Self-learning Approach Research Articles

Optimizing Automated Optical Inspection: An Adaptive Fusion and Semi-Supervised Self-Learning Approach for Elevated Accuracy and Efficiency in Scenarios with Scarce Labeled Data.

In the field of automatic optical inspection (AOI), this study presents innovative strategies to enhance object detection accuracy while minimizing dependence on large annotated datasets. We initially developed a defect detection model using a dataset of 3579 images across 32 categories, created in collaboration with a major Taiwanese panel manufacturer. This model was evaluated using 12,000 ambiguously labeled images, with improvements achieved through data augmentation and annotation refinement. To address the challenges of limited labeled data, we proposed the Adaptive Fused Semi-Supervised Self-Learning (AFSL) method. This approach, designed for anchor-based object detection models, leverages a small set of labeled data alongside a larger pool of unlabeled data to enable continuous model optimization. Key components of AFSL include the Bounding Box Assigner, Adaptive Training Scheduler, and Data Allocator, which together facilitate dynamic threshold adjustments and balanced training, significantly enhancing the model's performance on AOI datasets. The AFSL method improved the mean average precision (mAP) from 43.5% to 57.1% on the COCO dataset and by 2.6% on the AOI dataset, demonstrating its effectiveness in achieving high levels of precision and efficiency in AOI with minimal labeled data.

Dual Encoder–Decoder Shifted Window-Based Transformer Network for Polyp Segmentation With Self-Learning Approach

Dual Encoder–Decoder Shifted Window-Based Transformer Network for Polyp Segmentation With Self-Learning Approach

Automatic classification of emotions in speech: methods and data

The subject of this study is the data and methods used in the task of automatic recognition of emotions in spoken speech. This task has gained great popularity recently, primarily due to the emergence of large datasets of labeled data and the development of machine learning models. The classification of speech utterances is usually based on 6 archetypal emotions: anger, fear, surprise, joy, disgust and sadness. Most modern classification methods are based on machine learning and transformer models using a self-learning approach, in particular, models such as Wav2vec 2.0, HuBERT and WavLM, which are considered in this paper. English and Russian datasets of emotional speech, in particular, the datasets Dusha and RESD, are analyzed as data. As a method, an experiment was conducted in the form of comparing the results of Wav2vec 2.0, HuBERT and WavLM models applied to the relatively recently collected Russian datasets of emotional speech Dusha and RESD. The main purpose of the work is to analyze the availability and applicability of available data and approaches to recognizing emotions in speech for the Russian language, for which relatively little research has been conducted up to this point. The best result was demonstrated by the WavLM model on the Dusha dataset - 0.8782 dataset according to the Accuracy metric. The WavLM model also received the best result on the RESD dataset, while preliminary training was conducted for it on the Dusha - 0.81 dataset using the Accuracy metric. High classification results, primarily due to the quality and size of the collected Dusha dataset, indicate the prospects for further development of this area for the Russian language.

AI-Based Self-Learning Robotic Arm Using Microcontroller

Robotic arms are essential components of day-to-day life likely manufacturing and logistics operations to provide more speed, more flexible decision-making and reduce the possibility of errors. The objective of this project is to make a robotic arm with machine learning features to automate the assembly by object and colour detection using openCV, and TensorFlow library. The basic arm is made of light material to provide maximum mobility which is controlled by an ATmega328P microcontroller with the aid of an Arduino development environment for hassle-free interfacing sensors. A webcam is used for sensing the surroundings. Feed from waveform is processed by openCV and TensorFlow library Using a Python environment the real-time coordinates data and colour information will be fed to the microcontroller unit that controls the arm servos accordingly to grip the appropriate object and place them in the predetermined location. The arm is also capable of learning trajectory which is taught to it using human hands. The use of easily available components reduces the cost but the same time leaves much room for future improvement. There are two approaches to machine learning used in robotic systems, the first one is a conventional approach to machine learning which is a complicated, expansive asset that is not suitable for several types of automation industries. A self-learning approach, called a behaviour-based robotic arm with a machine learning feature to automate the assembly by the object and colour detection using openCV, a python environment to be built to address the limitations of conventional robotics methodologies. This approach aims to enhance the capability of robotic platforms to offer more flexible decision-making, cost-effectiveness, reduced human error, and higher precision in processes.

