Time-series Deep Learning: New AI Technology that Supports Safe Human Lives

Deep learning, attracting attention as a breakthrough in advancing artificial intelligence

Recently, deep learning is attracting attention as a breakthrough in advancing artificial intelligence (AI). Deep learning technology is, in the field of machine learning which is a central technology in artificial intelligence, a way to automatically interpret and assess phenomena without rules being taught manually.

In the IoT era, massive volumes of time-series data are being accumulated from devices. By classifying the data with a high degree of accuracy, further analyses and applications can be performed, and deep learning is expected to be highly promising as a means to realize it.

Deep learning is a potent machine learning technique, but so far it has only been able to be effectively applied to limited types of data, such as images and speech. In particular, for complex time-series data that is subject to severe oscillations and captured by sensors embedded in IoT devices, it has so far been difficult to achieve highly accurate classifications using deep learning or any other machine learning techniques.

Classifying time-series data as geometric diagrams

Now Fujitsu Laboratories has developed deep learning technology that uses advanced chaos theory* and topology** to automatically and accurately classify time-series data.

Time-series data can be represented as graphical diagrams by using a graphic approach in which characteristic tracks are drawn for each mechanism of movement based on the chaos theory. Features of the diagrams are converted into propriety vector representations using topological data analysis (TDA)*** based on the topology; then a newly designed convolutional neural network learns propriety vector representations, enabling highly accurate classification of complex time-series data.

Using the time-series data from gyroscopic sensors (angular rate sensors) built into wearable devices, this technology was found to achieve an accuracy of approximately 85%, an improvement of about 25% over existing techniques, on the benchmark test**** for classifying human activities. This technology extends the types of data to which deep learning can be applied to time-series data and allows integrated data handling, making it possible to classify time-series data— which people have difficulty discerning—with a high degree of accuracy.

* A field of study that studies complex and seemingly unpredictable phenomena, such as changes in the weather, vibrations in electric circuits, and neural systems of living beings.

** A field of geometry and a set theory. It studies properties of figures that are invariant and different figures under any continuous deformation of mutual disposition and connectivity of figures by ignoring quantitative relationships, such as length and size.

***A data-analysis technique that treats data as a set of points arrayed within a certain space, from which geometric information can be extracted.

****A benchmark test conducted by the UC Irvine Machine Learning Repository.

How this technology classifies time-series data

Creating new value and opening new business areas by deep learning

By classifying time-series data with a high degree of accuracy, it becomes possible to create new value and open new business areas. For example, using IoT devices, it could be used to accurately detect equipment anomalies or forecast breakdowns in factories, or could be used to analyze vital-signs data to assist with medical diagnoses and treatment. In ways such as these, it is expected that artificial intelligence can be applied to a variety of fields. It is also possible to support safe and secure lives of people through preventing failures in facilities and infrastructure by detecting abnormalities or more accurate treatment and preventing diseases.

Fujitsu Laboratories has actively promoted the use of advanced mathematical techniques through industry-university cooperation and acquired the know-how of using mathematical techniques. This new deep learning technology is the result of the accumulation of such techniques. Fujitsu Laboratories will work on further improving the accuracy of its time-series data classification technology with the aim of practical application of it in fiscal 2016 as a core part of Human Centric AI Zinrai. The company is also working to broaden the application of deep learning to types of data other than images, speech, and time series, and to perform more sophisticated data analysis.