines how to "see sounds"

The following project is based on the Rossman, a major European drug store, store sales kaggle competition. In this challenge, the goal was to predict 6 weeks of daily sales for 1000+ stores located across Germany. To solve this problem we applied time-series analysis techniques to be able to capture the important features of the dataset including trends, seasonality and cyclic variations.

Sentiment analysis is the process of examining user posted content to determine the emotional tone of the post. It can be very useful to manage hate speech, analyze customer feedback, survey responses and monitor social media and customer experience in general. In this project we build a sentiment analysis model using the IMDB reviews dataset to determine the overall reception for a given movie.

Deepfakes and Image generation models have seen an increase in popularity over the last couple of years. In this project we use generative adversarial networks (GAN) and train it on the celeb-a dataset, a collection of labelled images of celebrities, the resulting model is capable of creating new images of celebrities. The model is then extended to a conditional GAN to generate a new fake person with a given characteristic.

A language generation model based on the writings of the great J.R.R Tolkien. This model was trained on most of the Legendarium including: The hobbits, The lord of the rings and The Silmarillion.
Check out the Github repo to load the model and test it out.

This project is a proof of concept with the idea to apply reinforcement learning to telecommunications network. The goal was to develop an agent capable of optimizing traffic and make routing decisions in a network to reduce congestion and increase communication efficiency by also minimizing dropped packets.