A Self-Learning Approach to Heterogeneous Multi-Robot Coalition Formation Under Uncertainty

A Self-Learning Approach to Heterogeneous Multi-Robot Coalition Formation Under Uncertainty

Dual-model self-regularization and fusion for domain adaptation of robust speaker verification

Learning robust representations of speaker identity is a key challenge in speaker verification, as it results in good generalization for many real-world speaker verification scenarios with domain or intra-speaker variations. In this study, we aim to improve the well-established ECAPA-TDNN framework to enhance its domain robustness for low-resource cross-domain speaker verification tasks. Specifically, a novel dual-model self-learning approach is first proposed to produce robust speaker identity embeddings, where the ECAPA-TDNN is extended into a dual-model structure and then trained and regularized using self-supervised learning between different intermediate acoustic representations; Then, we enhance the dual-model by combining self-supervised loss and supervised loss in a time-dependent manner, thus enhancing the model’s overall generalization capabilities. Furthermore, to better utilize the complementary information in the dual-model’s outputs, we explore various methods for similarity computation and score fusion. Our experiments, conducted on the publicly available VoxCeleb2 and VoxMovies datasets, have demonstrated that our proposed dual-model regularization and fusion methods outperformed the strong baseline by a relative 9.07%–11.6% EER reduction across various in-domain and cross-domain evaluation sets. Importantly, our approach exhibits effectiveness in both supervised and unsupervised scenarios for low-resource cross-domain speaker verification tasks.

Understanding Cloud Systems’ Structure and Organization Using a Machine’s Self-Learning Approach

Abstract In this study, we introduce a self-supervised deep neural network approach to classify satellite images into independent classes of cloud systems. The driving question of the work is to understand whether our algorithm can capture cloud variability and identify distinct cloud regimes. Ultimately, we want to achieve generalization such that the algorithm can be applied to unseen data and thus help automatically extract relevant information important to atmospheric science and renewable energy applications from the ever-increasing satellite data stream. We use cloud optical depth (COD) retrieved from postprocessed high-resolution Meteosat Second Generation (MSG) satellite data as input for the network. The network’s architecture is based on the DeepCluster, version 2, and consists of a convolutional neural network and a multilayer perceptron, followed by a k-means algorithm. We explore the network’s training capabilities by analyzing the centroids and feature vectors found from progressive minimization of the cross-entropy loss function. By making use of additional MSG retrieval products based on multichannel information, we derive the optimum number of classes to determine independent cloud regimes. We test the network capabilities on COD data from 2013 and find that the trained neural network gives insights into the cloud systems’ persistence and transition probability. The generalization on the 2015 data shows good skills of our algorithm with unseen data, but results depend on the spatial scale of cloud systems. Significance Statement This study uses a self-supervised deep neural network to identify distinct cloud systems from cloud optical depth satellite images over central Europe. Satellite-retrieved products support the physical interpretation of the identified cloud classes and help optimize the number of identified classes. The trained neural network gives insights into cloud systems’ persistence and transition probability. The generalization capacity of the deep neural network with unseen data is promising but depends on the spatial scale of cloud systems.

Learning Program Synthesis for Integer Sequences from Scratch

We present a self-learning approach for synthesizing programs from integer sequences. Our method relies on a tree search guided by a learned policy. Our system is tested on the On-Line Encyclopedia of Integer Sequences. There, it discovers, on its own, solutions for 27987 sequences starting from basic operators and without human-written training examples.

Fingerprinting Network Sessions for the Discovery of Cyber Threats

A
 rtificial intelligence (AI) assisted cyber-attacks, within the network cybersecurity domain, have evolved to be more successful at every phase of the cyber threat lifecycle. This involves, amongst other tasks, reconnaissance, weaponisation, delivery, exploitation, installation, command & control, and actions. The result has been AI-enhanced attacks, such as DeepLocker, self-learning malware and MalGan, which are highly targeted and undetectable, and automatically exploit vulnerabilities in existing cyber defence systems
 . Countermeasures would require significant improvements in the efficacy of existing cyber defence systems to enable the discovery and detection of AI-enhanced attacks in networks in general. The challenge is that rule-and-anomaly-based intrusion detection approaches would need to be evolved into a dynamic self-learning approach before being able to discover “undetectable” network threats. The problem is that, when considering current state-of-the-art network cybersecurity countermeasures, this has not yet been achieved. One of the key challenges in achieving this is the inability to extract meaningful information from network packets. The novel solution proposed in this paper is to fingerprint network sessions. Each fingerprint is represented by a two-dimensional matrix that can be visualised, comprising a unique session key, the protocol discourse and the transmitted data. This is achieved by extracting information, summarising network session key events, encoding the received data, and merging it with existing fingerprints. The unique key and transmitted data are encoded using a Hilbert curve, while the protocol discourse is encoded into a tornado diagram. The resulting visualised network session fingerprints reveal hidden patterns that are ideal for subsequent pattern recognition, reinforcement learning (RL) or support vector machines (SVM) training to discover AI-enhanced cyber threats as they evolve.

