| Project/Area Number |
20K11955
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Multi-year Fund |
|---|
| Section | 一般 |
|---|
| Review Section |
Basic Section 61030:Intelligent informatics-related
|
|---|
| Research Institution | Iwate Prefectural University |
|---|
Principal Investigator |
藤田 ハミド 岩手県立大学, 公私立大学の部局等, 特命教授 (30244990)
|
|---|
| Project Period (FY) |
2020-04-01 – 2024-03-31
|
| Project Status |
Granted (Fiscal Year 2022)
|
| Budget Amount *help |
¥3,120,000 (Direct Cost: ¥2,400,000、Indirect Cost: ¥720,000)
Fiscal Year 2022: ¥650,000 (Direct Cost: ¥500,000、Indirect Cost: ¥150,000)
Fiscal Year 2021: ¥780,000 (Direct Cost: ¥600,000、Indirect Cost: ¥180,000)
Fiscal Year 2020: ¥1,690,000 (Direct Cost: ¥1,300,000、Indirect Cost: ¥390,000)
|
|---|
| Keywords | 知能情報学関連 / Machine Learning / Health Care System / Deep Learning / Health care System |
|---|
| Outline of Research at the Start |
Activity Recognition on Data Streams using Ensemble Learning for Health warning predictions is project to study several ensemble learning techniques to learn traceable and machine understanding representations and deep neural architectures to recognize entities and relations in data gathered from multi-sensing environment like ECG signals, hear beating and else. Successively, hybrid deep learning and ensemble techniques are to be examined to improve the health recognition on different extracted features. These are to be fused as aggregated prediction system for risk analysis in health care.
|
| Outline of Annual Research Achievements |
In this project, have used ensemble deep learning techniques by constructing Deep Neural Networks (DNNs) based on assembled CNN in architecture of two GPUs in cross layered connection. In addition, we have one GPU system, running as backup for training experiments using large scale data for comparison purpose.
I could achieve good research results using zero shot learning on multi-variate data. Also, I have trained the deep-learning architecture on dynamic data, and image data. The result was promising and therefore we publish international journal articles.
We also, implemented Multi-task Learning-based Attentional Feature Fusion Network for Scene Text Image Super-resolution, with good results.
Although we achieved good results using our reported algorithms (as shown in the list of publications), however, the accuracy for some health care data analytics needs to be enhanced.
The year2022 was hard due to the pandemic as traveling and moving around was tight, also materials become expensive than usual. Nevertheless, the results achieved is good, as evident from the publication enlisted in journals. However, we expect more progressing final outcome in 2023.
|
| Current Status of Research Progress |
Current Status of Research Progress
3: Progress in research has been slightly delayed.
Reason
Evidential form the list of publication reported results, that the developed system could provide high accuracy outcome, by testing it on the public data for learning system prediction, however due to the pandemic traveling restrictions, it was rather not easy to expand the results with European collaborators that I contribute with in terms of the better training mechanism to extract suitable hyper-parameters, and revise the idea in Deep learning based semi-supervised learning methods. My master student “Honda” had done extensive training by tuning of millions of hyper parameters to achieve better accuracy using public image data this is reported in: Kosuke Honda, Masaki Kurematsu, and Hamido Fujita, Ali Selamat "Multi-Task Learning for Scene Text Image Super-Resolution with Multiple Transformers," Electronics, 2022, 11(22), 3813, MDPI, https://doi.org/10.3390/electronics11223813 But such tuning was based on public data, in the year 2022 due to pandemic examining and validating the results with my collaborators in University of Hradec Kralove was tight due to pandemic restrictions. In the year 2023 we are examining for validating the results on real data that some is reported and published: in Dalibor Cimr, Hamido Fujita, Hana Tomaskova, Richard Cimler, Ali Selamat, "Automatic seizure detection by convolutional neural networks with computational complexity analysis," Computer Methods and Program in Biomedicine, Volume 229, February 2023, 107277, https://doi.org/10.1016/j.cmpb.2022.107277
|
| Strategy for Future Research Activity |
The remining part is to confirm the accuracy on the machine learning on real data. This year 2023 is to complete the results and validate it, the future work is relative to reporting the outcome on the year 2022 which has been delayed due to pandemic restrictions. part of the work is submitted to journal: Computer Methods and Program in Biomedicine which is under review. Title: Classification of health deterioration by geometric invariants. It is based on real data done jointly with my international collaborators at University of Hradec Kralove (Czech Republic) the data is enlisted in the link: D. Cimr, D. Bucsovsky, F. Studnicka, H. Fujita, R. Cimler, T. Hayashi,
BCG - patient deterioration impending death, mendeley Data, v2, doi:10.17632/4wrk4fr69w.2, 2023, The results will be expanded using the machine learning models developed in the past years of this project.
|
