Search

This paper deals with the challenging problem of simultaneously integrating interpretablility and reliability into prediction models in machine learning. It proposes to combine the interpretable models of decision rules with the reliable models based on conformal prediction. The result is a new technique of conformal decision rules. Given a test instance, the technique is capable of providing a point prediction, an explanation, and a confidence value for that prediction plus a prediction set. The experiments show when and how conformal decision rules can be used for interpretable and reliable machine learning.

Entity Resolution (ER) is the task of finding references that refer to the same entity across different data sources. Cleaning a data warehouse and applying ER on it is a computationally demanding task, particularly for large data sets that change dynamically. Therefore, a query-driven approach which analyses a small subset of the entire data set and integrates the results in real-time is significantly beneficial. Here, we present an interactive tool, called HiDER, which allows for query-driven ER in large collections of uncertain dynamic historical data. The input data includes civil registers such as birth, marriage and death certificates in the form of structured data, and notarial acts such as estate tax and property transfers in the form of free text. The outputs are family networks and event timelines visualized in an integrated way. The HiDER is being used and tested at BHIC center(Brabant Historical Information Center, https://¿www.¿bhic.¿nl); despite the uncertainties of the BHIC input data, the extracted entities have high certainty and are enriched by extra information.

Supplementing Vovk and V'yugin's `if' statement, we show that Bayesian compression provides the best enumerable compression for parameter-typical data if and only if the parameter is Martin-L\"of random with respect to the prior. The result is derived for uniformly discretizable statistical models, introduced here. They feature the crucial property that given a~discretized parameter, we can compute how much data is needed to learn its value with little uncertainty. Exponential families and certain nonparametric models are shown to be uniformly discretizable.

Edge computing is a paradigm in which data is intelligently processed close to its source. Along with advancements in deep learning, there is a growing interest in using deep neural networks at the edge for predictive analytics. Given the realistic constraints in computational resources of edge devices, this combination is challenging. In order to bridge the gap between deep learning models and efficient edge analytics, a container-based framework is presented that evaluates user-specified deep learning models for efficiency on the edge. The proposed framework is validated on a rotating machinery fault diagnosis use case. Conclusions on efficient state-of-the-art models for rotating machine fault diagnosis were drawn and appropriately reported. ispartof: pages:569-581 ispartof: Communications in Computer and Information Science vol:1752 pages:569-581 ispartof: Joint European Conference on Machine Learning and Knowledge Discovery in Databases location:Grenoble, France date:19 Sep - 23 Sep 2022 status: Published online

People using mobile devices for making phone calls, accessing the internet, or posting georeferenced contents in social media create episodic digital traces of their presence in various places. Availability of personal traces over a long time period makes it possible to detect repeatedly visited places and identify them as home, work, place of social activities, etc. based on temporal patterns of the person’s presence. Such analysis, however, can compromise personal privacy. We propose a visual analytics approach to semantic analysis of mobility data in which traces of a large number of people are processed simultaneously without accessing individual-level data. After extracting personal places and identifying their meanings in this privacy-respectful manner, the original georeferenced data are transformed to trajectories in an abstract semantic space. The semantically abstracted data can be further analyzed without the risk of re-identifying people based on the specific places they attend.

We present ProbLog2, the state of the art implementation of the probabilistic programming language ProbLog. The ProbLog language allows the user to intuitively build programs that do not only encode complex interactions between a large sets of heterogenous components but also the inherent uncertainties that are present in real-life situations. The system provides efficient algorithms for querying such models as well as for learning their parameters from data. It is available as an online tool on the web and for download. The offline version offers both command line access to inference and learning and a Python library for building statistical relational learning applications from the system’s components. ispartof: pages:312-315 ispartof: Lecture Notes in Computer Science vol:9286 pages:312-315 ispartof: ECML PKDD location:Porto, Portugal date:7 Sep - 11 Sep 2015 status: published

We demonstrate a system of tools for real-time detection of significant clusters of spatial events and observing their evolution. The tools include an incremental stream clustering algorithm, interactive techniques for controlling its operation, a dynamic map display showing the current situation, and displays for investigating the cluster evolution (time line and space-time cube).

We demonstrate interactive visual embedding of partition-based clustering of multidimensional data using methods from the open-source machine learning library Weka. According to the visual analytics paradigm, knowledge is gradually built and refined by a human analyst through iterative application of clustering with different parameter settings and to different data subsets. To show clustering results to the analyst, cluster membership is typically represented by color coding. Our tools support the color consistency between different steps of the process. We shall demonstrate two-way clustering of spatial time series, in which clustering will be applied to places and to time steps.

We present Viper, for Visual Pattern Explorer, an innovative, browser-based application for interactive pattern exploration, assisted by visualisation, recommendation, and algorithmic search. The target audience consists of domain experts who have access to data but not to ---potentially expensive--- data mining experts. The goal of the system is to enable the target audience to perform true exploratory data mining. That is, to discover interesting patterns from data, with a focus on subgroup discovery but also facilitating frequent itemset mining.