Exploring Algorithmic Reasoning with Petar Veličković at ICML 2022
We recently sat down with Petar Veličković, a research scientist at DeepMind, who, along with his co-authors, will be presenting their paper “The CLRS Algorithmic Reasoning Benchmark” at ICML 2022 in Baltimore, Maryland, USA.
My journey to DeepMind…
During my undergraduate studies at the University of Cambridge, I was fascinated by the limitations of modern deep learning systems, particularly when it came to playing complex games like Go. Witnessing DeepMind’s historic victory in a challenge match against a top Go player in 2016 sparked a fire in me. From that moment on, I was determined to join a company that could achieve the seemingly impossible. In early 2019, after completing my PhD, I embarked on my journey as a research scientist at DeepMind.
My role…
My role at DeepMind involves a continuous cycle of learning, researching, and communicating. I am constantly seeking out new knowledge, exploring areas like Category Theory, and collaborating with my team to translate our findings into impactful solutions. By presenting our work at conferences like ICML and mentoring students, I strive to drive innovation and create meaningful connections within the scientific community.
At ICML…
At ICML, we will be shedding light on our groundbreaking research on algorithmic reasoning, specifically our paper on “The CLRS algorithmic reasoning benchmark.” This benchmark aims to streamline efforts in neural algorithmic reasoning by tasking graph neural networks with executing a diverse set of algorithms from the “Introduction to Algorithms” textbook. Our goal is to foster collaboration and progress in this rapidly evolving field.
The future of algorithmic reasoning…
Our research ambitions revolve around embedding classical algorithmic computation into neural networks, enabling them to operate on raw data efficiently. This approach holds immense potential for advancing data-efficient reinforcement learning, bridging the gap between classical algorithms and modern computational challenges. Our blueprint for this integration has already shown promise, and we are excited to explore its applications further.
I’m looking forward to…
One aspect I am eagerly anticipating at ICML is the Workshop on Human-Machine Collaboration and Teaming, which aligns closely with my belief in synergistic AI-human partnerships. These collaborations have the power to revolutionize various domains, as demonstrated in our recent work on empowering mathematicians using AI. I am also honored to participate in a panel discussion with esteemed colleagues, including Sir Tim Gowers, a figure I deeply admire.
Looking ahead…
As we navigate through major conferences like ICML, I am committed to championing diversity and inclusion in AI. By teaching courses on Geometric Deep Learning in Africa and contributing to the Eastern European Machine Learning community, I aim to nurture a diverse and empowered cohort of AI practitioners. Together, we can shape a more inclusive and innovative future for artificial intelligence.