Jannes Gladrow, AI Resident
Jannes Gladrow, AI Resident
Hey! My name is Jannes Gladrow and I am an AI resident at MSR Cambridge (UK). In this blog post, I want to talk about how I got here. The path behind me is not direct but curved and seems to be starting in a place which in a way is not so far from where I am now. But, let me explain.
The story begins in high school, where I would often take part in science competitions. It so happens, that my last and most memorable project focused on optical data storage, which is located somewhere in the jungle between physics and computer science.
I must have enjoyed that because I went on to major in physics in Göttingen (Germany) with a minor in computer science. After my studies and a stint in the biotech sector, I wanted to go fundamental, at least for a while, and applied for a PhD in stochastic thermodynamics in Cambridge, UK.
The idea was to study the laws of energy and entropy on small scales using a holographic optical tweezers setup. However, lab work, it turns out, requires physical work and I soon found myself automating about every aspect of my optical setup. It essentially became a microrobot which moved tiny particles in parallel across distances of a few micrometres to carry out experiments. It was mesmerizing to watch, and it felt oddly satisfying to check into the lab via remote just to see the machine do what was supposed to be my work. Automation is awesome.
But I wanted to go one step further and started to explore machine-learning approaches to holography itself. I hoped that modern deep learning could help me with this. Fortunately, when I studied the literature, I didn’t have to start from scratch – much of the maths used in statistical physics is also relevant to AI such as linear algebra, probability, and information theory.
It was during this project that I became hooked on AI. I started to think about applying to MSR Cambridge, which I had been aware of for a while because of Christopher Bishop’s (the lab director) book on pattern recognition. But of course, I wasn’t sure if Microsoft would be interested in someone with my background. When I found out that MSR was active in optical data storage systems, I was of course reminded of my own little high-school project and I felt that applying to Microsoft was the right choice – I had to try and so I applied for the job.
MSR Cambridge hosts a huge variety of really interesting research including all flavours of AI, such as natural-language processing or computer vision. The residency is a great opportunity to explore entirely new topics for a while since the projects are cutting-edge but short-term.
Now, I am working on an exciting project in natural-language processing with state-of-the-art neural language processing tools. Together with another resident, I am using this as a platform to develop a novel recommender engine for a Microsoft product.
I have to say, it’s one thing to write a machine learning model on some clean data set, but quite another to build one for a real product with infrastructure and real-world constraints. Also, I didn’t anticipate the sense of fulfilment that comes from working on something that people use. The flipside of this, however, is a huge responsibility and I am very conscious of this.
Microsoft takes privacy and data protection very seriously. Even though we only work on internal test data, for some applications it is important, to experiment in compliance with Microsoft’s strict privacy standards, if the context requires it.
I don’t consider this as a drawback but as valuable experience: Privacy opens up a whole new engineering frontier in the field of AI and usually requires specialised solutions: How do you write algorithms for something you can never ever take a look at – not even when something crashes?
It is exciting to work with the complex infrastructure that Microsoft built for this purpose because I believe that handling of data privacy in AI is a question which will only grow in importance.
To be honest, as someone who had never worked with such specialised server infrastructure, I was intimated at first. However, my confusion cleared soon enough thanks to my supervisors who, whenever necessary, put me in touch with the engineers that are looking after the particular systems my co-worker and I are working on. I believe that learning the nitty-gritty engineering is as important in AI as keeping up with the fast-paced literature. To achieve the latter, we organise a weekly reading group where we present our own progress but also give talks on cutting-edge topics of our choice. The field of AI has been pretty strong during the last of couple of years – for good reasons. There is a steady stream of innovation and some exciting new developments are just beginning. Personally, I am really interested in natural-language processing, a type of model called ‘normalizing flows’, and causal inference.
Of course, there is life outside of work and Cambridge has a lot to offer. During summer, I like to go sailing off the east coast of the UK to recharge and (literally) disconnect. Whenever I find the time I also enjoy creating my own electronic music or you can find me exploring the numerous art galleries in London some of which even feature art work created by AI!
This shows that there are different ways to work with AI. But the one advice that I would give to anyone who is interested is that regardless of background, a good grasp of the fundamentals in mathematics and programming is important. It makes it much easier to keep up with new exciting developments of which I am sure there are many more to come.