Search

Deep learning (DL) algorithms have shown significant performance in various computer vision tasks. However, having limited labelled data lead to a network overfitting problem, where network performance is bad on unseen data as compared to training data. Consequently, it limits performance improvement. To cope with this problem, various techniques have been proposed such as dropout, normalization and advanced data augmentation. Among these, data augmentation, which aims to enlarge the dataset size by including sample diversity, has been a hot topic in recent times. In this article, we focus on advanced data augmentation techniques. we provide a background of data augmentation, a novel and comprehensive taxonomy of reviewed data augmentation techniques, and the strengths and weaknesses (wherever possible) of each technique. We also provide comprehensive results of the data augmentation effect on three popular computer vision tasks, such as image classification, object detection and semantic segmentation. For results reproducibility, we compiled available codes of all data augmentation techniques. Finally, we discuss the challenges and difficulties, and possible future direction for the research community. We believe, this survey provides several benefits i) readers will understand the data augmentation working mechanism to fix overfitting problems ii) results will save the searching time of the researcher for comparison purposes. iii) Codes of the mentioned data augmentation techniques are available at https://github.com/kmr2017/Advanced-Data-augmentation-codes iv) Future work will spark interest in research community., Comment: We need to make a lot changes to make its quality better

The development of computer vision and in-situ monitoring using visual sensors allows the collection of large datasets from the additive manufacturing (AM) process. Such datasets could be used with machine learning techniques to improve the quality of AM. This paper examines two scenarios: first, using convolutional neural networks (CNNs) to accurately classify defects in an image dataset from AM and second, applying active learning techniques to the developed classification model. This allows the construction of a human-in-the-loop mechanism to reduce the size of the data required to train and generate training data., Comment: 4 pages, accepted at the Irish Machine Vision and Image Processing Conference (IMVIP), Galway, August 2023

Recent developments in in-situ monitoring and process control in Additive Manufacturing (AM), also known as 3D-printing, allows the collection of large amounts of emission data during the build process of the parts being manufactured. This data can be used as input into 3D and 2D representations of the 3D-printed parts. However the analysis and use, as well as the characterization of this data still remains a manual process. The aim of this paper is to propose an adaptive human-in-the-loop approach using Machine Learning techniques that automatically inspect and annotate the emissions data generated during the AM process. More specifically, this paper will look at two scenarios: firstly, using convolutional neural networks (CNNs) to automatically inspect and classify emission data collected by in-situ monitoring and secondly, applying Active Learning techniques to the developed classification model to construct a human-in-the-loop mechanism in order to accelerate the labeling process of the emission data. The CNN-based approach relies on transfer learning and fine-tuning, which makes the approach applicable to other industrial image patterns. The adaptive nature of the approach is enabled by uncertainty sampling strategy to automatic selection of samples to be presented to human experts for annotation., Comment: 7 pages, 9 figures, 1 table. Presented at The 6th IEEE Workshop on Human-in-the-Loop Methods and Future of Work in BigData (IEEE HMData 2022) December 2022

In recent years we have seen significant advances in the technology used to both publish and consume Linked Data. However, in order to support the next generation of ebusiness applications on top of interlinked machine readable data suitable forms of access control need to be put in place. Although a number of access control models and frameworks have been put forward, very little research has been conducted into the security implications associated with granting access to partial data or the correctness of the proposed access control mechanisms. Therefore the contributions of this paper are two fold: we propose a query rewriting algorithm which can be used to partially restrict access to SPARQL 1.1 queries and updates; and we demonstrate how a set of criteria, which was originally used to verify that an access control policy holds over different database states, can be adapted to verify the correctness of access control via query rewriting.

We propose a new formal language for the expressive representation of probabilistic knowledge based on Answer Set Programming (ASP). It allows for the annotation of first-order formulas as well as ASP rules and facts with probabilities and for learning of such weights from data (parameter estimation). Weighted formulas are given a semantics in terms of soft and hard constraints which determine a probability distribution over answer sets. In contrast to related approaches, we approach inference by optionally utilizing so-called streamlining XOR constraints, in order to reduce the number of computed answer sets. Our approach is prototypically implemented. Examples illustrate the introduced concepts and point at issues and topics for future research., Comment: Appears in the Proceedings of the 15th International Workshop on Non-Monotonic Reasoning (NMR 2014)

Data provenance allows scientists in different domains validating their models and algorithms to find out anomalies and unexpected behaviors. In previous works, we described on-the-fly interpretation of (Python) scripts to build workflow provenance graph automatically and then infer fine-grained provenance information based on the workflow provenance graph and the availability of data. To broaden the scope of our approach and demonstrate its viability, in this paper we extend it beyond procedural languages, to be used for purely declarative languages such as logic programming under the stable model semantics. For experiments and validation, we use the Answer Set Programming solver oClingo, which makes it possible to formulate and solve stream reasoning problems in a purely declarative fashion. We demonstrate how the benefits of the provenance inference over the explicit provenance still holds in a declarative setting, and we briefly discuss the potential impact for declarative programming, in particular for instance-driven debugging of the model in declarative problem solving.

