Data Mining - Concepts and Techniques.

Statistics for Spatial Data.

“Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks”の学習報告

From the Publisher: 
The accessible presentation of this book gives both a general view of the entire computer vision enterprise and also offers sufficient detail to be able to build useful applications. Users learn techniques that have proven to be useful by first-hand experience and a wide range of mathematical methods. A CD-ROM with every copy of the text contains source code for programming practice, color images, and illustrative movies. Comprehensive and up-to-date, this book includes essential topics that either reflect practical significance or are of theoretical importance. Topics are discussed in substantial and increasing depth. Application surveys describe numerous important application areas such as image based rendering and digital libraries. Many important algorithms broken down and illustrated in pseudo code. Appropriate for use by engineers as a comprehensive reference to the computer vision enterprise.

Computer vision : a modern approach = 计算机视觉 : 一种现代的方法

Artificial neural networks (ANNs) constitute a class of flexible nonlinear models designed to mimic biological neural systems. In this entry, we introduce ANN using familiar econometric terminology and provide an overview of ANN modeling approach and its implementation methods. † Correspondence: Chung-Ming Kuan, Institute of Economics, Academia Sinica, 128 Academia Road, Sec. 2, Taipei 115, Taiwan; ckuan@econ.sinica.edu.tw. †† I would like to express my sincere gratitude to the editor, Professor Steven Durlauf, for his patience and constructive comments on early drafts of this entry. I also thank Shih-Hsun Hsu and Yu-Lieh Huang for very helpful suggestions. The remaining errors are all mine.

Artificial neural networks

In this technical report, we investigate the use of passive RFID technology for object localization and tracking. This work is motivated by the desire to design, and build, robust smart spaces (pervasive spaces), and in particular their application in the area of hazard aware spaces. In particular we are interested in building spaces that can detect hazards and autonomously take action to gain more information about the hazard and alert the users of the space. To enable a robot to operate in such a space, accurate localization and tracking must be carried out, allowing the robot to then sense and detect hazards and alert humans about when, where, and what hazards occur. We evaluate the strengths and weaknesses of passive RFID technology as a solution to this problem. In addition, we present a methodology for building a sensor model which is required for accurate localization, and we discuss one specific method of performing probabilistic localization. Finally, we present experimental results illustrating the successful global localization of a robot using these methods. We address the problem of building hazard aware spaces (HAS) to alert innocent people. We propose building a prototype HAS system for detecting fire, vision impairing light, earthquake, extreme sonic waves and dangerous gases. Our long term goal is to sense and detect indoor hazards and then alert humans with information about where, when and what hazards occur. In this project, we focus on incorporating passive radio frequency identification (RFID) technology to our existing prototype HAS system in order to improve localization of hazards, objects in a building and deployed wireless sensors.

/pdf/1-toward-hazard-aware-spaces-localization-using-passive-rfid-1rm39i00bi.pdf

1 Toward Hazard Aware Spaces: Localization using Passive RFID Technology

This paper addresses the problem of designing trojan detectors in neural networks (NNs) using interactive simulations. Trojans in NNs are defined as triggers in inputs that cause misclassification of such inputs into a class (or classes) unintended by the design of a NN-based model. The goal of our work is to understand encodings of a variety of trojan types in fully connected layers of neural networks. Our approach is: (1) to simulate nine types of trojan embeddings into dot patterns; (2) to devise measurements of NN states; and (3) to design trojan detectors in NN-based classification models. The interactive simulations are built on top of TensorFlow Playground with in-memory storage of data and NN coefficients. The simulations provide analytical, visualization, and output operations performed on training datasets and NN architectures. The measurements of a NN include: (a) model inefficiency using modified Kullback–Liebler (KL) divergence from uniformly distributed states; and (b) model sensitivity to variables related to data and NNs. Using the KL divergence measurements at each NN layer and per each predicted class label, a trojan detector is devised to discriminate NN models with or without trojans. To document robustness of such a trojan detector with respect to NN architectures, dataset perturbations, and trojan types, several properties of the KL divergence measurement are presented.

/pdf/designing-trojan-detectors-in-neural-networks-using-4njij4n1eo.pdf

Designing Trojan Detectors in Neural Networks Using Interactive Simulations.

This article introduces readers to a web-based solution useful for interactive nanoscale measurements of centimeter-sized specimens. This solution is a client-server system that promotes collaborative measurements and discovery. The system consists of multiple computational modules that enable uploading microscopy images, extracting metadata, assembling many nanometer-resolution images into an image covering a centimeter-sized area, and interactive viewing and measuring of objects of interest at multiple length scales over terabyte-sized images. We illustrate the use of the system on images of aerosolized nanoparticles and dye particles on printing paper.

/pdf/from-image-tiles-to-web-based-interactive-measurements-in-2eqznzlpr8.pdf

From Image Tiles to Web-Based Interactive Measurements in One Stop.

This paper addresses the problem of understanding preservation and reconstruction requirements for computer- aided medical decision-making. With an increasing number of computer-aided decisions having a large impact on our society, the motivation for our work is not only to document these decision processes semi-automatically but also to understand the preservation cost and related computational requirements. Our objective is to support computer-assisted creation of medical records, to guarantee authenticity of records, as well as to allow managers of electronic medical records (EMR), archivists and other users to explore and evaluate computational costs (e.g., storage and processing time) depending on several key characteristics of appraised records. Our approach to this problem is based on designing an exploratory simulation framework for investigating preservation tradeoffs and assisting in appraisals of electronic records. We have a prototype simulation framework called image provenance to learn (IP2Learn) to support computer-aided medical decisions based on visual image inspection. The current software enables to explore some of the tradeoffs related to (1) information granularity (category and level of detail), (2) representation of provenance information, (3) compression, (4) encryption, (5) watermarking and steganography, (6) information gathering mechanism, and (7) final medical report content (level of detail) and its format. We illustrate the novelty of IP2Learn by performing example studies and the results of tradeoff analyses for a specific image inspection task.

Understanding Challenges in Preserving and Reconstructing Computer-Assisted Medical Decision Processes

The objective of our project is to develop data mining and knowledge discovery in databases (KDD) techniques, using the “Data to Knowledge” (D2K) platform developed by National Center for Supercomputing Application (NCSA), to facilitate analysis, visualization and modeling of land-surface variables obtained from the TERRA and AQUA platforms in support of climate and weather applications. The project is developing capacity to access very large multivariate datasets; represent heterogeneous data types; integrate multiple GIS data sets stored in many GIS file formats; analyze variable relationships and model their dependencies using cluster and grid computing; and visualize input data, as well, as extracted features, integrated data sets and data mining results.

/pdf/using-d2k-data-mining-platform-for-understanding-the-dynamic-2au535igsz.pdf

Peter Bajcsy

Papers

1 Toward Hazard Aware Spaces: Localization using Passive RFID Technology

Designing Trojan Detectors in Neural Networks Using Interactive Simulations.

From Image Tiles to Web-Based Interactive Measurements in One Stop.

Understanding Challenges in Preserving and Reconstructing Computer-Assisted Medical Decision Processes

Using D2K Data Mining Platform for Understanding the Dynamic Evolution of Land-Surface Variables