In a significant move towards democratizing artificial intelligence (AI) technologies and promoting fairness in AI systems, Fujitsu Limited and the Linux Foundation have announced the launch of two open source software (OSS) projects: SapientML and Intersectional Fairness. These projects were officially unveiled at the “Open Source Summit Europe 2023” in Bilbao, Spain, and are set to transform the world of AI and machine learning.
Fujitsu, a pioneering Japanese AI developer with an impressive track record of 970 AI-related patents between 2014 and October 2022, has been at the forefront of AI innovation. Their commitment to open source innovation marks a significant milestone in their journey to share advanced AI technologies with developers worldwide.
The Significance of Open Source AI
AI technology is rapidly evolving and solving various societal and industrial challenges. However, its development and operation require specialized expertise, and concerns about AI fairness and transparency are rising. Fujitsu recognizes the need for a platform that openly shares AI technologies with engineers worldwide, making AI accessible to a broader community.
Fujitsu is making automated machine learning and AI fairness technologies available as open source projects through the Linux Foundation. This move allows developers worldwide to access Fujitsu’s cutting-edge technology at the source code level, accelerating technological advancement and creating new applications while addressing critical concerns related to AI ethics.
Introducing SapientML: Accelerating Machine Learning Development
One of the two projects, SapientML, focuses on Automated Machine Learning (AutoML) technology. This innovative technology generates code for rapidly building machine learning models using tabular data. Data scientists can leverage SapientML to quickly create precise and interpretable AI models, streamlining the model development process. This technology empowers data scientists to generate highly accurate models and fine-tune them for optimal performance using the generated code.
Intersectional Fairness: Mitigating Biases in AI
The second project, Intersectional Fairness, addresses fairness issues in AI systems. It specifically targets detecting and mitigating biases that may not be immediately apparent. In doing so, it combats discrimination against particular groups that can result from biased training data. This project’s primary goal is to create an AI fairness technology capable of identifying and mitigating “intersectional biases.”
Intersectional biases arise when multiple attributes, such as age, gender, and nationality, interact in complex ways, often overlooked in AI systems. For example, a pass rate difference among young women in an exam might only become apparent when considering the combination of various attributes. Intersectional Fairness technology aims to mitigate such biases while maintaining sufficient accuracy by adjusting the bias for each group and determining the acceptance line from a holistic perspective.
Fujitsu’s Ongoing Commitment to Open Source AI
Fujitsu has been providing its automated machine learning and AI fairness technologies, branded as “Fujitsu AutoML” and “Fujitsu AI Ethics for Fairness,” through the “Fujitsu Kozuchi – Fujitsu AI Platform.” Moving forward, Fujitsu will continue to offer technology updates for these projects on its AI platform.
This collaboration between Fujitsu and the Linux Foundation exemplifies their shared vision of making AI accessible to a global community of developers while upholding principles of fairness and transparency. By launching SapientML and Intersectional Fairness as open source projects, Fujitsu is contributing to the advancement and diffusion of AI, paving the way for a more equitable and innovative future in artificial intelligence.
- The Linux Foundation AI & Data Projects: click here
- “Fujitsu launches AI platform “Fujitsu Kozuchi,” streamlining access to AI and ML solutions to contribute to a sustainable society” (press release, April 20, 2023: https://www.fujitsu.com/global/about/resources/news/press-releases/2023/0420-02.html)