The Linked Open Data (LOD) Cloud forms a substantial and ever increasing portion of the global knowledge on theWorldWideWeb by contributing many distributed data sources. Due to the diversity of the knowledge that they expose, processing queries over these distributed autonomous sources can lead to interesting interdisciplinary discoveries and thus it is an important problem to address. As the cloud computing is gaining momentum, everything is expected to be provided as a service including query processing and knowledge extraction functionalities. This means a scalable, available and efficient implementation for query processing and knowledge extraction frameworks is required. However, Linked Data sources are distributed and autonomous and thus federating queries over them results in availability and performance problems. Caching and replication are the first class solutions to address availability and performance problems but lead to some trade-offs among quality metrics of provided response (i.e., quality of service). Quality metrics of response in a query processor with cache are the response latency and response consistency. In this thesis, I use caching to mitigate the availability and performance issues of processing queries over LOD. Therefore, this thesis addresses the problem of how to efficiently maintain a cache of distributed Linked Data by considering user-defined constraints on response consistency and latency. I decompose this problem into following sub-problems and address them throughout my thesis. First, to manage consistency/latency trade-off, metrics are required to quantify response consistency and latency. I discuss and motivate relevant metrics that I used to quantify them. Second, consistency/latency trade-off with consistency constraint is analyzed in the domain of Linked Data market systems. I propose a solution to estimate the response consistency on an existing state of the local cache. Moreover, I discuss how this solution can be leveraged to keep the maintenance at the bare minimum level and trigger it only if it is required to satisfy consistency constraint. Third, consistency/latency trade-off with latency constraints is analyzed in the domain of stream processing due to the critical importance of latency in this domain. My proposed maintenance framework refreshes the most influential cached entries on response freshness. Therefore, it maximizes the response freshness while respecting latency constraints. Fourth, I relax the assumption of having a big cache that can accommodate all required data. This adds completeness as an additional dimension for response consistency into the trade-off. The extended maintenance framework maximizes freshness (and completeness) when the latency and caching space is constrained. I show that increasing latency leads to higher freshness (and completeness) with a fixed cache size. Additionally, I show that the proposed maintenance framework outperforms the baselines in all evaluation metrics. The proposed policies in this thesis are contributed to an existing open source stream processor. Experimental results show that the performance of the extended system is boosted significantly.

The explosion of digital content and the heterogeneity of enterprise content sources have pushed existing data integration solutions to their boundaries. An alternative solution is to use the Resource Description Framework (RDF) together with the existing web infrastructure, commonly known as the Linked Data Web (LDW), as a means to integrate both public and private data. With the advent of SPARQL 1.1, it is possible not only to execute queries over the LDW, but also to use the SPARQL query language to maintain distributed graph data. However, such a decentralised architecture brings with it a number of additional challenges with respect to both data security and integrity. In this thesis, we focus on the problem of access control for the LDW. We are particularly interested in lifting both the data and the access control policies from existing line of business applications and enforcing and maintaining access control over linked data, irrespective of how it is published. We start by proposing a lifting strategy, which can be used to extract both data and access control policies from relational databases. The access permissions are represented using an extension of the RDF model known as annotated RDF (aRDF), which allows contextual information to be associated with RDF data. By using aRDF domain operations it is possible to combine different annotations for the same triple and to infer new annotations based on RDF Schema (RDFS) inference rules. We demonstrate how the proposed modelling, together with a set of inference rules, can be used to provide support for commonly used access control models, such as Role-Based Access Control, Attribute Based Access Control and View Based Access Control. With regards to the enforcement and administration of access control over RDF, we focus specifically on Discretionary Access Control (DAC). Given DAC allows users to delegate their permissions to others, it is particularly suitable for managing access control over distributed data. Although a number of authors demonstrate how they can used Semantic Web technology to represent DAC policies, the authors do not examine DAC from a data model perspective. In order to fill this gap, we provide a summary of access control requirements for the RDF data model, based on the different characteristics of the RDF data model compared to relational and tree data models. In order to support access control policy specification at the triple, resource, graph and schema level, authorisations are specified using graph patterns. In addition, we demonstrate how our proposed flexible graph based authorisation framework, which we call GFAF, can be used to cater for the specification, administration and enforcement of DAC policies over linked data. Given the proposed access control framework enforces access control at the query layer, when access to the requested RDF data is partially restricted, it is necessary to rewrite the query so that it behaves in the same manner as a query executed over a filtered dataset. Therefore, we propose a query rewriting strategy for both the SPARQL 1.1 query and update languages, that can be used to partially restrict access to unauthorised data. In addition, we demonstrate how a set of criteria, which was originally used to verify relational access control policies, can be adapted to ensure the correctness of access control over RDF via query rewriting.