| Presentation | 2017-06-24 Reinforcement learning for visual attention with scalable size of attentional field Yutaro Murata, Jun-nosuke Teramae, Naoki Wakamiya, |
|---|---|
| PDF Download Page |  PDF download Page Link |
| Abstract(in Japanese) | (See Japanese page) |
| Abstract(in English) | Humans percept visual scenes with focusing only on significant parts of visual scenes sequentially ratherprocess whole of the visual stimuli entirely at once. Sequential visual attention attracts much attention recently in thefield of computer vision since it may largely reduce computational cost of image processing. A previous study proposedthe hard attention model, in which neural networks are trained to select suitable attended areas of visual fields forvisual recognition by using reinforcement learning. The model, however, cannot properly select attended area if sizesof significant areas in visual scenes are different from that of training data since it cannot modulate size of attendedareas. To solve the problem, here, we propose a network model that can learn both position and size of attended areaof visual field. Since phase space features are significantly different between position and size of a selected visual scene, they contribute to reinforcement signals in different manner, which causes undesirable bias and failure of learning. Toreduce the bias, we propose a new learning algorithm that almost keep balance of their contribution. We show thatour model can learn to suitably modulate the size of attended area and visual recognition with sequential attention. |
| Keyword(in Japanese) | (See Japanese page) |
| Keyword(in English) | reinforcement learning / visual attention / image recognition / recurrent neural network |
| Paper # | NC2017-18 |
| Date of Issue | 2017-06-16 (NC) |
| Conference Information | |
| Committee | NC / IPSJ-BIO / IBISML / IPSJ-MPS |
|---|---|
| Conference Date | 2017/6/23(3days) |
| Place (in Japanese) | (See Japanese page) |
| Place (in English) | Okinawa Institute of Science and Technology |
| Topics (in Japanese) | (See Japanese page) |
| Topics (in English) | Machine Learning Approach to Biodata Mining, and General |
| Chair | Masafumi Hagiwara(Keio Univ.) / / Kenji Fukumizu(ISM) |
| Vice Chair | Yutaka Hirata(Chubu Univ.) / / Masashi Sugiyama(Univ. of Tokyo) |
| Secretary | Yutaka Hirata(Tokyo Inst. of Tech.) / (Nagoya Univ.) / Masashi Sugiyama / (Kyoto Univ.) |
| Assistant | Yoshihisa Shinozawa(Keio Univ.) / Keiichiro Inagaki(Chubu Univ.) / / Ichiro Takeuchi(Nagoya Inst. of Tech.) / Toshihiro Kamishima(AIST) |
| Paper Information | |
| Registration To | Technical Committee on Neurocomputing / Special Interest Group on Bioinformatics and Genomics / Technical Committee on Infomation-Based Induction Sciences and Machine Learning / Special Interest Group on Mathematical Modeling and Problem Solving |
|---|---|
| Language | JPN |
| Title (in Japanese) | (See Japanese page) |
| Sub Title (in Japanese) | (See Japanese page) |
| Title (in English) | Reinforcement learning for visual attention with scalable size of attentional field |
| Sub Title (in English) | |
| Keyword(1) | reinforcement learning |
| Keyword(2) | visual attention |
| Keyword(3) | image recognition |
| Keyword(4) | recurrent neural network |
| 1st Author's Name | Yutaro Murata |
| 1st Author's Affiliation | Osaka University(Osaka Univ.) |
| 2nd Author's Name | Jun-nosuke Teramae |
| 2nd Author's Affiliation | Osaka University(Osaka Univ.) |
| 3rd Author's Name | Naoki Wakamiya |
| 3rd Author's Affiliation | Osaka University(Osaka Univ.) |
| Date | 2017-06-24 |
| Paper # | NC2017-18 |
| Volume (vol) | vol.117 |
| Number (no) | NC-109 |
| Page | pp.pp.121-125(NC), |
| #Pages | 5 |
| Date of Issue | 2017-06-16 (NC) |