It is generally perceived that Dynamic Sparse Training opens the door to a new era of scalability and efficiency for artificial neural networks at, perhaps, some costs in accuracy performance for the classification task. At the same time, Dense Training is widely accepted as being the "de facto" approach to train artificial neural networks if one would like to maximize their robustness against image corruption. In this paper, we question this general practice. Consequently, \textit{we claim that}, contrary to what is commonly thought, the Dynamic Sparse Training methods can consistently outperform Dense Training in terms of robustness accuracy, particularly if the efficiency aspect is not considered as a main objective (i.e., sparsity levels between 10\% and up to 50\%), without adding (or even reducing) resource cost. We validate our claim on two types of data, images and videos, using several traditional and modern deep learning architectures for computer vision and three widely studied Dynamic Sparse Training algorithms. Our findings reveal a new yet-unknown benefit of Dynamic Sparse Training and open new possibilities in improving deep learning robustness beyond the current state of the art.

Deep neural networks have evolved as the leading approach in 3D medical image segmentation due to their outstanding performance. However, the ever-increasing model size and computational cost of deep neural networks have become the primary barriers to deploying them on real-world, resource-limited hardware. To achieve both segmentation accuracy and efficiency, we propose a 3D medical image segmentation model called Efficient to Efficient Network (E2ENet), which incorporates two parametrically and computationally efficient designs. i. Dynamic sparse feature fusion (DSFF) mechanism: it adaptively learns to fuse informative multi-scale features while reducing redundancy. ii. Restricted depth-shift in 3D convolution: it leverages the 3D spatial information while keeping the model and computational complexity as 2D-based methods. We conduct extensive experiments on AMOS, Brain Tumor Segmentation and BTCV Challenge, demonstrating that E2ENet consistently achieves a superior trade-off between accuracy and efficiency than prior arts across various resource constraints. %In particular, with a single model and single scale, E2ENet achieves comparable accuracy on the large-scale challenge AMOS-CT, while saving over 69% parameter count and 27% FLOPs in the inference phase, compared with the previous best-performing method. Our code has been made available at: https://github.com/boqian333/E2ENet-Medical.

Interpretable methods based on prototypical patches recognize various components in an image in order to explain their reasoning to humans. However, existing prototype-based methods can learn prototypes that are not in line with human visual perception, i.e., the same prototype can refer to different concepts in the real world, making interpretation not intuitive. Driven by the principle of explainability-by-design, we introduce PIP-Net (Patch-based Intuitive Prototypes Network): an interpretable image classification model that learns prototypical parts in a self-supervised fashion which correlate better with human vision. PIP-Net can be interpreted as a sparse scoring sheet where the presence of a prototypical part in an image adds evidence for a class. The model can also abstain from a decision for out-of-distribution data by saying “I haven't seen this before”. We only use image-level labels and do not rely on any part annotations. PIP-Net is globally interpretable since the set of learned prototypes shows the entire reasoning of the model. A smaller local explanation locates the relevant prototypes in one image. We show that our prototypes correlate with ground-truth object parts, indicating that PIP-Net closes the “semantic gap” between latent space and pixel space. Hence, our PIP-Net with interpretable prototypes enables users to interpret the decision making process in an intuitive, faithful and semantically meaningful way. Code is available at https://github.com/M-Nauta/PIPNet.

Hospitals often set protocols based on well defined standards to maintain the quality of patient reports. To ensure that the clinicians conform to the protocols, quality assurance of these reports is needed. Patient reports are currently written in free-text format, which complicates the task of quality assurance. In this paper, we present a machine learning based natural language processing system for automatic quality assurance of radiology reports on breast cancer. This is achieved in three steps: we i) identify the top-level structure (headings) of the report, ii) classify the report content into the top-level headings, and iii) convert the free-text detailed findings in the report to a semi-structured format (post-structuring). Top level structure and content of report were predicted with an F1 score of 0.97 and 0.94, respectively, using Support Vector Machine (SVM) classifiers. For automatic structuring, our proposed hierarchical Conditional Random Field (CRF) outperformed the baseline CRF with an F1 score of 0.78 versus 0.71. The determined structure of the report is represented in semi-structured XML format of the free-text report, which helps to easily visualize the conformance of the findings to the protocols. This format also allows easy extraction of specific information for other purposes such as search, evaluation, and research.