Low-Resource Neural Machine Translation Improvement Using Source-Side Monolingual Data

Despite the many proposals to solve the neural machine translation (NMT) problem of low-resource languages, it continues to be difficult. The issue becomes even more complicated when few resources cover only a single domain. In this paper, we discuss the applicability of a source-side monolingual dataset of low-resource languages to improve the NMT system for such languages. In our experiments, we used Wolaytta–English translation as a low-resource language. We discuss the use of self-learning and fine-tuning approaches to improve the NMT system for Wolaytta–English translation using both authentic and synthetic datasets. The self-learning approach showed +2.7 and +2.4 BLEU score improvements for Wolaytta–English and English–Wolaytta translations, respectively, over the best-performing baseline model. Further fine-tuning the best-performing self-learning model showed +1.2 and +0.6 BLEU score improvements for Wolaytta–English and English–Wolaytta translations, respectively. We reflect on our contributions and plan for the future of this difficult field of study.

Anomaly-based threat detection: Behavioural fingerprinting versus self-learning AI

When a malicious actor has access to a digital estate, they control compromised devices and user accounts to achieve their objectives. Given that an attacker’s objectives are often at odds with devices’ normal patterns of life, identifying deviations from these patterns can be used to detect an ongoing attack. This paper outlines and compares two approaches to anomaly-based threat detection: behavioural fingerprinting and self-learning artificial intelligence (AI). It argues that the self-learning approach is significantly superior in several important ways due to the fact it provides a more complex and accurate understanding of what is normal. The paper explains the motivation behind anomaly-based threat hunting, describes the fingerprinting approach and the self-learning approach to anomaly detection, and details real-world examples that demonstrate the advantages of the self-learning approach.

A self-learning approach for proactive resource and service provisioning in fog environment

A self-learning approach for proactive resource and service provisioning in fog environment

Quantum OPTICS and deep self-learning on swarm intelligence algorithms for Covid-19 emergency transportation.

In this paper, the quantum technology is exploited to empower the OPTICS unsupervised learning algorithm, which is a density-based clustering algorithm with numerous applications in the real world. We design an algorithm called Quantum Ordering Points To Identify the Clustering Structure (QOPTICS) and demonstrate that its computational complexity outperforms that of its classical counterpart. On the other hand, we propose a Deep self-learning approach for modeling the improvement of two Swarm Intelligence Algorithms, namely Artificial Orca Algorithm (AOA) and Elephant Herding Optimization (EHO) in order to improve their effectiveness. The deep self-learning approach is based on two well-known dynamic mutation operators, namely Cauchy mutation operator and Gaussian mutation operator. And in order to improve the efficiency of these algorithms, they are hybridized with QOPTICS and executed on just one cluster it yields. This way, both effectiveness and efficiency are handled. To evaluate the proposed approaches, an intelligent application is developed to manage the dispatching of emergency vehicles in a large geographic region and in the context of Covid-19 crisis in order to avoid an important loss in human lives. A theoretical model is designed to describe the issue mathematically. Extensive experiments are then performed to validate the mathematical model and evaluate the performance of the proposed deep self-learning algorithms. Comparison with a state-of-the-art technique shows a significant positive impact of hybridizing Quantum Machine Learning (QML) with Deep Self Learning (DSL) on solving the Covid-19 EMS transportation.

When firms go international: deliberate or random?

