We are thrilled to welcome Dr Felipe Tobar to the I-X Team! Felipe joined Imperial College London earlier this month, as a Senior Lecturer in Machine Learning at I-X and the Department of Mathematics.

 

Welcome to the I-X team! Could you tell us a bit about yourself and your academic background?

Thanks! I am very happy to have joined Imperial and I-X and looking forward to the collaborations ahead. My background is on Electrical Engineering. I obtained my undergrad and MSc from Universidad de Chile, and then my PhD in Signal Processing from Imperial’s Electrical Engineering Department in 2014. After that I did a postdoc in the Machine Learning Group at Cambridge and moved back to Universidad de Chile, where I spent the last 9 years as a Research Fellow, then Assistant Professor and, until recently, Associate Professor. During that time, I was part of the Center for Mathematical Modeling and the Initiative for Data & AI, where I had the opportunity to work on the interface between the theory of machine learning and Data Science applications to mining, banking, audio, and astronomy.

What projects are you working on currently?

One of the flagship projects we are working on now is what we called safe sampling. In Generative AI (GenAI), the current state of the art in image generation is based on diffusion models, which, despite their unparalleled ability to create synthetic images, can be hacked into producing discriminatory, violent or offensive content. As these models produce images by means of learning probabilistic models and “sampling” from them, we are set to design a sampling mechanism that operates in a “safe” manner, in the sense that the generated images do not contain offensive or violent content. In this project, I am very fortunate to work with Camilo Carvajal-Reyes, who recently started his PhD in Mathematics at Imperial under my supervision. We have been working during the last months on this problem with promising preliminary results, as a part of a larger research project that aims to align diffusion models with human criteria.

What motivated you to work in this area? And what are the biggest challenges to this project?

As GenAI models become more capable we need to ensure that they are used for the greater good and that in particular they do not constitute a means for violence or discrimination. Also, harmful use of AI systems can either be intentional or unintentional, this becomes particularly relevant since this technology is available to the public and used by non-experts such as underage persons. One of my motivations is making the Internet and the AI toolbox a safer place for families and kids. In the pursuit of this goal, there are multiple challenges including theoretical, related to the solution of the stochastic differential equation governing the data-generation process; technical, involving the data and infrastructure required to train and finetune these models; ethical, when it comes to use and produce violent/sensitive graphic content for supervised learning; and lastly societal, when validating the proposed models in public environments.

What attracted you to working at I-X?

I-X is an ambitious project which materialises Imperial’s current leadership in AI and its strategic vision to advance the field from a rigorous scientific stand. In the full interdisciplinary nature of AI, different expertises, backgrounds, and perspectives converge in I-X, thus making it a unique place to work as a researcher and educator. I am very excited about the multiple opportunities for collaboration provided by I-X through the interaction with other researchers within and beyond Imperial.

What are you interests outside of work?

I am a big music fan. I’ve been playing guitar since I was 13 and try to go to see live music whenever possible; my preferred genres are progressive rock and metal, but I am happy with heavier stuff too. I also enjoy watching films, particularly a nice thriller or a (psychological) horror film at the end of the day. 

When not working, I also try to be more connected with the real world, away from screens and abstract objects. I try to go to the park, for a cycle ride or a roadtrip with my wife and kids, or do some DIY over the weekend. Lastly, though I find little time to do it lately, I enjoy riding my motorcycle either just for commuting or for a day ride out of town: I can’t think of a better way clear up my head and recharge my batteries.