How I've Developed as a Researcher

and How that Benefits my Industrial Partner

My work focuses on using data provided by my industrial partner, collected from their manufacturing process over the life-cycle of production runs, and modelling this data such that any opportunities for optimisation can be uncovered. This change to applying my work to a real-world case was a difficult hurdle. Working with real-world data is very different to anything I had done while studying Physics as an undergraduate student, where everything is clean and sanitised, able to be done in a few seconds/minutes with pen and paper.

Now, a lot of data needs to be preprocessed, converted, and so on, and takes a considerable amount of time to run the code that does the modeling. Then, if any errors show up, this process has to be repeated.

This stage is just the foundation for analysis, where I aim to carry out the optimisation aspect of my work. This poses a new problem, which is having to essentially translate between real-world cause-and-effect, and mathematical theory. Obviously a manufacturing process is designed and set up with zero consideration for a potential future machine learning analysis, so this inserts a somewhat abstract layer between the two, and my job basically is to work in and communicate through that layer.

For my industrial partner, this has been quite a fresh take on the analysis of their process. Lots of companies/industries in the past few years have been hearing more and more about Machine Learning/Data Science (ML/DS), they don’t really know what the terms mean, but definitely see competitors having the same idea, so worry about being left behind. In my case, working with a company involved in manufacturing, an enormous amount of data collection goes on, but it’s all either analysed by hand, or goes unused. Building ML/DS methodology into an already existing workflow, especially one that is expensive to start AND run AND stop for maintenance is quite difficult. So, if any pauses like this were to be done, you must be seriously confident that the payoffs will be worth it, you can’t just try repeatedly and brute-force something usable.

What both myself and my industry partner have found rewarding about this was that, while it’s tough to start off trying to model in a domain you have no familiarity with, it meant that I had to start off with zero knowledge of things, which normally wouldn’t be great, but from a machine learning perspective, means I can model everything and anything without any subconscious biases dissuading me from trying certain techniques/datasets/models. If I were to have a thorough understanding of the production process, it’s more likely I would use that and, possibly without even realising it, make prior assumptions that aren’t suitable for the task at hand. This way, I can continue exploring new avenues that traditional analysis of the industry process have yet to find.

I’d like to think I picked up a lot of transferable skills throughout my undergraduate degree – reviewing literature, problem-solving, and so on. Moving to a project based in machine learning presented a unique set of challenges I had not encountered up to that point.

At every other level of education, the work required is presented to you as a clear step-by-step roadmap. At a PhD level, this approach didn’t really work well for me, considering the large amount of domain knowledge I was at that point unfamiliar with. Because of this, my first year consisted of what felt like spinning multiple plates, learning the basics and advanced theory of multiple different topic areas, tools, programming languages etc., some overlapping, most not. It can be a really overwhelming experience for many. And during this time, it becomes very easy to look around and feel like you’re falling behind because everyone else is at different stages in their own work, and takes a while for you to realise how different any two degrees are, and how these comparisons you’ve drawn don’t really hold up to any scrutiny.

The group holds a monthly seminar series for anyone to present their work, recent publications etc., which has recently expanded to inviting researchers from other departments or universities. The main barrier to collaboration across disciplines is simply not being aware of the work that goes on beyond your own. Having guest speakers come in to share their research with us is helping to overcome that. These are a relatively recent development, but I have already been in touch with one or two of the past speakers, following up on their presentations to request some more information, resources, etc. so I’m then able to look into that research myself, and look to start some collaborative work outside the CDT/my field/the university.

My aim is that hopefully, by the end of my PhD, I can have a fully realised project, results, thesis, everything expected of a postgraduate researcher, but alongside these, a good track record of collaborating beyond my immediate group and topic/research area. It can sometimes feel like you’re a little limited or kettled into your specific topic, and the idea of learning a different branch of your discipline for either little reason other than interest, or to have the possibility of collaborating on other work, both can seem like an inefficient use of time, but in a field like this, it’s quite commonplace, and when eventually moving to industry, will at times be more of a necessity. It begs reminding that the point of a PhD, especially one in a CDT, is to develop yourself as a researcher, and immediate results specific to one piece of work aren’t the be-all-end-all, otherwise it would be more convenient to recruit someone like a data scientist who would devote their entire focus on a single task.

I remember around the start my PhD being told that it’s “the most ‘free’ research you’ll ever get to do”, meaning that once you move to industry, or post-doctoral research, you have your domain and your project/task, and that is your focus until it’s finished. Whereas throughout the course of a PhD, you’re constantly exploring different topics, being invited to training courses regardless of how, for want of a better word, ‘useful’ it will be to your project.

Beyond the seminars mentioned, we have less frequent events like the CDT Showcase and ATI events that are larger in scale and place more of a focus on expanding your network to beyond that of the CDT. They both serve as a great opportunity to meet students, researchers, industry representatives and so on, far outside the circle you would normally have the opportunity to meet.

One of the strengths of the CDT programme is that it, by design, places a great amount of focus on your development as a future researcher, in terms of the skills developed, the connections you make, the domain knowledge you pick up, and more, and not just working in isolation on your project.