Explainable AI for mission-critical workloads with Machine Learning for IBM z/OS
Meeta Yadav Vouk; Elpida Tzortzatos; Shuangyu; Deng Ke Zhao; Abid Alam
We all are witnessing the transformative power of AI and the promise of exponential benefits it could bring to a wide range of applications and AI systems have become increasingly attractive to many enterprises. However, the broad adoption of AI systems will require humans to trust their output by understanding how it works and being able to assess that it’s reliable and compliant with the regulations.
Today, we are excited to share that the initial release of the Integrated Trustworthy AI capabilities — AI explainability is available in Machine Learning for IBM z/OS v3.1. This milestone represents a crucial step towards empowering organizations to fully leverage AI technologies for their mission-critical workloads on z/OS while ensuring the utmost trust, safety, and confidence.
The Trustworthy AI component of MLz brings a comprehensive suite of tools and capabilities designed to evaluate, monitor, and enhance the trustworthiness of AI models deployed to z/OS applications, following organizational regulations and requirements. Seamlessly integrated with the AI model lifecycle management and operationalization in MLz, the Trustworthy AI component is effortless to install, and configure and has minimal requirements on system resources.
We are committed to an iterative approach to delivering trustworthy AI capabilities. With this initial release in Q4, we are introducing the AI model explainability capability, which provides insights into the critical factors influencing the outcomes of both machine learning models and deep learning ONNX models. This empowers organizations to understand and interpret the decision-making process of their AI models.
Looking ahead, we have ambitious plans to expand the trustworthy AI capabilities in the year 2024. Our roadmap includes adding capabilities for drift detection, robustness assessment, and fairness evaluation. These forthcoming enhancements will further strengthen the trust and reliability of AI models deployed on z/OS.
Another notable addition in this 4Q update to MLz v3.1 is the integration of the Snap ML into MLz with end-to-end lifecycle management and operationalization. The integration of Snap ML brings high-performance training and inferencing of the most popular machine learning models to z/OS.
This feature has been thoughtfully designed to cater to various usage scenarios within MLz. Not only can users take advantage of the power and efficiency of Snap ML to train models in MLz, but they can also import the machine learning models trained elsewhere to MLz for high-performance inferencing using Snap ML whenever possible. In situations where the MLz scoring server is running on a z16 machine, the on-chip AI accelerator can also be leveraged for optimized performance.
To access the details of the PTFs that carry the above new features of MLz v3.1, you can visit the support page at: https://www.ibm.com/support/pages/node/7003489
To learn more about the Trustworthy AI capabilities, check out this detailed technical blog here and MLz documentation.
