論文一覧

2024

論文誌

Muhammad Ayat Hidayat, Yugo Nakamura, Yutaka Arakawa

Enhancing Efficiency in Privacy-preserving Federated Learning for Healthcare : Adaptive Gaussian Clipping with DFT Aggregator Journal Article

In: IEEE Access, pp. 1-1, 2024.

Links | BibTeX

Yuki Nishi, Yugo Nakamura, Shogo Fukushima, Yutaka Arakawa

LightSub: Unobtrusive Subtitles with Reduced Information and Decreased Eye Movement Journal Article

In: Multimodal Technologies and Interaction, vol. 8, no. 6, 2024, ISSN: 2414-4088.

Abstract | Links | BibTeX

@article{mti8060051,

title = {LightSub: Unobtrusive Subtitles with Reduced Information and Decreased Eye Movement},

author = {Yuki Nishi and Yugo Nakamura and Shogo Fukushima and Yutaka Arakawa},

url = {https://www.mdpi.com/2414-4088/8/6/51},

doi = {10.3390/mti8060051},

issn = {2414-4088},

year  = {2024},

date = {2024-06-01},

journal = {Multimodal Technologies and Interaction},

volume = {8},

number = {6},

abstract = {Subtitles play a crucial role in facilitating the understanding of visual content when watching films and television programs. In this study, we propose a method for presenting subtitles in a way that considers cognitive load when viewing video content in a non-native language. Subtitles are generally displayed at the bottom of the screen, which causes frequent eye focus switching between subtitles and video, increasing the cognitive load. In our proposed method, we focused on the position, display time, and amount of information contained in the subtitles to reduce the cognitive load and to avoid disturbing the viewer’s concentration. We conducted two experiments to investigate the effects of our proposed subtitle method on gaze distribution, comprehension, and cognitive load during English-language video viewing. Twelve non-native English-speaking subjects participated in the first experiment. The results show that participants’ gazes were more focused around the center of the screen when using our proposed subtitles compared to regular subtitles. Comprehension levels recorded using LightSub were similar, but slightly inferior to those recorded using regular subtitles. However, it was confirmed that most of the participants were viewing the video with a higher cognitive load using the proposed subtitle method. In the second experiment, we investigated subtitles considering connected speech form in English with 18 non-native English speakers. The results revealed that the proposed method, considering connected speech form, demonstrated an improvement in cognitive load during video viewing but it remained higher than that of regular subtitles.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

Yuichi Hattori, Yutaka Arakawa, Sozo Inoue

Evaluation of Functional Estimation Methods in IoT Devices at Intervals of a Few Seconds by Communication Traffic Analysis Journal Article

In: International Journal of Activity and Behavior Computing, vol. 2024, no. 1, pp. 5, 2024.

Links | BibTeX

Sopicha Stirapongsasuti, Francis Jerome Tiausas, Yugo Nakamura, Keiichi Yasumoto

Preserving Data Utility in Differentially Private Smart Home Data Journal Article

In: IEEE Access, vol. 12, pp. 56571-56581, 2024, ISBN: 2169-3536.

Links | BibTeX

Shaowen Peng, Kazunari Sugiyama, Tsunenori Mine

Less is More: Removing Redundancy of Graph Convolutional Networks for Recommendation Journal Article

In: Transactions on Information Systems, vol. 42, iss. 3, no. 85, pp. 1-26, 2024.

Abstract | Links | BibTeX

@article{transactions2024-peng,

title = {Less is More: Removing Redundancy of Graph Convolutional Networks for Recommendation},

author = {Shaowen Peng, Kazunari Sugiyama, Tsunenori Mine},

doi = {10.1145/3632751},

year  = {2024},

date = {2024-01-22},

urldate = {2024-01-22},

journal = {Transactions on Information Systems},

volume = {42},

number = {85},

issue = {3},

pages = {1-26},

abstract = {While Graph Convolutional Networks (GCNs) have shown great potential in recommender systems and collaborative filtering (CF), they suffer from expensive computational complexity and poor scalability. On top of that, recent works mostly combine GCNs with other advanced algorithms which further sacrifice model efficiency and scalability. In this work, we unveil the redundancy of existing GCN-based methods in three aspects: (1) Feature redundancy. By reviewing GCNs from a spectral perspective, we show that most spectral graph features are noisy for recommendation, while stacking graph convolution layers can suppress but cannot completely remove the noisy features, which we mostly summarize from our previous work; (2) Structure redundancy. By providing a deep insight into how user/item representations are generated, we show that what makes them distinctive lies in the spectral graph features, while the core idea of GCNs (i.e., neighborhood aggregation) is not the reason making GCNs effective; and (3) Distribution redundancy. Following observations from (1), we further show that the number of required spectral features is closely related to the spectral distribution, where important information tends to be concentrated in more (fewer) spectral features on a flatter (sharper) distribution. To make important information be concentrated in as few features as possible, we sharpen the spectral distribution by increasing the node similarity without changing the original data, thereby reducing the computational cost. To remove these three kinds of redundancies, we propose a Simplified Graph Denoising Encoder (SGDE) only exploiting the top-K singular vectors without explicitly aggregating neighborhood, which significantly reduces the complexity of GCN-based methods. We further propose a scalable contrastive learning framework to alleviate data sparsity and to boost model robustness and generalization, leading to significant improvement. Extensive experiments on three real-world datasets show that our proposed SGDE not only achieves state-of-the-art but also shows higher scalability and efficiency than our previously proposed GDE as well as traditional and GCN-based CF methods.},

keywords = {},

pubstate = {published},

tppubtype = {article}

}

While Graph Convolutional Networks (GCNs) have shown great potential in recommender systems and collaborative filtering (CF), they suffer from expensive computational complexity and poor scalability. On top of that, recent works mostly combine GCNs with other advanced algorithms which further sacrifice model efficiency and scalability. In this work, we unveil the redundancy of existing GCN-based methods in three aspects: (1) Feature redundancy. By reviewing GCNs from a spectral perspective, we show that most spectral graph features are noisy for recommendation, while stacking graph convolution layers can suppress but cannot completely remove the noisy features, which we mostly summarize from our previous work; (2) Structure redundancy. By providing a deep insight into how user/item representations are generated, we show that what makes them distinctive lies in the spectral graph features, while the core idea of GCNs (i.e., neighborhood aggregation) is not the reason making GCNs effective; and (3) Distribution redundancy. Following observations from (1), we further show that the number of required spectral features is closely related to the spectral distribution, where important information tends to be concentrated in more (fewer) spectral features on a flatter (sharper) distribution. To make important information be concentrated in as few features as possible, we sharpen the spectral distribution by increasing the node similarity without changing the original data, thereby reducing the computational cost. To remove these three kinds of redundancies, we propose a Simplified Graph Denoising Encoder (SGDE) only exploiting the top-K singular vectors without explicitly aggregating neighborhood, which significantly reduces the complexity of GCN-based methods. We further propose a scalable contrastive learning framework to alleviate data sparsity and to boost model robustness and generalization, leading to significant improvement. Extensive experiments on three real-world datasets show that our proposed SGDE not only achieves state-of-the-art but also shows higher scalability and efficiency than our previously proposed GDE as well as traditional and GCN-based CF methods.

国際会議

16.

Akihisa Takiguchi, Tsunenori Mine, Yutaka Arakawa

Investigation of the Diversity of News Reading Articles and Browsing Trends Using Sentence-BERT Proceedings Article

In: 16th International Congress on Advanced Applied Informatics (IIAI AAI 2024), 2024.

2024

論文誌

国際会議

研究会・シンポジウム