Google has released its code optimization and algorithm discovery agent, AlphaEvolve, to the public.
Google's AI agent ' AlphaEvolve ,' which can optimize existing code and information analysis methods and propose new methods, has finally been made publicly available. AlphaEvolve was only available to a limited audience for over a year after its announcement, during which time it presented numerous solutions to real-world analysis methods and unsolved mathematical problems.
AlphaEvolve is available for everyone | Google Cloud Blog
AlphaEvolve Overview | Gemini Enterprise | Google Cloud Documentation
https://docs.cloud.google.com/gemini/enterprise/docs/alphaevolve/developer-guide/overview?hl=ja
AlphaEvolve is a model described as a 'Gemini-based code optimization and algorithm discovery agent' that helps companies and research institutions solve algorithmic problems. While it was publicly available for a limited time, the model achieved various scientific and mathematical discoveries.
Google's AI 'AlphaEvolve' optimizes DNA analysis, power grids, quantum computing, and logistics in one year: A summary of the achievements of algorithm discovery AI - GIGAZINE
For example, in the field of life sciences, AlphaEvolve is used to improve DNA analysis models, reducing 'mutation detection errors'—where disease-related changes in DNA sequences are overlooked or non-existent changes are mistakenly detected—by 30%.
Chemical manufacturer BASF attempted to optimize its global supply chain by digitally recreating the entire structure, improving its existing planning and forecasting models by more than 80%. Logistics management company FM Logistics optimized warehouse delivery routes, achieving a level of efficiency to the point where further improvement is difficult.
In addition, he has been highly active in a wide range of fields, including successfully rediscovering cutting-edge solutions to approximately 75% of the more than 50 unsolved mathematical problems, improving the accuracy of AI models that convert observational data into a more manageable format for predicting natural disaster risks, and proposing quantum circuits for performing molecular simulations on quantum processors.
AlphaEvolve will be made publicly available as a Google Cloud service. When deploying AlphaEvolve in your own environment, you need to follow these four steps, from problem definition to completing code optimized for production.
- Define: Provides a baseline initial algorithm and problem definition, as well as background knowledge and contextual information about the problem to be solved.
• Evaluation (Measure): Define a scoring function to objectively evaluate candidate programs based on key indicators such as accuracy, performance, and operational constraints.
Optimize: Generates optimized code using AlphaEvolve's agent execution platform.
• Apply: Directly deploy the resulting highly optimized algorithm to production workloads and infrastructure.
Pushmeet Kohli, Chief Scientist at Google Cloud and Vice President of Science at Google DeepMind, said, 'AI is evolving from a productivity tool that accelerates the way we work to a discovery engine that expands the possibilities of what we can do. Tools like AlphaEvolve help researchers and engineers discover groundbreaking algorithms that complement traditional human intuition by autonomously exploring complex computational search spaces.'
Related Posts:
in AI, Posted by log1p_kr