Computer Science Journal (AGH University of Science and Technology, Krakow)
Not a member yet
476 research outputs found
Sort by
Enhanced Cluster Merging and Deep Learning Techniques for Entity Name Identification from Biomedical Corpus
Full text linkFor mining biomedical information identifying names is the prime task. Complex and uncertain naming styles of biomedical entities are the major setbacks here. Thus, state-of-the-art accuracy of biomedical name identification is reasonably inferior compared to general domain. This study includes machine learning and deep learning techniques to recognize names from biomedical corpus. In supervised classification, a classifier is built by finding required statistics from training corpus. Accordingly, performance of the system is primarily dependent on quantity and quality of training corpus. But manually preparing a large training dataset with enriched feature samples is laborious and time-taking. Therefore, various techniques were adopted in the literature to make effective use of raw corpora. We have incorporated a novel Cluster Merging technique and Attention Mechanism with BERT embedding for boosting machine learning and deep learning classifiers respectively. The suggested results outpour that profound techniques are competent and delineate signifying improvement over surviving methods
ATTENTION-BASED MULTIPLE-REPRESENTATION METHOD FOR FINGERPRINT-PRESENTATION-ATTACK DETECTION
Full text linkFingerprint biometrics are one of the most common authentication mechanisms. However, such systems are often compromised by presentation attacks by presentation attack instruments. Most of the fingerprint presentation attack detection approaches show poor performance due to the large variation in presentation attack instruments and limited feature representation of input fingerprint. Therefore this article proposes a hybrid model of shallow and deep features with multiple representations of input fingerprints. To obtain these shallow and deep features first we have enhanced the texture of the input fingerprint through a novel median adaptive local binary pattern filter and existing binarised statistical image feature. After that, the input fingerprint image and two textured enhanced images are concatenated along with the channel dimension for multiple representations. Finally, an extended ResNeXt architecture with channel and spatial attention (EResNeXt) has been used for relevant feature extraction and presentation attack detection. The proposed model (EResNeXt) has been assessed on LivDet-2015 and Livdet-2017 datasets and provides significant results in unknown presentation attack instrument scenarios
A NETWORK-BASED COMPUTATIONALPIPELINE TO STUDY THE VARIABILITY OFTRANSCRIPTOME PROFILES FOR HUMANDISEASES
No full textMachine learning applications to high-throughput data in medicine – one of thebiggest resource for understanding complex disease – so far have been limited.Here we present a computational approach for assessing the intrinsic variabilityin the most prominent data type, transcriptomics data for disease cohorts. Ourstudy looks at situations, where multiple datasets for the same disease areavailable. We leverage concepts of network medicine to assess, how the matchbetween a biological network and a set of differentially expressed genes variesacross different networks and experiments. Our results show that differentbiological networks yield markedly different results. Also, the clustering ofdiseases depends strongly on the choice of parameters contained in the dataanalysis and network processing
ANAPHORA SOLVED AD-DL-BERT MODEL FOR TEXT SUMMARIZATION WITH AUTO ENCODING USING THE TOPIC DESCRIPTION AND SEVERAL PRIORS (ATDS) APPROACH
Full text linkOwing to the large amount of digital text content in articles, novels stories, and so on, Automatic Text Summarization (ATS) is becoming a significant task. Abstractive or extractive summaries of single or multi documents have been generated by various researchers. Although several models were generated, there are still limitations like the anaphora problem that occurred during the summarization. To overcome such limitations, this paper proposes the Added dropout-Deleted Layer norm-Bidirectional Encoder Representations from Transformers (Ad-DL-BERT)-based Extractive Text Summarization (ETS). Primarily, the input document’s sentences are prepared for accurate summarization by pre-processing; then, the unwanted sentences are removed. Afterward, with the Auto encoding using the Topic Description and Several priors (ATDS) approach, the sentences under the same topic are clustered. Moreover, keywords for summarization are extracted with an Anaphora-POS (An-POS) extractor. Thereafter, for removing the redundant sentences, the ranking with Exponential Linear Unit-Generative Adversarial Network (ELU-GAN) and saliency score assignment processes are performed. Also, assignments for sentences are performed to enhance the coherency, sorting, and cosine similarity score. Lastly, the Ad-DL-BERT generated summary and the proposed technique’s performance are evaluated on the Document Understanding Conference (DUC2002) dataset. Regarding clustering time, execution time, Recall-Oriented Understudy for Gisting Evaluation (ROUGE-1) scores of recall, F-measure, and precision, the experimental outcomes exhibited the proposed techniques’ dominance over the conventional approaches.
