ML-Master 2.0 - An autonomous machine learning agent launched by SciMaster

What is ML-Master 2.0?

ML-Master 2.0 is an autonomous intelligent agent developed by the SciMaster team, comprised of members from the School of Artificial Intelligence at Shanghai Jiao Tong University, the Shanghai Institute for Algorithm Innovation, and Deepwise Technology, for real-world machine learning research tasks. Based on the domestically developed open-source large-scale model DeepSeek, ML-Master 2.0 possesses ultra-long-range autonomous capabilities, enabling it to continuously learn through trial and error, accumulate experience, and evolve over extended research tasks. ML-Master 2.0's hierarchical cognitive caching mechanism efficiently manages knowledge and intelligence, significantly improving research efficiency. In OpenAI's MLE-bench test, ML-Master 2.0 outperformed top international teams such as Google and Meta, achieving the world's number one score, demonstrating China's strong capabilities in autonomous AI research. It has already been applied in cutting-edge fields such as embodied intelligence and theoretical physics.

Main functions of ML-Master 2.0

• Independent exploration of long-term scientific research tasksIt can work continuously for tens of hours in complex tasks, exploring the same scientific research goal.
• Experience accumulation and knowledge accumulationLearn from failures, transform experience into reusable knowledge, and transfer it to new tasks.
• Code generation and debuggingIt automatically generates and debugs code, completing a full closed loop of experimental design, code implementation, and result analysis.
• Multi-tasking adaptabilityBy employing a hierarchical cognitive caching mechanism, high-level strategies can be reused across tasks, thereby improving task adaptability.
• High-efficiency resource managementMaintain a stable research pace over long periods of time to avoid context explosion or forgetting historical experience.

Technical principles of ML-Master 2.0

• Ultra-Long-Horizon AutonomyML-Master 2.0 simulates real scientific research processes and designs an ultra-long-range autonomous mechanism that can maintain goal consistency and actively avoid ineffective paths during long-term tasks.
• Hierarchical Cognitive Caching (HCC)Cognition is divided into three levels: Experience, Knowledge, and Wisdom. Experience is used for immediate decision-making, Knowledge consists of verified and stable conclusions, and Wisdom comprises high-level strategies that can be reused across tasks. Each level of cognition plays a specific role in a task, and through dynamic filtering and improvement, valuable information is preserved while noisy information is eliminated.
• Deep exploration and reasoning integrationIt combines exploration and reasoning abilities, selectively capturing and summarizing key information through an adaptive memory mechanism, ensuring that the two reinforce each other.
• Based on the domestically developed open-source large model DeepSeekIt utilizes the domestically developed open-source large model DeepSeek-V3.2-Speciale, combined with high-performance AI infrastructure, to achieve efficient computing and inference capabilities.

Project address for ML-Master 2.0

• Project official websitehttps://sjtu-sai-agents.github.io/ML-Master/
• GitHub repositoryhttps://github.com/sjtu-sai-agents/ML-Maste

Application scenarios of ML-Master 2.0

• Embossed Intelligent Robot TrainingML-Master 2.0 helps robots learn and optimize their behavior strategies autonomously in complex environments, improving their adaptability and decision-making capabilities.
• Theoretical Physics Simulation and DiscoveryThe system can design complex physical simulation experiments, helping scientists discover new physical laws and accelerating the process of theoretical physics research.
• Machine learning engineering tasksML-Master 2.0 can automate engineering tasks such as machine learning model development and optimization, and improve development efficiency by efficiently managing knowledge and experience.
• Complex System Modeling and OptimizationIt is used for modeling and optimizing complex systems such as financial risk models and climate models, and improves model accuracy by adapting to dynamic changes in the system.
• Automation scientific researchML-Master 2.0 can assist scientists in designing experiments, analyzing data, and formulating hypotheses, thereby promoting research and development in fields such as biomedicine and materials science.