The rapid increase of graph data calls for efficient analysis on massive graphs. Pregel is a popular distributed computing model for dealing with large-scale graphs. However, it can be tricky to implement graph algorithms correctly and efficiently in its vertex-centric model, especially when the algorithm has multiple computation stages, complicated data dependencies, or even communication over dynamic internal data structures. We propose a new domain-specific language Palgol to help users handle all of these complexities in Pregel programming, and it also ensures high efficiency even compared with well-optimized hand written code.
B4：Stabilizing GMRES Using the Normal Equation Approach for Severely Ill-Conditioned Problems Liao Zeyu
Consider using GMRES to solve severely ill-conditioned problems, where the condition number of the Hessenburg matrix and the resulting upper triangular system becomes huge, so that the back substitution process becomes unstable and the convergence of GMRES deteriorates. We propose solving the normal equation corresponding to the above triangular system using standard Cholesky decomposition. This has the effect of 'hiding' tiny singular values, making the process stable, rendering better convergence and a more accurate solution.。
公共機関や研究機関の情報を再利用可能な形式で公開するオープンデータ・オープンサイエンスの取り組みが広まっていますが、高度な利活用を促進するためには情報の構造化や語彙の統一などのセマンティック技術の導入が必要不可欠です。本ブースではLinked Open Data（LOD）を用いた専門分野の用語辞書構築や、表形式で記述された情報の自動同定に関する研究を紹介します。
C4: Deletion-based sentence compression A Simple Language Model based Evaluator for Sentence Compression Yang ZHAO
Deletion-based sentence compression aims to delete unnecessary words from the original sentence to form a short sentence while retaining grammaticality and important information. In this work, we present a language-model-based evaluator for deletion-based sentence. More specifically, a syntactic neural language model is first built and then, a series of trial-and-error deletion operations are conducted on source sentences via a reinforcement learning framework to find the best target compression. Empirical study shows that the proposed model can effectively generate more readable compressed sentences, comparable or superior to several strong baseline methods.
C10: Security of machine learning processes Characterizing Adversarial Subspaces Using Local Intrinsic Dimensionality Michael E. Houle
Deep Neural Networks (DNNs) have recently been shown to be vulnerable against adversarial examples, which are carefully crafted instances that can mislead DNNs to make errors during prediction. To better understand such attacks, a characterization is needed of the properties of regions (the so-called `adversarial subspaces') in which adversarial examples lie. We tackle this challenge by characterizing the dimensional properties of adversarial regions, via the use of Local Intrinsic Dimensionality. Our analysis not only motivates new directions of effective adversarial defense, but also opens up more challenges for developing new attacks to better understand the vulnerabilities of DNNs.
C11: Characterizing the complexity of data analysis tasks Inlierness, Outlierness, Hubness and Discriminability: an Extreme-Value-Theoretic Foundation Michael E. Houle
To date, no unifying theory of data mining has been proposed. I Many ad-hoc techniques have been designed for individual problems, such as classification or clustering. I Solutions involve much invention and reinvention, with few guidelines. A theoretical framework that ties together different fundamental machine learning and data mining tasks (including indexing, clustering, classification, data discriminability, subspace methods, etc.) could help the discipline, and serve as a basis for future investigation.
E5: Transformation on Computer-Generated Facial Image to Avoid Detection by Spoofing Detector Nguyen Hong Huy
Making computer-generated (CG) images more difficult to detect is an interesting problem in computer graphics and security. While most approaches focus on the image rendering phase, this paper presents a method based on increasing the naturalness of CG facial images from the perspective of spoofing detectors. The proposed method is implemented using a convolutional neural network (CNN) comprising two autoencoders and a transformer and is trained using a black-box discriminator without gradient information. Over 50% of the transformed CG images were not detected by three state-of-the-art spoofing detectors. This capability raises an alarm regarding the reliability of facial authentication systems, which are becoming widely used in daily life.
E6: Cyber security Gait Anonymization Using Deep Learning Tieu Thi Ngoc Dung
Nowadays, internet users can easily upload photos and videos to the social networks with or without permission of people captured in the video. A serious problem may occur if someone captured in those videos is identified unintentionally by gait recognition systems and his/her personal information is revealed eventually. The objective of this research is to address this privacy problem, which is to helps users upload their videos safely and effectively. The proposed method modifies the original gait so that the gait recognition systems cannot identify the people in the videos, while maintaining the naturalness of the gaits.
The Cooking Recipes Without Border project will demonstrate an innovative social media platform for sharing photos and ratings of dishes. We will show a new approach that uses Linked Open Data to build a Cooking Recipe Knowledge Base. This Cooking Recipe Knowledge Base is a collection of generalized cooking recipes semantic graphs. The cooking recipes will be generated modifying theses generalized cooking semantic graphs.
F10: Inducing Temporal Relations from Time Anchor Annotation A new approach to obtain temporal relation annotation, which requires less annotation effort, induces inter-sentence relations easily, and increases informativeness of temporal relations. 金融スマートデータ研究センター
Recognizing temporal relations among events and time expressions has been an essential but challenging task in natural language processing. Conventional annotation of judging temporal relations puts a heavy load on annotators. In reality, the existing annotated corpora include annotations on only "salient" event pairs, or on pairs in a fixed window of sentences. In this paper, we propose a new approach to obtain temporal relations from absolute time value (a.k.a. time anchors), which is suitable for texts containing rich temporal information such as news articles. We start from time anchors for events and time expressions, and temporal relation annotations are induced automatically by computing relative order of two time anchors. This proposal shows several advantages over the current methods for temporal relation annotation.