Vincent Jarasse

Vincent Jarasse

Applied Machine Learning @ Yelp

Master's degrees from Imperial College London & Telecom Paris

Contact Me

About Me

I am an applied data scientist at Yelp, where my mission is to analyse data, design experiments and build models that improve our understanding of sales' and customers' retention.

Graduate in Artificial Intelligence and Machine Learning from Imperial College London, and in Maths and Computer Science from Telecom Paris, my main interests and projects in academics were in deep learning and its applications to the fields of robotics and NLP.

Apart from my job, I enjoy practicing a wide range of sports, from running to cycling, kitesurfing or mountain hiking.

Work Experience

Intern in Applied Machine Learning - Yelp (Oct. 2019 - Present)

Within the Customer & Sales Intelligence team, we aim at applying Machine Learning to better understand our customers. This encompasses building feature stores, building models, design and run experiments, and maintain production predictions to optimize our customers' acquisition, retention, and understanding in general.

Cybersecurity & Digital Trust Consultant - Wise Partners (Aug. 2017 - Jul. 2018)

With clients operating in a variety of areas, from national public transport to retail and energy, this consulting experience provided me with a valuable insight on the industry's IT system organisation. I had the opportunity to progress on multinational environments, strengthening both my project management and communication skills.

Latest Projects

Automatic behaviour characterisation in robotics using deep learning

Master Thesis within the Adaptative and Intelligent Robotics Lab at Imperial College

Hexapod robot illustration

If the robots are to take part in our daily life, they must be versatile and adaptive. Antoine Cully, director of the Adaptative and Intelligent Robotics Lab at ICL has developped a framework combining genetic and dimensionality reduction algorithms to make robots aware of their behavioral capacities, in complete autonomy. My work consists in exploring how Deep Learning (especially VAE and LSTM) can be used to learn a low-dimensional representation of the robot's behaviour repertoire, that will then be used by evolutionary algorithms to create new behaviours and compare them with existing ones.

LANL kaggle challenge

Kaggle: LANL Earthquake Prediction

In this competition, we have predicted the time remaining before laboratory earthquakes occur from real-time seismic data. The feature extraction, the tuning of XBG, LBG and a fully connected neural network and the combination of the three models have ranked us in the top 29%.

Find out more

Offensval challenge

Offensive language detection in tweets using deep learning

This challenge, hosted by Codalab, consists of 3 steps. First, classify a tweet as being offensive or not. If it is, classify if it is directed towards somebody or not. Finally if it is, classify whether it is towards a single individual, a group of individuals or an organisation.

Find out more