It’s no secret that the gender gap still exists within STEM. Despite a slight increase in recent years, studies show that women only make up about a quarter of the overall STEM workforce in the UK. While the reasons vary, many women report feeling held back by a lack of representation, clear opportunities and information on what working in the sector actually involves.
Closing the gap within STEM is not a quick fix but a collective effort of everyone in the industry. Various organisations like Women in Machine Learning (WiML) actively work to help create a more inclusive environment where the successes of women are amplified. They also stand as an important point of information for the many women who want to learn more about what it’s like to work in STEM.
That’s why for this year’s International Women in Engineering Day, we asked the WiML community to share with us the most common questions they receive about technical interviewing. To share their perspectives and to discuss what it’s actually like to work at DeepMind, we brought together Mihaela Rosca (Research Engineer), Feryal Behbahani (Research Scientist) and Kate Parkyn (Recruitment Lead - Research & Engineering).
Mihaela: It’s not uncommon to have self doubts or feel as if you’re under prepared for a position in the field. There will never be a perfect time to apply and you can easily convince yourself that there’s more to learn but that shouldn’t be a deterring factor in your decision to apply.
Of course the right skillset will depend on the specific role you’re after, but if you’re keen to work on the future of machine learning research, read research papers and implement state of the art algorithms - you’re ready...so apply!
Kate: We recruit for many roles across the organisation so the qualities we focus on differ accordingly.
The majority of research scientist hires we make are post PhD level, so we don’t over index on publications. We also don’t have a specific marker for degree achievement or GPA. When it comes to experience, we’re always interested in reading about a candidate's past internships and/or voluntary industry experiences. We look for proven ability not only in ‘research’ but also in implementation, engineering and application. Reading about side projects and open source contributions are also great to see when looking at potential candidates, so feel free to link your Github, side projects or code.
For research engineers, it’s important to remember that the role is part research and part engineer, so we’re always looking for people that enjoy putting theory into computational form.
For software engineers, we look for the clear ability to communicate problems and solutions. Software engineers at DeepMind regularly deal with ambiguous problems which also have underlying engineering complexities. Evidence of working on similar projects, or experiences in accelerating research and harnessing tools to augment research, is key.
Kate: Creating the perfect CV or resume is a big job. Luckily there are a countless number of resources out there that can help you get the job done. To keep it simple, we’d suggest focusing on the following points:
Feryal: There are a wide range of resources available to help you learn and develop your skills in machine learning. These include open-access introductory courses on YouTube (i.e. Nando de Freitas’s course on Deep Learning, David Silver’s course on Reinforcement Learning and DeepMind x UCL Lecture Series), blog posts which provide overviews of particular techniques (e.g. Distill) and more advanced machine learning conference proceedings such as NeurIPS, ICML and ICLR.
There are also a number of summer schools (i.e. MLSS and DLRLSS) that help support students and professionals who are interested in learning from leading experts in the field. Many of the summer schools also host videos and practical exercises from previous years which can act as excellent resources for learning at your own pace.
It’s also great to look to organisations like Women in Machine Learning (WiML) that specifically help women in the field build their technical confidence and voice while amplifying their achievements to the wider community.
Feryal: The interview process at DeepMind can vary depending on the particular role you’re applying for. From my experience, the interview process for a Research Scientist role consisted of four phases:
This is to cover your background, experience, the motivation for applying and future plans. At this stage, you will also have the opportunity to ask any questions that you may have about the role or the interview process.
This part of the process involves several sessions - including one with a technical quiz that covers a large breadth of topics in computer science, statistics, mathematics and machine learning. It’s key that you revise broadly for this session! At this stage there will also be a coding interview where you [in your chosen language] will have to work through a few questions and a specific problem with the end goal of coming to a solution implementation.
This stage is made up of various short [i.e. ~30min] interviews with researchers and leads about your specific research background and interests. Here you will have the opportunity to give a talk about your research, which gives the interviewers a better idea of your overall research direction. At this point, try to show your technical understanding of the field and feel free to bring up your own achievements and research ideas. It’s not necessary, but I would also suggest reading through recent papers published by the DeepMind team to try to frame your strengths better!
Towards the end of the interview process, you will once again connect with the recruitment team to discuss DeepMind’s culture and mission. I recommend that you read about DeepMind’s mission and think about how your career goals can fit within it.
Mihaela: Due to the versatility required to do machine learning research, the interview process has a relatively even split between coding and assessing research skills. The first stage focuses on mathematics, statistics, machine learning and computer science knowledge, while later stages focus on coding. Keep in mind that throughout the interview process, the interviewer is trying to assess your problem solving skills, so focus on communication and explain your answers.
For my own interview, I prepared by reviewing some of the notes from my university lectures - including a statistics course I had taken. At the time I didn’t know a lot about reinforcement learning, so I did some additional research and watched David Silver’s UCL course on the topic. For my coding interview, I chose python. To prepare and to practice my speed I solved a few coding questions without using an integrated development environment (IDE) or my favourite editor - only a simple text editor.
Mihaela: Absolutely! Research Engineers at DeepMind - and elsewhere - often lead projects of all sizes. They can lead as first authors of conference papers, or as larger team efforts which involve groups of different sizes and take place over multiple months.
There are plenty of examples, but here are a few: AlphaZero, improving exploration in reinforcement learning using generative modeling, and open sourcing of core libraries such as Reverb.
Feryal: Being a research scientist means that my day never really looks the same. My time is often spent thinking about my research projects, coding, meeting and discussing ideas with others, reading papers and attending presentations or reading groups.
As always in research, what I’m doing can change depending on if I’m working towards a paper deadline, working on a specific project, or thinking about what to do next. Luckily DeepMind is really flexible in how one can organise their time and schedule. We use a “milestone system” which organises research into smaller, measurable chunks (e.g. 3-6 weeks) so this really helps with planning research and breaking it down into concrete steps.