Citadel AI Inc., the leader in trustworthy AI infrastructure, and Suntory System Technology Ltd. (100% owned by Suntory Holdings Ltd.) have successfully completed a joint verification of Citadel AI’s automated AI monitoring and testing products.
Based on this verification, Citadel AI and Suntory Holdings have entered into a capital and business alliance agreement on Wednesday, March 30.
Moving forward, the Suntory Group will promote the use of Citadel AI’s systems as the data governance foundation across its company-wide AI platform. This platform will enhance the training and operational performance of Suntory’s AI applications across the ML lifecycle, leading to business growth and profitability.
Background of the joint verification
The Suntory Group uses AI extensively in R&D, SCM, and other fields, including production planning  and anomaly detection on production lines. Their goal is to maximize the performance of AI in real-world use cases, but this is difficult due to challenges such as data drift and model drift, as Japanese government agencies have also documented in their AI guidelines .
For AI systems to continuously provide business value, it’s critical to continuously validate the quality of input data and output predictions, ensure that models are being properly re-trained, and monitor real-world model performance across time, model versions, and dataset versions. Before the introduction of Citadel AI’s products, these steps were completed manually, which required significant engineering time and cost.
 Suntory Foods and Hitachi have developed an AI-based production planning system. https://www.suntory.co.jp/softdrink/news/pr/article/SBF0756.html
 From “Machine Learning Quality Management Guidelines,” AIST, 2020. https://www.aist.go.jp/aist_j/press_release/pr2020/pr20200630_2/pr20200630_2.html
Details of the joint verification
Suntory System Technology Ltd., the Suntory Group’s IT subsidiary, has successfully verified Citadel AI’s products in Suntory’s R&D and SCM applications. Our products help Suntory to:
- Instantly and comprehensively reveal AI weaknesses and monitor real-world model performance, which are difficult and time-consuming to detect manually.
- Easily and visually analyze AI problems including model bias, data drift, and explainability at training-time and serving-time.
- Compare and improve model performance across versions through our version management and data logging functionality.
- Can be easily installed and do not require major modifications to Suntory’s existing AI models and development workflows.
- The above features can be utilized as an integrated, managed service.
Moving forward, the Suntory Group will promote the use of Citadel AI’s systems as a data governance foundation in its company-wide AI platform to maximize business value.
About Citadel AI’s products
Unlike traditional software, the accuracy of AI software quickly deteriorates in a real-world, rapidly changing environment. It’s essential for companies to have quality management tools in place to maintain AI reliability and manage business and security risks.
Citadel AI offers two easy-to-use products that improve reliability at different steps of the ML lifecycle: Citadel Radar (serving time) and Citadel Lens (training time). Our products can be used as a hosted SaaS service, in a private cloud, or on-premise.
In contrast to low-code tools that automate AI model creation, Citadel AI’s products validate and monitor AI models across the full ML lifecycle, from testing to deployment.
Citadel Radar and Citadel Lens follow the principles of “data-centric AI”, and can be applied to any AI model through a black-box testing approach, without modifying the customer’s existing AI models and development workflows.
- Create and share Lens Model Reports from a simple UI, or automatically test models with an API (for CI/CD pipelines).
- Automatically reveal model weak spots such as model bias, robustness, backward compatibility, etc.
- Actionable insights: label this data next, prioritizing model/data mistakes, and enforcing QA standards across a company.
- Automatically analyze and send alerts for data drift, model drift, and prediction quality (in individual customer segments) over time.
- Improve prediction quality by validating new data points in real time, with hundreds of automated tests for invalid, outlier, and adversarial data.
- Easily explain any prediction, of any model, at any time – from feature attributions to counterfactuals.
About Suntory System Technology Ltd.
- Representative Director: Hiroaki Kawamura
- Headquarters: Dojima, Kita-ku, Osaka
- Established: March 29, 1990
- Company URL: https://www.suntory.co.jp/sst
About Citadel AI Inc.
Citadel AI is an MLOps startup based in Tokyo, and raised seed funding from UTokyo IPC and ANRI in 2021. The co-founder & CTO was the former product manager for ML infrastructure at Google Brain, and is currently leading Citadel AI’s product and engineering teams.
We’re actively recruiting talented software engineers to join our team – apply here!
|December 10, 2020