skip to content

Department of Chemistry


Ed Pyzer-Knapp working at his desk, courtesy EPK

As IBM Research Global Lead for AI Enriched Modelling and Simulation, and a visiting Professor at Liverpool, alumnus Ed Pyzer-Knapp was already a busy man. Instead of cutting back, he has now launched a new journal highlighting AI research.

“This is a labour of love, and I found that you don’t give things up for that,” he says of Applied AI Letters, launched by Wiley in February with Ed as Editor-in-Chief. “The journal was developed specifically to provide peer review and validation in a rigorous academic way for researchers, not only in academia but also in industry where so much AI [Artificial Intelligence] and ML [Machine Learning] research is occurring,” he says.

“I was drawn to working at IBM because I could see real world applications and benefits from using new AI and ML methodologies,” says Ed. But as a researcher working within industry he was frustrated by the limited opportunities for publishing peer-reviewed articles in these quickly expanding fields.  

“Previously the journal choices for publishing ML and AI papers have been limited,” he says, noting that many journals overlook papers from industry, and with others the turnaround times are too slow. “The typical way was to go to a conference and then have your paper published.”  But as interest in AI exploded in recent years, the number of papers available soon exceeded the ability of conferences to get them published. “Because of the growth of the field, some of the big conferences have reject rates similar to Nature!” he says.

The idea for Applied AI Letters was formed when Ed was working on a textbook on machine learning and mentioned his frustration to his editor at Wiley. “She said I might be exactly the right person to pursue it, and that I should see someone on the journal side, so I did.”

Because of his early involvement, Ed was able to include several features he saw as critical to the type of journal he wanted to create. Most important was that the research should be accessible to all.  “We were very passionate about the fact that if we were going to do this, it is not good enough to hide it behind a firewall and make people pay to see it,” he says. “This is a community we are trying to build and we want this research to be accessible.” As a result, Wiley agreed to make the journal Gold open access, and has waived all open access fees through the end of 2020 – “So if anyone applying AI to real world problems out there have papers to submit, now’s the time!” he says.

Another feature of which Ed is proud is that the journal has its own pre-print server. “As soon as you submit your paper, it becomes visible to the world, and if it’s accepted by the journal, the DOI and electronic tagging flow through with your paper.” This means that all of a paper’s metrics for citations and other factors are kept together. If the paper is not accepted, it will stay on the pre-print server, but will drop out of the journal’s ‘collection’ of accepted papers.

Ed also addressed his concerns about turnaround time by initiating a limit on article length. “We deliberately limit the length of articles, which makes them easier to review (the journal will publish primarily short-format letters).” And unlike conferences, which could have the near-impossible task of reviewing thousands of submissions in one month, the journal has a steadier stream of submissions and can ‘load balance.’

Applied AI Letters was launched at the height of Covid-19, which Ed admits has been “an extra challenge.”  “What’s worth pointing out is that I have three fabulous editors who are doing a lot of the heavy lifting.  We have great support from Wiley and we’ve got editors across a mix of academia -- you can’t do something worthwhile without having a great team.”

Ed became interested in computers at a very early age. “When I was in primary school I saw my first computer, which was a BBC Micro. I remember being the only person in school who could wire it together properly. It didn’t have a mouse, which is maybe why I still don’t like to use a mouse today!” Ed wrote his first programme on that computer when he was about five years old.

Some years later, while completing a Master’s in Supramolecular Chemistry at Durham, Ed discovered that he preferred to analyse data rather than work with chemicals (he refers to his lab table as being a “mess of fluorescence”). He completed his PhD in Computational Chemistry here (Selwyn 2013), supervised by Graeme Day and Bill Jones, and also worked closely with David Wales.  “I can’t say enough good things about them and I continue to work with Graeme (Bill Jones has recently retired).  I still call David up with questions from time to time, and he always makes time for me, even after all these years.“

 “Cambridge is one big highlight for me. I loved my time there – and Chemistry is a great department full of really, really great people.” Ed and his wife, who was at Jesus and did research within the Nitschke group, still reminisce about the formal college dinners which they enjoyed so much. And his best man at their wedding was one of his lab mates [Hugh Thompson (Churchill 2006-2014), now an IP lawyer in Switzerland]. Edward admits he sometimes got frustrated about his work “but if you didn’t get frustrated, it would mean you didn’t care.”

After a prestigious stint at Harvard working on the Materials Genome Initiative Clean Energy project, where he further developed his interest in machine learning, Ed returned to the UK to join a new team being started by IBM for large-scale machine learning research in the Hartree Centre near Liverpool.  

Ed‘s description of the difference between ML and AI is succinct: “AI encompasses all, ML is a subset of AI, and Deep Learning is a subset of ML.”

As a Visiting Professor of Industrially Applied AI at the University of Liverpool, Ed is involved in the Materials Innovation Factory run by Andy Cooper, whom he first worked with at Cambridge. “We are researching different technologies to improve the speed at which we can discover new things --  we are asking can we be more efficient, can we build better models?” He is also interested in the underlying methodology, and is working with Simon Maskell, a professor in Electronics and Electrical Engineering, on a particular algorithm for Bayesian inference known as SMC samplers. “I want to make a prediction, and in the real world I also want to know what the uncertainty is around that prediction. SMC can give me this information at a higher quality than before, whilst also running faster due to its great scaling properties”

Ed has always felt the most interesting research work happens on the interface between different areas. “Obviously you need people to go deep into a particular area, but that doesn’t suit me. I’ve always been the guy that tries to find all the pieces that interlock together. My skill set has been to come up with analogies that allow us to think about what works together.”

What ties all of Ed’s roles together is his conviction that “a job is only done when you can demonstrate that it works in the real world.” He says: “My job isn’t finished until I can say that an idea I came up with on the blackboard also works in the real world, which means I am providing some value back to the community.”