Some of the Projects by our graduates

Fighting Malaria with Machine Learning

Eduardo Peire used deep learning, paired with a cheap smartphone microscope, to create an app that can diagnose malaria from a blood sample – which used to take a technician hours using state-of-the-art equipment.This project is now successfully going through crowdfunding, and will soon be making a real difference in the fight against malaria.

Virtual gym trainer

Project by Ahmad Kurdi and Artur Silicki. The app observes your training with the camera, recognizing the exercise. It evaluates your technique, and generates an animation of your form superimposed with corrected posture.  human pose estimators interpret pixel information into human joint positions in space and time (a temporal-spatial encoding).
Our chosen model approaches this by utilizing a temporal convolutional network to construct the 3D points. They abstracted this a level further by inferring a semantic encoding, which can be used for classification of the exercise and for generation of idealized form.

PUPAL - USING DEEP LEARNING TO DETECT TIREDNESS AND DISTRACTION WHILE DRIVING OR AT WORK

Project by Garret O’Connell, Catherine Chaput, Filipe Conceicao. Becoming distracted during a task is costly to goal success and happens to virtually everyone, every day. Our project attempts to warn people becoming distracted by using the known change in eye pupil size to becoming distracted. they used image-based deep learning to characterize this response using a standard webcam and send this signal back to the user to re-focus on them on their task.

DIAGNOSING SKIN CANCER USING MACHINE LEARNING

Dr. Juliana Du is a doctor in Human Genetics. She enrolled at DSR as part of Batch 13, and ended up building an app that uses computer vision to diagnose melanoma in humans.

FAKE NEWS DETECTOR FOR THE TECH INDUSTRY WITH DEEP learning

A pertinent project for a Post-Truth Era, Diego Amicabile has worked across Europe in the IT industry for nearly two decades before coming to Data Science Retreat Batch 12 and starting work on a Fake News Detector (no, not “Generator”!) to expose fictitious news stories in the tech world.

FORECASTING SOLAR ENERGY PRODUCTION USING MACHINE LEARNING

Dr. Setareh Sadjadi was a chemical engineer in DSR Batch 13. She wanted to produce a project relating to climate change and the promotion renewable energy usage. She wrote a script to predict and forecast the photovoltaic power from a solar-panelled roof.

BUILD A ROAD SIGN DETECTOR FROM SCRATCH

Dr. James Ryan was a theoretical physicist from Ireland when he joined Batch 13 of DSR in 2018. His time at the Retreat saw him produce a real-time road sign detector from scratch using a Raspberry Pi to be used in cars in a mere two weeks. How did he do…? Find out!

PREDICTING THE PETROL PRICES IN GERMANY

Christian Lübbe is a PhD holder in Math, Lecturer at the University of Oxford and a participant of Data Science Retreat, Batch 14. His project at DSR aimed at “Predicting the petrol prices in Germany”

If you would rather see some code

Have a look at some projects’ source and think how you could make them cooler. We want our participants to collaborate, and build a vibrant community.

RECOMMENDER SYSTEMS FOR GITHUB USERS

Collaborative filtering (Alternating least squares (ALS) and the Universal Recommender (UR)) are evaluated to recommend interesting repositories for GitHub users according to their preferences: forks and stars.


https://github.com/fradia/github_recommender_systems

CoMNIST: CYRILLIC-ORIENTED MNIST, used to learn the alphabet by comparing hand drawn doodles to letters

A repository of images of hand-written Cyrillic and Latin alphabet letters for machine learning applications. CoMNIST also makes available a web service that reads drawing and identifies the word/letter you have drawn.


https://github.com/GregVial/CoMNIST

IPALOT: AN INTELLIGENT PARKING LOT

Control system for a parking lot based on reinforcement learning, where an AI can take control of the customer’s car and dispach/retrieve it to/from a designated parking spot.


https://github.com/orla84/IPaLot

SENSOR FUSION BY DEEP LEARNING FOR AUTONOMOUS DRIVING

This repo shows how to train a toy self driving car to run laps on a circuit. This was before ‘donkey cars’ were a thing; all hardaware an software lovilingly crafted by Marcus Jones. It’s a CNN model to decide the direction while driving, and a deconvolution mask for visulization to diagnose driving errors – what did the trained CNN model ‘see’ that caused ‘confusion’?


https://github.com/MarcusJones/ai_drive

REAL-TIME AI-POWERED TRANSLATION OF AMERICAN SIGN LANGUAGE TO TEXT

It translates American Sign Language (ASL) to fingerspelled alphabet (26 letters) using transfer learning to extract features. This model is a real-time system with OpenCV – reading frames from a web camera and classifying them frame-by-frame.


https://github.com/BelalC/sign2text

THE RAIMONES - MAKING PUNK ROCK GREAT AGAIN

With 130 songs in MIDI format of Ramones and the lyrics of all their 178 songs, a deep learning model was trained to write guitar and bass lines for Ramones-inspired music.


https://github.com/mfkomakino/raimones