Fujitsu Laboratories, a subsidiary research center at Fujitsu Limited, announces the development of a deep learning technology that analyzes time series data - a collection of observations made sequentially over time - with a high level of accuracy. Such data can be subject to severe volatilities, which makes it difficult for people to standardize them. Considered a promise for Internet of Things applications, the novelty allows the information to be analyzed in a deeper way, with the perspective of creating new values and opening up new business areas.
Deep learning technology is an area of research related to machine learning based on the adoption of a multi-layered neural network, allowing a computer to identify patterns and automatically organize information and learn tasks. It has attracted attention as an innovation in the advancement of artificial intelligence and has achieved extremely high accuracy in image and language recognition. However, the types of data it can be applied to still make it limited. This is due to the complexity of accurately and automatically classifying data from volatile time series, such as extracted from IoT devices, which are difficult to have patterns discerned by people.
The deep learning approach developed by Fujitsu Laboratories uses an advanced mathematical technique and extracts geometric characteristics from time series data, allowing for an accurate classification of varying time intervals. In everyday life, for example, the use of technology could be used to accurately detect anomalies in equipment, predict interruptions in a factory, or to analyze vital sign data in order to assist medical diagnoses and treatments. In this sense, the expectation is that the technology will bring advances to a series of segments through artificial intelligence.
In tests conducted at UC Irvine Machine Learning - a world-renowned repository that provides numerous data sets for comparative machine learning assessments - that classified gyroscopic time series data on wearable devices, the new technology achieved approximately 85% accuracy, an improvement around 25% compared to existing technology.
Deep learning technology is an area of research related to machine learning based on the adoption of a multi-layered neural network, allowing a computer to identify patterns and automatically organize information and learn tasks. It has attracted attention as an innovation in the advancement of artificial intelligence and has achieved extremely high accuracy in image and language recognition. However, the types of data it can be applied to still make it limited. This is due to the complexity of accurately and automatically classifying data from volatile time series, such as extracted from IoT devices, which are difficult to have patterns discerned by people.
The deep learning approach developed by Fujitsu Laboratories uses an advanced mathematical technique and extracts geometric characteristics from time series data, allowing for an accurate classification of varying time intervals. In everyday life, for example, the use of technology could be used to accurately detect anomalies in equipment, predict interruptions in a factory, or to analyze vital sign data in order to assist medical diagnoses and treatments. In this sense, the expectation is that the technology will bring advances to a series of segments through artificial intelligence.
In tests conducted at UC Irvine Machine Learning - a world-renowned repository that provides numerous data sets for comparative machine learning assessments - that classified gyroscopic time series data on wearable devices, the new technology achieved approximately 85% accuracy, an improvement around 25% compared to existing technology.
EN 
PT 