 
TREEXTRUST:TOPIC-AWARE COMPUTATIONAL TRUST BASED ON INTERACTION EXPERIENCE, REPUTATION OF USERS WITH SIMILARITY, AND PATH ALGEBRA OF GRAPH IN SOCIAL NETWORKS
Full text linkTrust measure is confidence or reliability among users or peers and has been studied widely in online social networks. Most trust models currently are based on the concepts of interaction trust and reputation trust. However, various forms of interaction and analysis of interaction contexts have been not considered fully for trust estimation. Moreover, the mechanism for computing reputation trust based on propagation lacks a clear foundation and is expensive in computation. The purpose of this paper is to present a family of models of computational trust, named TreeXTrust, for estimating a trust degree of a user truster on another user trustee. Our model is a mathematical formulation based on aggregation of the topic-aware experience trust with various forms of interaction and the topic-aware reputation trust with users’ similarity and operators on path algebra in graph. We conduct experiments to evaluate how impact of interaction forms and users’ interests on experience trust and the correlation of experience trust and reputation trust on overall trust estimation. Our experimental results have demonstrated that: (i) Interest degrees influence on experience trust more than interaction ones; (ii) Community evaluation of some trustee affects the overall trust estimation more than the truster’s individual evaluation. Our family of models outperforms the state of art methods presented in the literature and is a framework for selecting and implementing a suitable model of computational trust for our problem at hand
INTRUSION DETECTION USING FEDERATED LEARNING WITH NEURAL NETWORKS
No full textThe amount of information shared amongst different devices and the variety of novel methods of network crimes have exponentially increased in recent years because of the widespread use of the internet. Quick identification of all types of attacks would not be possible with conventional methods including firewalls, which focused on data filtering. Dealing with the timely recognition of these types of assaults is very successful for intrusion detection systems (IDS) grounded on ML algorithms. They can efficiently manage the enormous amount of data in order to identify any harmful behaviour. Every network activity is searched for any possibly dangerous activity using IDS based on machine learning. The main objective of the planned effort is to provide analytical analyses of such current intrusion detection systems. Furthermore, examined in this work are the useful data sets and several techniques already in use to develop an effective IDS using single, hybrid, and ensemble machine learning algorithms. The approaches in the literature have then been investigated under several criteria in line to provide a clear road and direction for the next projects that will be successful. Nowadays, companies of all kinds include an intrusion detection system (IDS), which inhibits cybercrime to protect the network, resources, and private data. Many strategies have been suggested and implemented up till now to prevent uncivil behaviour. Since machine learning (ML) approaches are successful, the proposed approach applied several ML models for the intrusion detection system. The CIC IOT 2023 Dataset is the one applied in this paper. Tested were several techniques including random forest, XG Boost, logistic regression, MLP model, and RNN. Following fine-tuning, the federated learning model using neural networks had the best accuracy—99.84%
A Proposal of Digital Contents Copyright Protection by using Blockmarking Technique
Full text linkRecently, blockmarking technique \cite{blockmarking} is proposed for a new hybrid model based on the combination of blockchain and watermarking method. In this model, it not only achieves the goal of image copyright protection but also stores the image into the blockchain network such as IPFS system. In this paper, we propose a new DRM system by inheriting the idea of blockmarking. The copyright contents can be distributed via IPFS blockchain, then be restored by using the reconstruction license for each legal user. Also, in our method, based on the reconstruction licenses, the distributed contents can be reconstructed from IPFS with various watermarking patterns. It helps us can manage the legal users and trace the traitor if a dispute occurs. The experimental results show that our method successfully achieved the purpose of digital copyright protection
Character/Word Modelling: A Two-Step Framework for Text Recognition in Natural Scene Images
Full text linkText recognition from images is a complex task in computer vision. Traditional text recognition methods typically rely on Optical Character Recognition (OCR); however, their limitations in image processing can lead to unreliable results. However, recent advancements in deep-learning models have provided an effective alternative for recognizing and classifying text in images. This study proposes a deep-learning-based text recognition system for natural scene images that incorporates character/word modeling, a two-step procedure involving the recognition of characters and words. In the first step, Convolutional Neural Networks (CNN) are used to differentiate individual characters from image frames. In the second step, the Viterbi search algorithm employs lexicon-based word recognition to determine the optimal sequence of recognized characters, thereby enabling accurate word identification in natural scene images. The system is tested using the ICDAR 2003 and ICDAR 2013 datasets from the Kaggle repository, and achieved accuracies of 79.8% and 81.5%, respectively
MODIFIED HONEY BEE ALGORITHM WITH RANDOM SELECTION OF VIRTUAL MACHINES FOR DYNAMIC LOAD BALANCING
Full text linkCloud workloads can overwhelm load balancers, leading to inefficiencies and performance issues. To address these challenges, the Honey Bee Load Balancing algorithm is highly effective in enhancing cloud resource allocation. Inspired by the foraging behavior of honey bees, this algorithm offers a dynamic approach to resource distribution, adapting to changing workloads in real-time. This paper delves into the key features and advantages of Honey Bee Load Balancing, focusing on its dynamic resource allocation, overall response time, and data center processing time. Through a comparative study of existing methodologies, we propose a modified Honey Bee Load Balancing algorithm that incorporates the random selection of virtual machines. Utilizing the CloudAnalyst tool for simulation, we compare traditional and proposed Honey Bee Load Balancing algorithms to evaluate overall response time and data center processing time. The proposed algorithm demonstrates superior performance in these parameters compared to the traditional approach
STABLE AND LOW ASSOCIATIVE LEFT-RIGHT HASHING
Full text linkHashing is indispensable for efficient search operations, captivating the interest of numerous researchers. Among the diverse array of techniques, Cuckoo Hashing has emerged as particularly effective across a wide range of applications. It is implemented in two primary forms: Parallel Cuckoo Hashing and Sequential Cuckoo Hashing. Nonetheless, Cuckoo Hashing encounters significant challenges, including high insertion latency, inefficient memory usage, and high data migration costs. The concept of Combinatorial Hashing has inspired this research. Our proposed scheme enhances Combinatorial Hashing and introduces an innovative collision resolution technique called Left-Right Random Probing. This advanced variant of random probing strategically utilizes prime numbers and Fibonacci sequences to improve performance. This paper introduces two performance indicators, the degree of dexterity and table reference count per key. This paper identifies switching cost as a new challenge in Cuckoo Hashing and quantifies this switching cost using the parameter, table reference count per key