22/10/2024

Novello is the new visiting researcher at Google DeepMind.

Tiago Novello, a researcher at IMPA and member of Visgraf (IMPA's Computer Graphics Laboratory), is the new visiting researcher at Google DeepMind – an advanced research company in Artificial Intelligence. The British company, acquired by Google, has as its CEO and director, respectively, Demis Hassabis and John Jumper, winners of the 2024 Nobel Prize in Chemistry.

"I have the privilege of working at an institute that has already had a Fields Medal winner, and now I have the opportunity to collaborate with a company that brings together Nobel Prize laureates. I think this award can further motivate the use of machine learning in other areas of science – something we are already doing at IMPA ," Novello assessed.

With the company's growing interest in the field of computer vision , Novello joins DeepMind to work on problems involving the reconstruction of 3D images from a small number of photos. “If we take a picture of a bottle, for example, we only have an image of the front of it. It would be difficult to reconstruct it without machine learning . The neural network is trained using photos of other objects and can thus understand the back of that bottle.”

Novello was the first Brazilian researcher to be invited to join the Google DeepMind team. The partnership was established for a period of nine months and had the support of researcher André Araújo, who was part of the Google Research team before the merger of the two companies. “I met André last year when we were presenting a paper at the ICCV [International Conference on Computer Vision]. I was walking through the poster sessions and saw his name. I said, 'it has to be Brazilian'. We started talking until he invited me to submit a project to Google DeepMind and I was accepted as a visiting researcher,” he said.

Novello's admission is not the only news. Last month, the IMPA researcher submitted a project to the company proposing the reconstruction of objects using the Gaussian splatting technique. The proposal was accepted, and now IMPA's work, developed by Visgraf, will receive US$20,000 in funding from the company. The funds will be used to secure various collaborations for the project.