School of Technology PhD student, Andrei Margeloiu, worked for Huawei Research UK on a project to investigate novel methods for compressing data (photos and videos) without affecting the quality. The fellowship complements Andrei’s ongoing PhD in Machine Learning.

I self-found the fellowship . I wanted to work with Huawei because they are one of the global leaders in Artificial Intelligence. I chose this project in particular because I am interested in generative modelling, and I wanted to learn more about it. The project offered me the opportunity to work on the cutting-edge of generative modelling.

Within two weeks, I fully accommodated the work environment. My project supervisor presented me with the working environment, the required software and the necessary internal tools (e.g., GPU cluster, internal Github etc). I mainly worked with people with a similar background in Computer Science and Mathematics, and I had a great experience.

With the rapid adoption of high-quality cameras, users generate an increasing number of photos and videos. The storage cost of user-generated content is becoming prohibitive. Cloud storage providers (such as Huawei) seek to reduce the cost of storing users’ data. My fellowship project aimed to investigate novel methods for compressing data without affecting the quality, thus reducing costs for storage providers.

We investigated the proposed hypothesis, but we obtained negative results. The fellowship unfortunately did not lead to publication but has nevertheless been an incredibly worthwhile experience.

The skills and experiences acquired in the fellowship will serve in my PhD and future career. I learned how to run large-scale computational experiments on GPU clusters. I understood the usefulness of rapid prototyping by working alongside experienced researchers. This fellowship validated my hypothesis that I would like to become a researcher in an industrial lab.

The fellowship made a big difference in my research by teaching me to prioritize quick experimentation. For example, I learned that tagging each experiment will make it easier to analyze the results later. I had the opportunity to run hundreds of experiments during the three months, and I developed skills for developing sound hypotheses and experiments. With hindsight, the research presented significant uncertainty on its own, and I would have benefited from a clearer structure and a backup plan if the hypothesis turned out to be negative, which was the case.

In terms of my future career, this fellowship validated that I would like to be a researcher in an industrial lab. I enjoyed being around researchers and working on technical challenging problems.
The fellowship taught me numerous transferable skills, the most important being:

  • Crafting sound hypotheses and experiments
  • Presenting and disseminating research findings
  • Working in teams and pursuing a common research agenda

I used my previous academic knowledge (in machine learning algorithms and the related literature) to contribute ideas and hypotheses to the investigation. I provided a new perspective to the algorithm by bringing my experience and knowledge along with my enthusiasm for working on this project with the corporation.

I would recommend doing fellowships because it provides relevant work experience and a low-risk way to validate whether a specific career suits you or not.

Andrei Margeloiu’s placement with Huawei Research UK took place between September and December 2021

Pin It on Pinterest

Share This