By understanding the tremendous opportunities to work with social media data and the acknowledgment of the negative effects social media messages can have, a way of assessing truth in claims on social media would not only be interesting but also very valuable. By making use of this ability, applications using social media data could be supported, or a selection tool in research regarding the spread of false rumors or 'fake news' could be build. In this paper, we show that we can determine truth by using a statistical classifier supported by an architecture of three preprocessing phases. We base our research on a dataset of Twitter messages about the FIFA World Cup 2014. We determine the truth of a tweet by using 7 popular fact types (involving events in the matches in the tournament such as scoring a goal) and we show that we can achieve an F1-score of 0.988 for the first class; the Tweets which contain no false facts and an F1-score of 0.818 on the second class; the Tweets which contain one or more false facts.

Social media accounts are valuable for hackers for spreading phishing links, malware and spam. Furthermore, some people deliberately hack an acquaintance to damage his or her image. This paper describes a classification for detecting hacked Twitter accounts. The model is mainly based on features associated with behavioural change such as changes in language, source, URLs, retweets, frequency and time. We experiment with a Twitter data set containing tweets of more than 100 Dutch users including 37 who were hacked. The model detects 99% of the malicious tweets which proves that behavioural changes can reveal a hack and that anomaly-based features perform better than regular features. Our approach can be used by social media systems such as Twitter to automatically detect a hack of an account only a short time after the fact allowing the legitimate owner of the account to be warned or protected, preventing reputational damage and annoyance.

Existing concurrency control systems cannot execute transactions with overlapping updates concurrently. This is especially problematic for bulk updates, which usually overlap with all concurrent transactions. To solve this, we have developed a concurrency control mechanism based on lazy evaluation, which moves evaluation of operations from the writer to the reader. This allows readers to prioritize evaluation of those operations in which they are interested, without loss of atomicity of transactions. To handle bulk operations, we dynamically split large transactions into transactions on smaller parts of the data. In this paper we present an abstract lazy index structure for lazy transactions, and show how transactions can be encoded to effectively use this data structure. Moreover, we discuss evaluation strategies for lazy transactions, where trade-offs can be made between latency and throughput. To evaluate our approach, we have implemented a concurrent lazy trie, on which we performed a number of micro benchmarks.

Web content changes rapidly [18]. In Focused Web Harvesting [17] which aim it is to achieve a complete harvest for a given topic, this dynamic nature of the web creates problems for users who need to access a set of all the relevant web data to their topics of interest. Whether you are a fan following your favorite idol or a journalist investigating a topic, you may need not only to access all the relevant information but also the recent changes and updates. General search engines like Google apply several techniques to enhance the freshness of their crawled data. However, in focused web harvesting, we lack an efficient approach that detects changes for a given topic over time. In this paper, we focus on techniques that can keep the relevant content to a given query up-to-date. To do so, we test four different approaches to efficiently harvest all the changed documents matching a given entity by querying web search engines. We define a document with changed content or a newly created or removed document as a changed document. Among the proposed change detection approaches, the FedWeb method outperforms the other approaches in finding the changed content on the web for a given query with 20 percent, on average, better performance.

Abstract Twitter is a rich source of continuously and instantly updated information. Shortness and informality of tweets are challenges for Natural Language Processing tasks. In this paper, we present TwitterNEED, a hybrid approach for Named Entity Extraction and Named Entity Disambiguation for tweets. We believe that disambiguation can help to improve the extraction process. This mimics the way humans understand language and reduces error propagation in the whole system. Our extraction approach aims for high extraction recall first, after which a Support Vector Machine attempts to filter out false positives among the extracted candidates using features derived from the disambiguation phase in addition to other word shape and Knowledge Base features. For Named Entity Disambiguation, we obtain a list of entity candidates from the YAGO Knowledge Base in addition to top-ranked pages from the Google search engine for each extracted mention. We use a Support Vector Machine to rank the candidate pages according to a set of URL and context similarity features. For evaluation, five data sets are used to evaluate the extraction approach, and three of them to evaluate both the disambiguation approach and the combined extraction and disambiguation approach. Experiments show better results compared to our competitors DBpedia Spotlight, Stanford Named Entity Recognition, and the AIDA disambiguation system.

