83 research outputs found
Dynamic Changes in Eastern and Northern Chinese Dialects During the Han Dynasty
No full text本文的主要内容曾在“ICSTLL-55”（题目為“談揚雄《方言》中東齊海岱方言詞彙的特徴”，2022年9月16日，於京都大學）以及“Workshop: Chinese Language and its surroundings”（題目為“淺談漢代北方方言的形成過程”，2022年12月17日，於京都大學。京大中國語學研究會主辧) 上宣讀。This paper focuses on the Former Han Dynasty's Eastern and Northern Chinese dialects, using Yang Xiong's “Fangyan” as linguistic material. It also discusses the distance between these and other dialects. Additionally, it estimates the historical changes that occurred in these Eastern and Northern dialects during or before the Former Han Dynasty. Specifically, it employs the method of “frequency of shared or distinct vocabularies among dialects” and highlights two main points: (1) The “Dongqi-Haidai” dialect in the East is closer linguistically to the geographically distant “Qin- Jin” dialect than to neighboring ones like “Lu, ” “Song, ” and “Zhao.” (2) The northern “Beiyan” and the “Chaoxian” dialects are extremely close linguistically. However, only the “Chaoxian” dialect exhibits an extremely distant linguistic relationship with “Dongqi-Haidai.” Based on these points, phenomenon (1) likely resulted from dialects like “Lu, ” “Song, ” and “Zhao” being influenced by groups situated to their south. Meanwhile, phenomenon (2) is inferred to originate from the “Beiyan” dialect influenced by southern groups. In contrast, the “Chaoxian” dialect seems to have been minimally influenced. Additionally, the paper makes two assertions regarding the Chinese dialects of the Western Han period: (3) Vocabulary distributed in the “Beiyan” and “Chaoxian” dialects displays distinctive formal characteristics, with a higher proportion of non-monosyllabic words compared to other dialects. This tendency is particularly evident in “Yan zhi jiaowai, ” a part of the “Beiyan” dialect. (4) Wang Qiming (1993) argued that an overlap exists in the distribution areas of Yang Xiong's “Dongqi” and “Qi” dialects. However, after conducting an analysis using the “frequency of shared or distinct vocabularies among dialects” method, this paper disagrees and considers both to refer to different regional dialects
Improving exploration in reinforcement learning with temporally correlated stochasticity
No full text
抽象化及び推論し行動計画を立案することが可能な認知脳型ロボットの構築に向けて：階層的で部分観測な環境での強化学習及び能動推論
No full textOkinawa Institute of Science and Technology Graduate UniversityDoctor of PhilosophyThe thesis aims to advance cognitive decision-making and motor control using reinforcement learning (RL) with stochastic recurrent neural networks (RNNs). RL is a computational framework to train an agent, such as a robot, to select the actions that maximize immediate or future rewards. Recently, RL has undergone rapid development by introducing artificial neural networks as function approximators. RL using neural networks, also known as deep RL, have shown super-human performance on a wide range of virtual and real-world tasks, such as games, robotic control, and manipulating nuclear fusion devices. There would not be such a success without the efforts of numerous researchers who developed and improved the deep RL algorithms. In particular, most of the works focus on designing or revising the RL objective functions by mathematical analysis and heuristic ideas. While the well-formulated loss functions are critical to the RL performance, relatively fewer efforts have been paid to developing and improving the architecture of the neural network models used in deep RL. The thesis discusses the benefits of using novel network architectures for deep RL. In particular, the thesis includes two of the authors’ original studies about developing novel stochastic RNN architectures for RL in partially observable environments. The first work proposes a novel, multiple-level, stochastic RNN model for solving tasks that require hierarchical control. It is shown that an action hierarchy, characterized by consistent representation for abstracted sub-goals in the higher level, self-develops during the learning in several challenging continuous robotic control tasks. The emerged action hierarchy is also observed to enable faster relearning when the sub-goals are recomposed. The second work introduces a variational RNN model for predicting state transitions in continuous robotic control tasks in which the environmental state is partially observable. By predicting subsequent observations, the models learn to represent the underlying states of the environment that are indispensable but not observable. A corresponding algorithm is proposed to facilitate efficient learning in partially observable environments. The proposed studies suggest that the performance of RL agents can be improved by adequate usage of stochastic RNNs structures, which provides novel insights for designing better model architectures for future deep RL studies
GEAD_CCS24
No full text<p>This repository contains the source code, processed datasets, and evaluation results and scripts for the paper "Rules Refine the Riddle: Global Explanation for Deep Learning-Based Anomaly Detection in Security Applications" accepted by ACM CCS 2024.</p>
Data for the paper "Synergizing habits and goals with variational Bayes".