PurposeThe net outward investment position (NOIP) indicator is insufficient for the purposes of understanding firms’ internationalization decision-making behaviour. The indicator does not allow for the withdrawal of insights into the structure of an economy and is a weak predictor of the degree of foreign direct investment. The purpose of this paper is to argue that a typology of firms aggregated according to intrinsic characteristics of those firms is a better predictor of the degree of internationalization of an economy than the NOIP.Design/methodology/approachThis paper uses a database of 2,133 firms located in Portugal with international operations, made available by AICEP, a government agency. This paper uses multiple correspondence and cluster analyses to build a typology of firms and obtains evidence of common characteristics of the constituent groups.FindingsThis paper identifies a typology of firms characterized by five types differentiated by firm age, length of internationalization process, sector of economic activity, legal status and psychological/cultural proximity. These variables suggest an evolutionary, iterative, self-learning approach to internationalization, which can be better explained by the combined use of the investment development path (IDP) framework, the Uppsala Evolutionary School and Vernon’s product life cycle theory. Additionally, this paper finds that the most striking differences between developed and developing host countries are in terms of the economic sector, legal status of the firm and belonging (or not) to an economic group.Originality/valueThis paper establishes a link between the IDP framework, the Uppsala Evolutionary School and Vernon’s product life cycle theory, using a categorization of firms made according to selected characteristics to understand the internationalization of firms.

A novel self-learning approach to overcome incompatibility on TripAdvisor reviews

Among social media networks, TripAdvisor acts as the main role because everyone is eager to share and review their thoughts on their travel experiences in different destinations. Sentiment analysis is a method that can be used to analyze people's behaviors and opinions on public and social media platforms. In this study, hotel reviews are extracted from the five most attractive Sri Lankan cities, and user-written reviews are compared over user bubble ratings, which define overall travelers' experiences as a numerical scale that ranks from 1 to 5. We find that the compatibility between user-written reviews and bubble ratings has a low correlation because bubble ratings may not represent the overall idea of users' genuine opinions expressed in their reviews. To address this problem, a two-phase approach is proposed: (1) the ensemble method to improve the performance of lexicon-based outputs and identify the correctly matching user review and bubble rating; (2) the self-learning approach to finding the sentiment of a review that does not properly label by the user. The performance is studied by considering reviews incompatible with the sentiment of user bubble rating and the sentiment generated by the proposed model. For example, regardless of bigram “not good”, the average percentages of the word “good” for each negatively identified review from the proposed model and bubble rating are 25.63% and 38.85%, respectively. Thereby, it is apparent that the negative sentiments derived by bubble rating have significantly more positive words compared to the proposed model.

Implementation of Problem Based Learning to Solve Real Life Problems

Problem based learning (PBL) has been used for many years and it is one of the effective educational methodologies applicable for distinguished engineering disciplines. It is a widely used student-centric approach through which students get engaged to solve real world problems by research, continuous study, and regular practices along with knowledge utilization during implementation. This study explores the PBL with an advanced computer engineering course of pre-final year students. This study centers around the implementation of a system to solve one of the real-life problems and distinguishes a portion of the difficulties that lie ahead for PBL. Here, to examine the impact of PBL approach for the specified subject, the result of B.Tech. (CE and IT branches) sixth semester students (A.Y. 2019-20 and 2020-21) are considered with various known parameters. Students' feedback is taken at the end of course study. Problem title, hook of the problem, background or history of the problem, clarifying/providing constraints, list of deliverables, strategy to solve the problem, assessment criteria/rubric etc. are provided well in advance to the students to be comprehensible for the problem. Learning sequence sessions are also arranged on a regular basis during problem solving by the facilitator to fill out the gap. As a result, the PBL approach is found to be intended more strongly for engineering students than for other students; in contrast, with traditional MCQ based examination students mostly focus on mug up of concepts rather than implementation and critical thinking. Engineering students are able to work smartly along with hard work, and as a result students have developed their technical skills, collaborative skills, self-learning approach with advanced computer engineering courses. Assessment pattern has changed into advancement with the proposed PBL approach over a traditional MCQ based approach. Continuous assessment is performed on covering all segments of the exam. Results found with traditional approach are contrasted with results found with PBL approach and tracked down that a more prominent number of students had secured more than over average score (>=60%) with the proposed PBL approach and statistically proven for the same. This correlation shows that students are more noticeable and curious about the PBL approach as opposed to robbing up hypothetical ideas and choosing the appropriate answers with a traditional MCQ based approach. In Addition, student’s feedback is observed to be very satisfactory. It has been seen that through PBL, students' retaining level has extremely increased and their learning abilities and understanding level too has improved altogether. PBL plays out a fundamental role for achieving the student's involvement and commitment towardslearning. Logical, critical thinking, teamwork, technical etc. skills for engineering students can be developed effectively with the PBL approach. Keywords— Problem-based learning, Continuous assessment, PBL, self-regulated learning, self-directed learning