This paper presents a benchmark for measuring the blocking behavior of schema transformations in relational database systems. As a basis for our benchmark, we have developed criteria for the functionality and performance of schema transformation mechanisms based on the characteristics of state of the art approaches. To address limitations of existing approaches, we assert that schema transformations must be composable while satisfying the ACID guarantees like regular database transactions. Additionally, we have identified important classes of basic and complex relational schema transformations that a schema transformation mechanism should be able to perform. Based on these transformations and our criteria, we have developed a benchmark that extends the standard TPC-C benchmark with schema transformations, which can be used to analyze the blocking behavior of schema transformations in database systems. The goal of the benchmark is not only to evaluate existing solutions for non-blocking schema transformations, but also to challenge the database community to find solutions that allow more complex transactional schema transformations.

With the goal of harvesting all information about a given entity, in this paper, we try to harvest all matching documents for a given query submitted on a search engine. The objective is to retrieve all information about for instance "Michael Jackson", "Islamic State", or "FC Barcelona" from indexed data in search engines, or hidden data behind web forms, using a minimum number of queries. Policies of web search engines usually do not allow accessing all of the matching query search results for a given query. They limit the number of returned documents and the number of user requests. These limitations are also applied in deep web sources, for instance in social networks like Twitter. In this work, we propose a new approach which automatically collects information related to a given query from a search engine, given the search engine's limitations. The approach minimizes the number of queries that need to be sent by analysing the retrieved results and combining this analysed information with information from a large external corpus. The new approach outperforms existing approaches when tested on Google, measuring the total number of unique documents found per query.

Relation extraction involves different types of uncertainty due to the imperfection of the extraction tools and the inherent ambiguity of unstructured text. In this paper, we discuss several ways of handling uncertainties in relation extraction from social media. Our study case is to extract tennis games' results for two Grand Slam tennis tournaments from tweets. Analysis has been done to find to what extent it is useful to use semantic web, domain knowledge, facts repetition, and authors' trustworthiness to improve the certainty of the extracted relations.

Trajectories of people contain a vast amount of information on users' interests and popularity of locations. To obtain this information, the places visited by the owner of the device on such a trajectory need to be recognized. However, the location information on a point of interest (POI) in a database is normally limited to an address and a coordinate pair, rather than a polygon describing its boundaries. A region of interest can be used to intersect trajectories to match trajectories with objects of interest. In the absence of expensive and often not publicly available detailed spatial data like cadastral data, we need to approximate this ROI. In this paper, we present several approaches to approximate the size and shape of ROIs, by integrating data from multiple public sources, a validation technique, and a validation of these approaches against the cadastral data of the city of Enschede, The Netherlands.

With the pervasiveness of computers and mobile devices, digital forensics becomes more important in law enforcement. Detectives increasingly depend on the scarce support of digital specialists which impedes efficiency of criminal investigations. Tracks Inspector is a commercial solution that enables non-technical investigators to easily investigate digital evidence using a web browser. We will demonstrate how Tracks Inspector can be used to discover the most important persons and groups in case data by investigators without requiring the help of digital forensics experts.

With the increasing amount of data in deep web sources (hidden from general search engines behind web forms), accessing this data has gained more attention. In the algorithms applied for this purpose, it is the knowledge of a data source size that enables the algorithms to make accurate decisions in stopping the crawling or sampling processes which can be so costly in some cases [14]. This tendency to know the sizes of data sources is increased by the competition among businesses on the Web in which the data coverage is critical. In the context of quality assessment of search engines [7], search engine selection in the federated search engines, and in the resource/collection selection in the distributed search field [19], this information is also helpful. In addition, it can give an insight over some useful statistics for public sectors like governments. In any of these mentioned scenarios, in the case of facing a non-cooperative collection which does not publish its information, the size has to be estimated [17]. In this paper, the suggested approaches for this purpose in the literature are categorized and reviewed. The most recent approaches are implemented and compared in a real environment. Finally, four methods based on the modification of the available techniques are introduced and evaluated. In one of the modifications, the estimations from other approaches could be improved ranging from 35 to 65 percent.

