- Anomaly detection
- Media data analysis
Deep Embedded Clustering
Highly accurate, unsupervised clustering of small volumes of failure data
- Even imbalanced data can be clustered accurately using novel evaluation functions.
- Toshiba achieved the world’s highest level of clustering performance for both public data sets and real data.
- Click the “Play” button to play a video on YouTube.
- YouTube is a service unrelated to Toshiba Corp. Use of this service is subject to relevant terms and conditions.
Applications
- Yield analysis at manufacturing sites
Benchmarks, strengths, and track record
- Achieved the world’s highest level of unsupervised clustering accuracy. (ACML2018)
- Unsupervised clustering accuracy for semiconductor manufacturing data with bias in data quantities: 84.1% (conventional methods: 69.9%)
- Unsupervised clustering accuracy for public datasets of handwritten digits (MNIST): 98.4% (conventional methods: 93.8%)
Inquiries
Please include the title “Toshiba AI Technology Catalog: Deep Embedded Clustering” 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. Tao, et al., “RDEC: Integrating Regularization into Deep Embedded Clustering for Imbalanced Datasets,” Proc. of the 10th Asian Conference on Machine Learning (ACML2018), pp.49-64, 2018.
- K. Nakata et.al.; “Reducing analysis work time with ‘Yield News’ support system for yield analysis using big data”; Digital Practice, Vol. 10, No. 2, pp. 304-321, 2019.
- Development of Deep Embedded Learning technologies for highly accurate, unsupervised clustering of small volumes of defect data. (in Japanese)
- Toshiba Review; Vol.72, No.1, p.31 (March 2017) (PDF) *Application (in Japanese)
- Toshiba Review; Vol.73, No.3, pp.18-21 (May 2018) (PDF) *Application
