講演抄録/キーワード |
講演名 |
2020-12-17 16:30
少量データにおける不変性を持つ潜在特徴量の抽出手法 ○Mohit Chhabra・Quan Kong・Tomoaki Yoshinaga(日立) PRMU2020-49 |
抄録 |
(和) |
The remarkable effectiveness of neural networks on vision tasks has led to an interest in adapting neural network models to limited data cases. It is also desired that low dimensional representations of the data efficiently represent the data distribution. We propose to minimize ordinal energy of the code produced by encoder model of de-noising auto-encoder and add stochastic non-linear units. Proposed modifications lead to an increase in the classification performance in the semi-supervised
setting on MNIST, improved lung segmentation results, failure prediction capability on chest scans of COVID19 patients and improved anomaly detection scores on MIMII dataset. |
(英) |
The remarkable effectiveness of neural networks on vision tasks has led to an interest in adapting neural network models to limited data cases. It is also desired that low dimensional representations of the data efficiently represent the data distribution. We propose to minimize ordinal energy of the code produced by encoder model of de-noising auto-encoder and add stochastic non-linear units. Proposed modifications lead to an increase in the classification performance in the semi-supervised
setting on MNIST, improved lung segmentation results, failure prediction capability on chest scans of COVID19 patients and improved anomaly detection scores on MIMII dataset. |
キーワード |
(和) |
Representation learning / Anomaly detection / Small Data / Stochastic nonlinearity / De-noising auto-encoder / Segmentation / Ordinal Energy / |
(英) |
Representation learning / Anomaly detection / Small Data / Stochastic nonlinearity / De-noising auto-encoder / Segmentation / Ordinal Energy / |
文献情報 |
信学技報, vol. 120, no. 300, PRMU2020-49, pp. 63-68, 2020年12月. |
資料番号 |
PRMU2020-49 |
発行日 |
2020-12-10 (PRMU) |
ISSN |
Online edition: ISSN 2432-6380 |
著作権に ついて |
技術研究報告に掲載された論文の著作権は電子情報通信学会に帰属します.(許諾番号:10GA0019/12GB0052/13GB0056/17GB0034/18GB0034) |
PDFダウンロード |
PRMU2020-49 |