We demonstrate ROX, a run-time optimizer of XQueries, that focuses on finding the best execution order of XPath steps and relational joins in an XQuery. The problem of join ordering has been extensively researched, but the proposed techniques are still unsatisfying. These either rely on a cost model which might result in inaccurate estimations, or explore only a restrictive number of plans from the search space. ROX is developed to tackle these problems. ROX does not need any cost model, and defers query optimization to run-time intertwining optimization and execution steps. In every optimization step, sampling techniques are used to estimate the cardinality of unexecuted steps and joins to make a decision which sequence of operators to process next. Consequently, each execution step will provide updated and accurate knowledge about intermediate results, which will be used during the next optimization round. This demonstration will focus on: (i) illustrating the steps that ROX follows and the decisions it makes to choose a good join order, (ii) showing ROX's robustness in the face of data with different degree of correlation, (iii) comparing the performance of the plan chosen by ROX to different plans picked from the search space, (iv) proving that the run-time overhead needed by ROX is restricted to a small fraction of the execution time.

Optimization of complex XQueries combining many XPath steps and joins is currently hindered by the absence of good cardinality estimation and cost models for XQuery. Additionally, the state-of-the-art of even relational query optimization still struggles to cope with cost model estimation errors that increase with plan size, as well as with the effect of correlated joins and selections.

IMPrECISE is an XQuery module that adds probabilistic XML functionality to an existing XML DBMS, in our case MonetDB/XQuery. We demonstrate probabilistic XML and data integration functionality of IMPrECISE. The prototype is configurable with domain knowledge such that the amount of uncertainty arising during data integration is reduced to an acceptable level, thus obtaining a "good is good enough" data integration with minimal human effort.

Relational XQuery systems try to re-use mature relational data management infrastructures to create fast and scalable XML database technology. This paper describes the main features, key contributions, and lessons learned while implementing such a system. Its architecture consists of (i) a range-based encoding of XML documents into relational tables, (ii) a compilation technique that translates XQuery into a basic relational algebra, (iii) a restricted (order) property-aware peephole relational query optimization strategy, and (iv) a mapping from XML update statements into relational updates. Thus, this system implements all essential XML database functionalities (rather than a single feature) such that we can learn from the full consequences of our architectural decisions. While implementing this system, we had to extend the state-of-the-art with a number of new technical contributions, such as loop-lifted staircase join and efficient relational query evaluation strategies for XQuery theta-joins with existential semantics. These contributions as well as the architectural lessons learned are also deemed valuable for other relational back-end engines. The performance and scalability of the resulting system is evaluated on the XMark benchmark up to data sizes of 11GB. The performance section also provides an extensive benchmark comparison of all major XMark results published previously, which confirm that the goal of purely relational XQuery processing, namely speed and scalability, was met.

In mobile and ambient environments, devices need to become autonomous, managing and resolving problems without interference from a user. The database of a (mobile) device can be seen as its knowledge about objects in the 'real world'. Data exchange between small and/or large computing devices can be used to supplement and update this knowledge whenever a connection gets established. In many situations, however, data from different data sources referring to the same real world objects, may conflict. It is the task of the data management system of the device to resolve such conflicts without interference from a user. In this paper, we take a first step in the development of a probabilistic XML DBMS. The main idea is to drop the assumption that data in the database should be certain: subtrees in XML documents may denote possible views on the real world. We formally define the notion of probabilistic XML tree and several operations thereon. We also present an approach for determining a logical semantics for queries on probabilistic XML data. Finally, we introduce an approach for XML data integration where conflicts are resolved by the introduction of possibilities in the database.

This article is a proposal for a database index structure, the XPath accelerator , that has been specifically designed to support the evaluation of XPath path expressions. As such, the index is capable to support all XPath axes (including ancestor, following, preceding-sibling, descendant-or-self, etc.). This feature lets the index stand out among related work on XML indexing structures which had a focus on the child and descendant axes only. The index has been designed with a close eye on the XPath semantics as well as the desire to engineer its internals so that it can be supported well by existing relational database query processing technology: the index (a) permits set-oriented (or, rather, sequence-oriented) path evaluation, and (b) can be implemented and queried using well-established relational index structures, notably B-trees and R-trees.We discuss the implementation of the XPath accelerator on top of different database backends and show that the index performs well on all levels of the memory hierarchy, including disk-based and main-memory based database systems.

No abstract available.