Presentation | 2019-01-24 An Error Correction Technique with the Viterbi Algorithm for a Machine-Learning-Based Classifier Yoshito Watanabe, Yozo Shoji, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | This paper proposes a novel technique to correct errors that can be caused in the decisions made by a machine learning (ML)-based classifier. Considering the log-likelihood ratio (LLR) of the posterior probability emitted from a classifier for each class as the observed variable, a hidden Markov model (HMM) is introduced to numerically model the sequential operation of classification process, and the Viterbi algorithm is then applied to the sequence of the LLR to make the final decisions. A case of binary classification, in which mobile-sensing-based anomaly detection employing a support vector machine (SVM), is first investigated. The method is extended to an example case of multiclass classification, in which an application of human activity recognition from the acceleration data obtained by body-worn smartphones by a neural network is demonstrated. The simulation results show that the proposed technique is effective for correcting the misclassifications made by conventional ML-based classifiers. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | machine learning / sequential classification / the Viterbi algorithm / hidden Markov model / the EM algorithm |
Paper # | SR2018-105 |
Date of Issue | 2019-01-17 (SR) |
Conference Information | |
Committee | SR |
---|---|
Conference Date | 2019/1/24(2days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | Corasse, Fukushima city (Fukushima prefecture) |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | cognitive radio, machine learning application, heterogeneous network, SDN, IoT etc. |
Chair | Kenta Umebayashi(Tokyo Univ. of Agric. and Tech.) |
Vice Chair | Masayuki Ariyoshi(NEC) / Suguru Kameda(Tohoku Univ.) |
Secretary | Masayuki Ariyoshi(NICT) / Suguru Kameda(ATR) |
Assistant | Gia Khanh Tran(Tokyo Inst. of Tech.) / Syusuke Narieda(Mie Univ.) / Koji Ohshima(Kozo Keikaku Engineering) / Mai Ohta(Fukuoka Univ.) / Teppei Oyama(Fujitsu Lab.) |
Paper Information | |
Registration To | Technical Committee on Smart Radio |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | An Error Correction Technique with the Viterbi Algorithm for a Machine-Learning-Based Classifier |
Sub Title (in English) | |
Keyword(1) | machine learning |
Keyword(2) | sequential classification |
Keyword(3) | the Viterbi algorithm |
Keyword(4) | hidden Markov model |
Keyword(5) | the EM algorithm |
1st Author's Name | Yoshito Watanabe |
1st Author's Affiliation | National Institute of Information and Communications Technology(NICT) |
2nd Author's Name | Yozo Shoji |
2nd Author's Affiliation | National Institute of Information and Communications Technology(NICT) |
Date | 2019-01-24 |
Paper # | SR2018-105 |
Volume (vol) | vol.118 |
Number (no) | SR-421 |
Page | pp.pp.57-61(SR), |
#Pages | 5 |
Date of Issue | 2019-01-17 (SR) |