- Operation and Control
- Placement and Design
High-dimensional Bayesian optimization that adjusts multiple parameters quickly and automatically
Automatically optimizes multiple parameters that would be difficult to search manually.
- Dramatically reduces human time constraints by automating parameter adjustment tasks.
- It can be used in device design work, for example, HDD servo systems and power devices.
- By appropriately narrowing down training data and scope of searches for high-dimensional parameters, it improves search efficiency and also reduces computational costs.
Applications
- Device design works (for HDD servo systems, power device and so on)
- Identifying system parameters
- Adjust hyper-parameters for machine learning models
Benchmarks, strengths, and track record
- In practical time, adjust more than 100 types of design parameters that are difficult to search using typical Bayesian optimization.
- By applying this method to the adjustment of design parameters for an HDD servo system, compared its performance with that of an HDD servo system with manually adjusted parameters, it reduces parameter adjustment time by two thirds while reducing the head positioning error by 13%.
- By applying this method to the structural design of power semiconductor devices, we developed new structures that reduce on-resistance.
Inquiries
Contact the Toshiba Corporate Research & Development Center
Please include the title “Toshiba AI Technology Catalog: High-dimensional Bayesian optimization that adjusts multiple parameters quickly and automatically” or the URL in the inquiry text.
Please note that because this technology is currently the subject of R&D activities, immediate responses to inquiries may not be possible.
References:
- Y. Taguchi, H. Gangi, "Bayesian Optimization that Limits Search Region to Lower Dimensions utilizing Local GPR," 2023 22nd International Conference on Machine Learning and Applications (ICMLA), Jacksonville, FL, USA, 2023, pp. 202-209, doi: 10.1109/ICMLA58977.2023.00036.
- Toshiba Corporation developed an AI-based "High-Dimensional Bayesian Optimization Technology" to Automatically Optimize many Parameters that are Difficult to Tune Manually -Promoting digital evolution and transformation in data-driven design of high-performance power semiconductors and other advanced applications-; R&D Center; Toshiba (global.toshiba)
- “High-Dimensional Design Parameter Optimization for HDD Servo Systems”; TOSHIBA REVIEW SCIENCE AND TECHNOLOGY HIGHLIGHTS 2023 (global.toshiba) (PDF)
- Hiro Gangi, Yasunori Taguchi, Tomoaki Inokuchi; “Design Automation System for Si Power MOSFETs Using Machine Learning”; Toshiba Review, Vol. 77, No. 6, pp.52-55, 2022.
- H. Gangi et al., "Design Optimization of Multiple Stepped Oxide Field Plate Trench MOSFETs with Machine Learning for Ultralow On-resistance," 2021 33rd International Symposium on Power Semiconductor Devices and ICs (ISPSD), Nagoya, Japan, 2021, pp. 151-154, doi: 10.23919/ISPSD50666.2021.9452194.
