Improving artificial intelligence infrastructure
AlphaEvolve has been pilot tested and has become a key element of our infrastructure. AlphaEvolve was used as a regular next-generation design optimization tool TPU. It also helped discover greater efficiency cache replacement rulesachieving in two days what previously required a coordinated, human effort spanning months.
“” – Jeff Dean, Chief Scientist, Google DeepMind and Google Research
AlphaEvolve has improved performance Google key by improving it Log-structured merge tree compaction heuristics. This optimization reduced “write gain” – the ratio of data written to memory compared to the original request – by 20%. She also provided insights for new compiler optimization strategies which reduced the amount of disk space occupied by software by almost 9%.
Scaling commercial applications
Together with Google Cloudwe are now bringing AlphaEvolve capabilities to a variety of commercial enterprises across a variety of industries.
- In financial services, Klarna used the system to optimize one of its largest transformer models – doubling the training speed while improving model quality.
- In semiconductor production Subsoil applied AlphaEvolve to its computational lithography environment, achieving a multi-fold augment in operating speed, enabling much larger simulations of advanced semiconductors.
- In logistics FM logistics used the technology to optimize intricate routing challenges such as the traveling salesman problem, finding a 10.4% improvement in routing performance compared to previous highly optimized solutions, saving over 15,000 kilometers traveled per year.
- In advertising and marketing, WPP used AlphaEvolve to refine AI model components, navigate intricate, multi-dimensional campaign data, and achieve a 10% augment in accuracy over competing manual model optimizations.
- In computational materials and life sciences, Schrödinger used AlphaEvolve to achieve approximately 4x speedup in both training and inference using machine learning force fields (MLFF).
— Gabriel Marques, technical lead for machine learning at Schrödinger.
The future of AlphaEvolve
The past year has shown how AlphaEvolve is quickly becoming a versatile, general-purpose system. It shows that the next breakthroughs will be driven by algorithms that can learn, evolve and optimize themselves. As we look to the future, we are excited to expand these capabilities and leverage the power of this technology in the face of an even broader set of external challenges.
Thanks
AlphaEvolve was developed by Matej Balog, Alexander Novikov, Ngân Vũ, Marvin Eisenberger, Emilien Dupont, Po-Sen Huang, Adam Zsolt Wagner, Sergey Shirobokov, Borislav Kozlovskii, Francisco JR Ruiz, Abbas Mehrabian, M. Pawan Kumar, Abigail See, Swarat Chaudhuri, George Holland, Alex Davies, Sebastian Nowozin and Pushmeet Kohli. This study was developed as part of a broader initiative focusing on the apply of artificial intelligence for algorithm discovery. After initial development, Alexey Cherepanov, Anindya Basu, Becky Evangelakos, Jamie Smith, and Mario Pinto joined the team to expand AlphaEvolve’s impact.
Adam Connors, Alex Bäuerle, Anna Trostanetski, Fernanda Viegas, Gabi Cardoso, Jonathan Caton, Lucas Dixon, Mariana Felix, Martin Wattenberg, Matin Akhlaghinia, Richard Green, Yosuke Ushigome, and Yunhan Xu worked with our team to develop the AlphaEvolve UI, with support from many others.
Anant Nawalgaria, Diego Ballesteros, Gemma Jennings, Jakob Oesinghaus, Kartik Sanu, Laurynas Tamulevičius, Nicolas Stroppa, Nishta Dhawan, Oliver Hilsenbeck, Reah Miyara, Skander Hannachi, Tom Beyer, and Vishal Agarwal worked with our team to develop the AlphaEvolve API and connect with Google Cloud customers, with support from many others.
We are grateful to our collaborators for leading applications of AlphaEvolve to critical problems and contributing to this report: Aaron Wenger, Abhradeep Guha Thakurta, Akanksha Jain, Alex Vitvitskyi, Amir Yazdan Bakhsh, Andrew Carroll, Aranyak Mehta, Arthur Conmy, Anshli, Anshli Martin, Pere Martin, Eric Hasler Thurston, Hongzheng Chen, Jack Mason, János Kramár, Jeremy Ratcliff, Jessica Sapick, Johannes Bausch, Jonathan Katz, Kevin Miller, Kim Stachenfeld, Mark Kurzeja, Mircea Trofin, Myriam Khan, Nero Geng, Samuel Castro, Petar Veličković, Pi-Chuan Chang, Praghav Pablo Raghav Raghav Gupta, Rohin Shah, Sasha Vezhnevets, Sébastien Lahaie, Sergio Guadarrama, Shravya Shetty, Shruthi Gorantala, Terence Tao, Todd Lipcon, Tom O’Brien, Vinod Nair, Ziyue Wang, Zun Li and many other AlphaEvolve users.
Finally, we thank our management for their guidance and support: Amin Vahdat, Ankur Jain, Demis Hassabis, Jeff Dean, Parthasarathy Ranganathan, Pushmeet Kohli, Saurabh Tiwary and Sundar Pichai. We also express our gratitude to our partner teams in Google DeepMind, Google Cloud, Google Labs, Google Research, and other product areas for enabling AlphaEvolve-powered applications and products.