International Journal of Service and Knowledge Management

Adversarial Attacks for Time Series Classification using Partial Perturbations

Jun Teraoka, Keiichi Tamura — Mon, 28 Oct 2024 01:43:49 +0000

Adversarial attacks using adversarial examples have recently become a significant threat that intentionally misleads deep-learning models beyond human recognition. Adversarial examples have primarily been studied in the field of image recognition; however, they have recently been applied in other fields, including time series data classification. To generate adversarial examples, small perturbations unrecognizable by humans are typically added to all the data regions. However, adding perturbations to the entire time series data results in time series data that are clearly manipulated for time series classification. In this case, adversarial attacks are immediately apparent to humans and do not pose a significant threat. This study shows that unidentifiable adversarial examples of time series can be identified as adversarial examples in time series data classification by adopting partial perturbations. The fast gradient sign method (FGSM) and projected gradient descent (PGD) attack methods, which are originally proposed for generating adversarial examples of image data, are applied to time series data classification models. In this study, partial-FGSM and partial-PGD attacks are proposed which utilize only a part of the perturbations to generate fewer unreliable adversarial examples of time series data that are easily recognized as adversarial examples. To evaluate partial-FGSM and partial-PGD attacks, the 2 Class-Based-Detecting adversarial detection method is employed, as its effectiveness for protecting adversarial attacks against time series classification has been proven. The performance is evaluated, and the results show that attacks are possible with a small degradation in attack performance for some datasets, even if the perturbation ratio is 1/10.

The Influence of Linguistic Attribute Differences in Multilingual Datasets on Sarcasm Detection

Linshuo Yang, Daisuke Ikeda — Mon, 16 Dec 2024 13:06:15 +0000

Presently, social media is crucial for sentiment analysis tasks with machine learning. However, the presence of sarcasm presents a challenge to this task by concealing the true intent of a text. Consequently, there has been a surge in research on automatic sarcasm detection. Furthermore, studying with sarcasm detection with multilingual datasets is becoming indispensable, because it can sovle data scarcity problem for low-resource languages and can also reduce the cost of training models for different languages. Past research has largely overlooked the influence of language diversity within training datasets on model performance. This study assumes that linguistic differences may influence sarcastic expressions and employs two datasets: English- Arabic dataset, which belongs to the same category of morphological typology, and English-Chinese dataset which belongs to different categories. Subsequently, models were trained with BERT, BERT-BiLSTM, and BERT-RCNN architectures. Finally, results were compared using two English test datasets with different patterns. The outcomes revealed superior training results for English-Arabic in contrast to English-Chinese, signifying the influence of morphological typology. In addition, BiLSTM and RCNN architectures can enhance the performance of multilingual sarcasm detection models. And the RCNN structure appears to be beneficial for detecting sarcasm in different patterns.

Migrant Entrepreneurs and Regional Revitalization: A Qualitative Case Study

Osamu Nakajima — Mon, 09 Dec 2024 06:40:02 +0000

In Japan, the number of people involved in community activities is decreasing due to population outflow; thus, the survival of some communities is in jeopardy. Although various local governments, residents, and businesses are attempting to ensure communities’ survival, some limitations remain. However, the changes in lifestyles and values resulting from the COVID-19 pandemic have reduced the psychological hurdles for people to migrate to rural areas. Under these circumstances, migrants who arrive in rural regions and start businesses have an economic impact on their destination, increasing the value of the region. To identify these agents, in this paper, I propose the concept of “migrant entrepreneurs,” which has diverse meanings depending on these individuals’ relationship with the local community and the form of their entrepreneurship. Taking the town of Higashikawa, in the prefecture of Hokkaido, as a study case, I clarify the characteristics of migrant entrepreneurs, who comprise a new entity responsible for regional revitalization.

The Redesigning Road Network in the Era of Decreasing Population by Traffic Simulation

Mon, 28 Oct 2024 01:41:34 +0000

In light of an aging society in which the number of expressway users declines while infrastructure maintenance costs remain high, we propose a new method for evaluating whether expressway routes should be closed when their cost-benefit ratios become unfavorable owing to reduced usage. Moreover, we conducted a case study in which this method was applied to analyze the Kawasaki route (Kanagawa No. 6, Kawasaki route, a metropolitan expressway in Japan). The analysis results projected the cost-benefit ratio of this route to fall below 1 by 2025 owing to the reduced number of users. However, closing the Kawasaki route would lead to significant congestion on its alternative route, National Route 409. Restoring the congestion rate of National Route 409 to its previous level requires its expansion to 10 lanes, making the closure of the Kawasaki route impractical.

Intellectual Property Strategy Required for Startups from Joining in a New Business Ecosystem to Growing as Large Companies: A Case Study of Tesla

Jun Oya, Naoshi Uchihira — Mon, 28 Oct 2024 01:38:25 +0000

This article conducts a case study of Tesla’s intellectual property (IP) strategy to clarify the IP strategy required for startups from joining in a new business ecosystem to growing as large companies. As a result, it was clarified that to establish new businesses quickly, it may be effective for startups with limited own resources to focus on patents for not only core areas developed on their own but also the interface between core and outsourcing areas to collaborate with companies that have a substantial patent portfolio in outsourcing areas and open the interfaces to diffuse their core technologies. In the process of growing into large companies, startups are exposed to stiff competition as competitors try to catch up. Therefore, it is necessary to increase their capability, formulate added value through in-house research and development and design in the outsourcing areas, and apply for patents to increase its competitive advantage. It was further indicated that a pledge of non-exercise of patent rights may be an effective option to avoid being sued for patent infringement. This study is unique in identifying effective IP strategies for startups in terms of outsourcing, diffusion of their technologies, increasing their capability and avoiding patent infringement lawsuits.

Generation and Comparison of Distributed Representations of Words from Japanese Wiktionary with BERT Sentence-Embedding

Ryota Nishiura, Seiji Tsuchiya, Hirokazu Watabe — Fri, 20 Dec 2024 04:19:02 +0000

In this paper, we introduce a novel approach to generate distributed representations of words. Our approach makes use of structured contents of Wiktionary, in contrast to existing methods such as Word2Vec, which are learned from unstructured corpora of plain text. Our approach generates distributed representations of headwords in Wiktionary by obtaining sentence embeddings of the corresponding contents of the entries with a pre-trained BERT model. We demonstrate that our proposed method outperforms a Word2Vec model in a XABC test, potentially due to its ability to leverage both the quality of a pre-trained BERT model and expert-moderated knowledge from Wik-tionary. We also conducted a comparative study on a variety of different BERT models to find the model conditions most suitable for our purpose.