George | Jiří Čevora PhD

  • As an experienced technology leader I focus on building and leading teams that develop enterprise-grade software underpinned by AI / Data Science. While my teams are well versed in the aboslute cutting edge of AI research, the emphasis is very much on delivering high-quality and high-performance software in the most pragmatic fashion.
  • I lead work across the entire life-cycle of AI-powered software including designing the application logic, user interfaces, database management, infrastructure and maintenance, training users.
  • The AI techniques I use include all major Machine Learning paradigms. I pride myself in being able to find the perfect tool for any job, not trying to fix all problems with a single tool.
  • I am currently CTO of, taking care of the entire technology stack underpinning the leading product of Anti-Money Laundering and Customer Due Diligence sector.
  • During 2021-2024 I was Chief Data Scientist & Partner at Artefact UK – one of the largest independent Data Consulting companies in the world. Together with my team we have pushed the boundaries of AI deployed in some of the largest enterprises.
  • I do hold an Advisory Board roles at illumr and Rey House where I direct the companies’ strategies in respect to Artificial Intelligence and Data Science
  • I have obtained a PhD in Computational Neuroscience in Rik Henson’s lab at the Univeristy of Cambridge.

My recent research projects

Deep Learning

Success at Deep Learning is greatly dependent on weight initialisation. In this paper Marcel Marais, Dr Mate Hartstein and I introduce a novel weight initialisation technique for autoencoders coined the Straddled Matrix. In the poper we report experiments which indicate that our development outperforms existing state-of-the-art techniques.

In this paper Oliver Turnbull and I explain that this instability in computer vision arises due to the type of tasks we ask computer vision to solve and is therefore unavoidable. However, our analysis of the causes also points to a number of ways how the issue can be alleviated to a large degree.

Reinforcement LEarning

Reinforcement Learning is one of the most exciting techniques within data science and has huge potential to transform entire industries by solving some of the most significant commercial problems. However, it can be tricky to deploy due to safety concerns.

In our paper on the topic Evan Hurwitz, Nelson Peace and I have demonstrated a simple batch learning technique to improve the stability of RL agents. This is an important step on the path to fully unlocking the real world potential of Reinforcement Learning.

Explainable AI

Dr Savio Rozario and I have einvestigated the techniques of explainable AI for their match with the needs of AI systems end-users. In the article published here we have concluded that current techniques are deeply inadequate for this purpose and suggested some alternatives.

small-data science

On an example of workforce reskilling Dr Evan Hurwitz and I demonstrated how even very small datasets can be used to improve decision-making and policy.

Upon an unusual request from a team of rowers preparing for a transatlantic race Dr Mate Hartstein and I developed a general method for routing rowing boats across oceans that aims to find the optimal compromise between shortest path and most favourable winds. Despite the very low number of recorded crossings of the Atlantic with a rowing boat of race specifications this project achieved significant improvement over existing methods.

Fair ML

I’m proud to introduce Rosa, a system that fights discrimination in any Data Analytic pipeline. Rosa is currently available online for free to all in a limited version.

If you are interested in how Rosa works under the hood you may wish to read my paper explaining the underlying methodology – Fair Adversarial Networks – that I have developed over last two years.

If you want to know how Rosa can help you fight discrimination in your Data Science project see my paper providing practical examples of using Rosa to fight discrimination .

You can also watch my talk about Rosa at Kings College London Data Science Society that summarises the current issues with discrimination in Data Analytics and Machine Learning, introduces the methodology Rosa is based on and demonstrates performance of Rosa in real-world scenarios.

Watch my talk about the need for Explainability in AI at ODSC conference 2018:


Check out my criticism of the hype around Biological inspiration behind AI:The relationship between Biological and Artificial Intelligence

Check out my PhD thesis about the role of prediction error in learning associations:The role of Prediction Error in Probabilistic Associative Learning

Read about my argument why the current evidence suggesting that people learn through correcting their errors may in fact be inconclusive:Reconsidering the Imaging Evidence Used to Implicate Prediction Error as the Driving Force behind Learning