London, England, United Kingdom
Always looking for ways to keep learning, curiosity has driven me to study the inner workings of our universe using computers and abstract mathematics. After my PhD in theoretical physics (string theory) I started working in industry doing applied machine learning for fraud prevention, marketing and cyber security. More recently I've been working on fundamental challenges in artificial intelligence.
Develop production-grade software solutions for research. Accelerate scientific breakthroughs with artificial intelligence. Fundamental research in deep reinforcement learning at scale.
Developed adversarial agents for real-time cyber defense, which were trained with Reinforcement Learning. Led efforts to open-source agent framework (coax.readthedocs.io) and contributed to our simulation environment (github.com/microsoft/CyberBattleSim).
Using reinforcement learning for cyber security.
I worked on the Windows Defender research team. My core role was to create prediction models for detecting and mitigating the impact of malicious and unwanted software at scale. These prediction models are deployed either on the client or in the cloud. The kinds of models I worked on range from simple lightweight ones like logistic regression to sophisticated neural nets for learning meaningful latent space representations (autoencoders, Siamese networks, etc.)
Much of Booking.com's success can be attributed to search-engine optimization and performance marketing. Booking.com is one the world's biggest players on Google's AdWords platform. I work on optimizing the bidding strategy for our AdWords campaigns using new developments in artificial intelligence. I also give internal workshops held each month to help out our ever growing community of data scientists. The topics of the workshops are: - "Python for Data Scientists" : numpy, pandas, sklearn - "Spark for Data Scientists" : pyspark, rdd's, dataframes, Spark SQL, Spark ML - "Deep Learning with Tensorflow" : basics, wide&deep, CNNs for images, RNNs for text - "Online Machine Learning with Vowpal Wabbit" : vw format, FTRL and other optimizer settings, NLP with vw I'm one of the main contributors to our in-house data science package for Python. This package contains a lot of nice functions and APIs that are specific to our infrastructure. My aim is to minimize the need for duplicated "boiler plate" code, whilst making sure that our resources are optimally (and securely) used.
Design, build and implement fraud-prevention engines using artificial intelligence. I write production-level machine learning software as well as everything else needed to get raw data in a shape that a prediction model can understand it.
Algorithm design and business analytics: I optimized our real-time bidding platform for mobile advertisement by designing and implementing machine learning algorithms that maximize conversion rates as well as perform automatic audience targeting. I developed these products from start to finish, from algorithm design to back-end (and front-end) implementation. Besides predictive modelling, I also analyzed and answered business intelligence questions concerning e.g. attribution or customer retention.
Gained valuable business insights via clever webscraping and aggregation. As a team of four, we gained some valuable business insights using only publicly available data. I was responsible for most of the scraping and applying machine learning to predict missing values in the resulting dataset.