Presentation | 2019-12-06 Interpretation of Multi-Label Learning by combining two probability models Kurebayashi Kosuke, Morizumi Tetsuya, Kinoshita Hirotsugu, |
---|---|
PDF Download Page | PDF download Page Link |
Abstract(in Japanese) | (See Japanese page) |
Abstract(in English) | We have already proposed a security model with a three-layer structure of AI architecture. However, there is a certain lack of certainty whether a security policy or the like determined by humans can be reflected in access control even if it is included in teacher data. In this paper, the weighting in Multi-label Learning is separated from the probabilistic model for the purpose of surely reflecting the security policy and security rules newly established by humans, and the tensor components (words) composed of labels and topics are separated. On the other hand, we propose a model weighted manually. The manual [label, topic] tensor weighting is obtained by extracting a label latent random variable and a topic latent random variable tensor defined by multi-label learning from a multi-label learning probability model. In this proposal, the [Label, Topic] tensor is extracted from the Multi Label probabilistic model, so that the entire probabilistic model is separated into two probabilistic models (LDA). These correspond to the first and second layers of the three-layer AI architecture, respectively. By performing this weighting, it is possible to reliably reflect the security policy to what is intended by humans. In this paper, we confirmed by experiment how much utility the weighting can be obtained. |
Keyword(in Japanese) | (See Japanese page) |
Keyword(in English) | Artificial intelligence / Machine learning / Access control / Probabilistic security model / Probabilistic model / LDA / Tensor Decomposition |
Paper # | SITE2019-81 |
Date of Issue | 2019-11-29 (SITE) |
Conference Information | |
Committee | SITE |
---|---|
Conference Date | 2019/12/6(1days) |
Place (in Japanese) | (See Japanese page) |
Place (in English) | |
Topics (in Japanese) | (See Japanese page) |
Topics (in English) | |
Chair | Tetsuya Morizumi(Kanagawa Univ.) |
Vice Chair | Masaru Ogawa(Kobe Gakuin Univ.) / Takushi Otani(Kibi International Univ.) |
Secretary | Masaru Ogawa(Toyo Eiwa Univ.) / Takushi Otani(KDDI Research) |
Assistant | Nobuyuki Yoshinaga(Yamaguchi Pref Univ.) / Daisuke Suzuki(Hokuriku Univ.) |
Paper Information | |
Registration To | Technical Committee on Social Implications of Technology and Information Ethics |
---|---|
Language | JPN |
Title (in Japanese) | (See Japanese page) |
Sub Title (in Japanese) | (See Japanese page) |
Title (in English) | Interpretation of Multi-Label Learning by combining two probability models |
Sub Title (in English) | An approach that interprets evaluate texts by regarding labels as teacher data |
Keyword(1) | Artificial intelligence |
Keyword(2) | Machine learning |
Keyword(3) | Access control |
Keyword(4) | Probabilistic security model |
Keyword(5) | Probabilistic model |
Keyword(6) | LDA |
Keyword(7) | Tensor Decomposition |
1st Author's Name | Kurebayashi Kosuke |
1st Author's Affiliation | Kanagawa University(Kanagawa Univ.) |
2nd Author's Name | Morizumi Tetsuya |
2nd Author's Affiliation | Kanagawa University(Kanagawa Univ.) |
3rd Author's Name | Kinoshita Hirotsugu |
3rd Author's Affiliation | Kanagawa University(Kanagawa Univ.) |
Date | 2019-12-06 |
Paper # | SITE2019-81 |
Volume (vol) | vol.119 |
Number (no) | SITE-329 |
Page | pp.pp.7-12(SITE), |
#Pages | 6 |
Date of Issue | 2019-11-29 (SITE) |