Loading...
Please wait, while we are loading the content...
Similar Documents
Transfer Learning for Mining Digital Phenotype by SAS® Viya®
| Content Provider | Semantic Scholar |
|---|---|
| Author | Fujita, Satoki Kiguchi, Ryo Yoshida, Yuki Hirano, Katsunari Kitanishi, Yoshitake |
| Copyright Year | 2020 |
| Abstract | With the remarkable development of various AI and machine learning methods in recent years, various advanced technologies based on collected data have been born in each industry. One of the technologies represented by them is deep learning. Image recognition, speech recognition, and natural language processing are well-known usage applications for deep learning. Computers automatically extract important features that affect the results from data without human intervention. This method achieves recognition and identification accuracy that is much higher than that of the method, and is not inferior to humans. By using this, from data collected from digital devices that are now widespread worldwide, such as posts to social media such as SNS and blogs, call logs from smartphones, and accelerometer data obtained from sensors, it may be possible to discover and quantify the human phenotypes, so-called digital phenotypes, that characterize the owner, which cannot be easily discovered by human hands. However, the implementation of deep learning requires a large amount of training data. Although barriers to obtaining big data have disappeared from several years ago, there are still many cases where it is difficult to obtain a sufficient amount of data depending on the target problem and the environment where it is located. One way to solve such a problem is transfer learning. In this paper, we implemented transfer learning in SAS® Viya etc. as a method that can be expected to be applied to digital phenotyping in the situation where the amount of data is insufficient. In particular, we examine the usefulness of transfer learning by dealing with cases where deep learning is used to determine whether or not a sticky note is attached in a document image. Furthermore, we try to find out the characteristics of the author and the story by applying transfer learning to the case that identifies the author from a sentence of the book. It imitates mining phenotype. |
| File Format | PDF HTM / HTML |
| Alternate Webpage(s) | https://www.sas.com/content/dam/SAS/support/en/sas-global-forum-proceedings/2020/5029-2020.pdf |
| Language | English |
| Access Restriction | Open |
| Content Type | Text |
| Resource Type | Article |
