Petar Veličković, a research scientist at DeepMind, is gearing up to present his paper, The CLRS Algorithmic Reasoning Benchmark, at ICML 2022 in Baltimore, Maryland. His journey to DeepMind began during his undergraduate studies at the University of Cambridge, where he witnessed DeepMind’s historic victory over a top Go player. This inspired him to join the company and contribute to its groundbreaking work.
In his role as a research scientist, Petar engages in a continuous cycle of learning, researching, communicating, and advising. He stays up-to-date with the latest developments in the field, explores new ideas with his team, and employs theoretical analysis and programming to validate their hypotheses. The results of their research are shared through papers and presentations, empowering others to make use of their findings.
At ICML, Petar will be presenting The CLRS algorithmic reasoning benchmark, a research project that focuses on neural algorithmic reasoning. They task graph neural networks with executing a variety of algorithms from the Introduction to Algorithms textbook. This benchmark aims to address the challenge of tracking progress and entry barriers in this emerging field by providing a readily accessible dataset and code.
The ultimate goal of Petar’s research on algorithmic reasoning is to enable the execution of classical algorithms using high-dimensional neural executors. This would open up possibilities for data-efficient reinforcement learning, where classical algorithms can be directly applied to raw or noisy data representations. They have already published a working prototype at NeurIPS 2021, and Petar is eager to see what the future holds.
Aside from his work at DeepMind, Petar is enthusiastic about the ICML Workshop on Human-Machine Collaboration and Teaming. He believes that the true power of AI lies in its synergy with human domain experts. He has also been actively involved in outreach efforts, teaching a course on Geometric Deep Learning at the African Master’s in Machine Intelligence (AMMI) and organizing the EEML Serbian Machine Learning Workshop in his hometown of Belgrade. These initiatives aim to foster diversity, equity, and inclusivity in the AI community.
As Petar prepares for ICML, he emphasizes the importance of reflecting on diversity and inclusion in the field of AI. While hybrid and virtual conference formats have made events more accessible, there is still work to be done to ensure that underrepresented communities are included in the conversation. Petar’s dedication to teaching and outreach demonstrates his commitment to building a diverse and vibrant AI ecosystem.