No full text<p>Data used for the paper "Synergizing habits and goals with variational Bayes". </p>
Improving exploration in reinforcement learning with temporally correlated stochasticity
No full textReinforcement learning is a useful ap-proach to solve machine learning problems by self-exploration when training samples are not provided.However, researchers usually ignore the importance ofthe choice of exploration noise. In this paper, I showthat temporally self-correlated exploration stochastic-ity, generated by Ornstein-Uhlenbeck process, can sig-nificantly enhance the performance of reinforcementlearning tasks by improving exploration.conference pape
淺談漢代東部、北部方言的動態變化
No full textThis paper focuses on the Former Han Dynasty's Eastern and Northern Chinese dialects, using Yang Xiong's “Fangyan” as linguistic material. It also discusses the distance between these and other dialects. Additionally, it estimates the historical changes that occurred in these Eastern and Northern dialects during or before the Former Han Dynasty. Specifically, it employs the method of “frequency of shared or distinct vocabularies among dialects” and highlights two main points: (1) The “Dongqi-Haidai” dialect in the East is closer linguistically to the geographically distant “Qin- Jin” dialect than to neighboring ones like “Lu, ” “Song, ” and “Zhao.” (2) The northern “Beiyan” and the “Chaoxian” dialects are extremely close linguistically. However, only the “Chaoxian” dialect exhibits an extremely distant linguistic relationship with “Dongqi-Haidai.” Based on these points, phenomenon (1) likely resulted from dialects like “Lu, ” “Song, ” and “Zhao” being influenced by groups situated to their south. Meanwhile, phenomenon (2) is inferred to originate from the “Beiyan” dialect influenced by southern groups. In contrast, the “Chaoxian” dialect seems to have been minimally influenced. Additionally, the paper makes two assertions regarding the Chinese dialects of the Western Han period: (3) Vocabulary distributed in the “Beiyan” and “Chaoxian” dialects displays distinctive formal characteristics, with a higher proportion of non-monosyllabic words compared to other dialects. This tendency is particularly evident in “Yan zhi jiaowai, ” a part of the “Beiyan” dialect. (4) Wang Qiming (1993) argued that an overlap exists in the distribution areas of Yang Xiong's “Dongqi” and “Qi” dialects. However, after conducting an analysis using the “frequency of shared or distinct vocabularies among dialects” method, this paper disagrees and considers both to refer to different regional dialects.本文的主要内容曾在“ICSTLL-55”（题目為“談揚雄《方言》中東齊海岱方言詞彙的特徴”，2022年9月16日，於京都大學）以及“Workshop: Chinese Language and its surroundings”（題目為“淺談漢代北方方言的形成過程”，2022年12月17日，於京都大學。京大中國語學研究會主辧) 上宣讀
AdHoneyDroid-Capture malicious android advertisements
No full textIn this paper we explore the problem of collecting malicious smartphone advertisements. Most smartphone app contains advertisements and also suffers from vulnerable advertisement libraries. Malicious advertisements exploit the ad library vulnerability and attack victim smartphones. Similar to the traditional honeypots, we need an effective way to capture malicious ads. In this paper, we provide our approach named AdHoneyDroid. We build a crawler to gather apps on the android marketplaces and manually collect ad libraries and their vulnerabilities. Then AdHoneyDroid executes the apps and detects malicious advertisements. In our approach, we adopt the idea of API sandbox and TaintDroid to detect the attack event. We store the malicious advertise-ments in a database for future analysis. Malicious ads can help security analysts have a better understanding of current mobile attacks and also disclose the attack payloads. Copyright is held by the owner/author(s).EI
- …