Creation and implementation of a set of game strategies based on training neural networks with reinforcement learning

The study explores the problems of reinforcement learning and finding non-obvious play strategies using reinforcement learning. Two approaches to agent training (blind and pattern-based) are considered and implemented. The advantage of the self-learning approach with reinforcement using patterns as applied to a specific game (tic-tac-toe five in a row) is shown. Recorded and analyzed the use of unusual strategies by an agent using a pattern-based approach.

An ontology self-learning approach for CNC machine capability information integration and representation in cloud manufacturing

Manufacturing activities in cloud-based environments strongly rely on the online integration and description of the capability of machining resources. In the authors’ previous work, STEP-NC schema was applied to construct an ontology model to support the integration and reasoning of machine tool information and capability. For the maintenance and update of the model, in this paper, a self-learning method is proposed to explore correlations from STEP-NC process planning documents to obtain machining knowledge to improve the comprehensiveness of the model. In this method, a Map/Reduce-based Apriori algorithm is developed incombination with the built ontological model. First, a dataset is extracted from the document according to the importance analysis results of the model. Then, a mining procedure that combinesApriori algorithm and Map/Reduce framework is developed. Finally, two representation modes are adopted to embed the mined results into the model. According to the outcomes of a preliminary experiment with standard STEP-NC documents, this method effectively enables the ontology learning mechanism from the aspects of time consumption and mined associations, which improves the suitability of the enriched ontology model to handle information integration and industrial applications.

The degree to which kindergarten curricula in the Kingdom of Saudi Arabia include some concepts of digital citizenship in light of the requirements of the digital age: درجة تضمين مناهج رياض الأطفال في المملكة العربية السعودية لبعض مفاهيم المواطنة الرقمية في ضوء متطلبات العصر الرقمي

The study aimed to determine the degree to which kindergarten curricula in the Kingdom of Saudi Arabia include some concepts of digital citizenship in light of the digital age requirements. To achieve this goal, the researcher used the descriptive and analytical approach to the study sample represented in 8 books from the kindergarten curricula in Saudi Arabia. A list of digital citizenship concepts, needed in light of the digital age, was built based on a review of educational literature and previous studies related to the topic. The list included a total of 30 indicators categorised into 6 areas: digital behavior, digital communication, digital knowledge, digital health and safety, digital law, and digital security. The researcher analyzed the content based on the list of standards and concluded that the inclusion of digital citizenship concepts was very limited or almost absent in the studied sample. Inclusion was highest for digital knowledge and digital health and safety at 50% for each area, but with only two repetitions. Other areas, however, were not included at all (0%). The most important recommendation of the study is to highlight the need for specialists in kindergarten curricula to familiarize themselves with the current global standards of curricula, and to include the most important and relevant concepts of the digital age, including digital citizenship. It is also necessary to review the weaknesses and the lack of kindergarten books on the self-learning approach in the curriculum developed in Saudi Arabia, as well as to strengthen the available advantages.

Pedometer-free Geomagnetic Fingerprinting with Casual Walking Speed

The geomagnetic field has been wildly advocated as an effective signal for fingerprint-based indoor localization due to its omnipresence and local distinctive features. Prior survey-based approaches to collect magnetic fingerprints often required surveyors to walk at constant speeds or rely on a meticulously calibrated pedometer (step counter) or manual training. This is inconvenient, error-prone, and not highly deployable in practice. To overcome that, we propose Maficon, a novel and efficient pedometer-free approach for geo ma gnetic fi ngerprint database con struction. In Maficon, a surveyor simply walks at casual (arbitrary) speed along the survey path to collect geomagnetic signals. By correlating the features of geomagnetic signals and accelerometer readings (user motions), Maficon adopts a self-learning approach and formulates a quadratic programming to accurately estimate the walking speed in each signal segment and label these segments with their physical locations. To the best of our knowledge, Maficon is the first piece of work on pedometer-free magnetic fingerprinting with casual walking speed. Extensive experiments show that Maficon significantly reduces walking speed estimation error (by more than 20%) and hence fingerprint error (by 35% in general) as compared with traditional and state-of-the-art schemes.