Categorization Training Research Articles

Research on the Classification of Sun-Dried Wild Ginseng Based on an Improved ResNeXt50 Model

Ginseng is a common medicinal herb with high value due to its unique medicinal properties. Traditional methods for classifying ginseng rely heavily on manual judgment, which is time-consuming and subjective. In contrast, deep learning methods can objectively learn the features of ginseng, saving both labor and time. This experiment proposes a ginseng-grade classification model based on an improved ResNeXt50 model. First, each convolutional layer in the Bottleneck structure is replaced with the corresponding Ghost module, reducing the model’s computational complexity and parameter count without compromising performance. Second, the SE attention mechanism is added to the model, allowing it to capture feature information more accurately and precisely. Next, the ELU activation function replaces the original ReLU activation function. Then, the dataset is augmented and divided into four categories for model training. A model suitable for ginseng grade classification was obtained through experimentation. Compared with classic convolutional neural network models ResNet50, AlexNet, iResNet, and EfficientNet_v2_s, the accuracy improved by 10.22%, 5.92%, 4.63%, and 3.4%, respectively. The proposed model achieved the best results, with a validation accuracy of up to 93.14% and a loss value as low as 0.105. Experiments have shown that this method is effective in recognition and can be used for ginseng grade classification research.

Content and design of respectful maternity care training packages for health workers in sub-Saharan Africa: Scoping review.

Training health workers might facilitate respectful maternity care (RMC); however, the content and design of RMC training remain unclear. To explore the content and design of RMC training packages for health workers in sub-Saharan Africa. MEDLINE, EMBASE, CINAHL Complete, Web of Science Core Collections, SCOPUS, and grey literature sources (including websites of RMC-focused key organizations and Ministries of Health) were searched for journal papers, reports, and training guides from January 2006 up to August 2022. There were no restrictions on study designs, language, or health-worker cadre. Two reviewers independently screened results. Key data, including training content and methods used, were extracted and summarized. Thirty-two citations from 26 studies/programs were identified (24 journal papers, 5 manuals/guides, 2 reports and 1 PhD thesis), with 27 citations from 22 studies informing the review findings. About half of all conducted studies were from East Africa. The most common topics in RMC trainings were communication, privacy and confidentiality, and human resources. Most trainings were multicomponent and appear to be largely in-service training. Health workers providing direct care to women, compared with non-clinical staff such as receptionists and cleaners, were the only recipients of training in most studies (81.8%). Two broad categories of training methods/tools were identified: workshop-based and action-based. Over 90% of the studies assessed impact of the training, with a majority focused on impacts on maternal health and care; however, half of the latter studies did not appear to have feedback mechanisms in place for implementing change. The content and design of RMC training in sub-Saharan Africa are multifaceted, suggesting the complexity of implementing/promoting RMC. Some progress has been made; however, missed opportunities in training remain with respect to study populations, training topics, cadres, and feedback mechanisms.

Perceived Self-Efficacy among Special Education Students Enrolled in a Practical Training Program During the COVID-19 Pandemic in Jordan

The current study aimed to investigate the level of perceived self-efficacy among special education students enrolled in the practical training program during the COVID-19 pandemic in Jordan. The study sample consisted of (150) male and female students from various public and private Jordanian universities, who were purposefully chosen, and enrolled in practical training during the semesters that continued under the pandemic. To achieve the objectives of the study, the descriptive approach and the perceived self-efficacy scale prepared by the researcher were used. The results of the study showed that the level of perceived self-efficacy among students of practical training in the specialization of special education came at a high level; the results also indicated that the level of perceived self-efficacy concerning the employment of modern technology ranked first and at a high degree, while the self-efficacy related to teaching skills and behavior modification, as well as management, communication, and social responsibility came to a medium degree. The results also showed that there are statistically significant differences between the arithmetic averages of the level of perceived self-efficacy among students of field training in the specialization of special education who are registered in Jordanian universities according to the type of university (public/private) on the total degree and all dimensions except for the dimension (management, communication, and social responsibility) and in favor of private universities. The results also showed that there were statistically significant differences between the arithmetic averages of the level of self-efficacy among students of special education according to the type of training and its duration on the total degree and dimensions and in favor of the two training categories (full and continuous semester" and "full and intermittent semester"). In contrast, the results did not show any statistically significant differences between the arithmetic averages of the level of perceived self-efficacy among special education students enrolled in the practical training program during the COVID-19 pandemic in Jordan, according to the student's gender.

Association between training methods and owner-reported problems behaviours in dogs enrolled in the ‘Generation Pup’ longitudinal study in the UK

This study aimed to explore factors associated with owners reporting problem behaviour in dogs. Owner-completed surveys collected as part of a longitudinal dog health/behaviour study were analysed. Owner-selected training methods were classified into four training categories on a scale from ‘Rewards only’ to ‘Aversive only’ that reflected the proportion of aversive- and rewards-based methods selected. ‘Owner-reported behaviour at 12 months’ (problem reported/ not) was modelled with logistic regression. Predictive variables included: dog/owner demographics; owner-reported behaviour at three timepoints (12/16 weeks, 6 and 9 months), the training category and change in the training category (to a more aversive/less aversive method or maintenance) at three timepoints (12/16 weeks, 9 and 12 months). The model was repeated for a subset of dogs whose owners completed a personality questionnaire (International Personality Item Pool-50). The second model included the predictive variables listed for the first model and owners’ personality scores. Dogs trained with only aversive methods were excluded (n <10). In the first model (X2=312, R2=0.33, p=<0.001, n=1130), owner-reported problem behaviour at 12 months was associated with the increased odds of: owner-reported problem behaviour at 12/16 weeks (OR=1.88, CI%1.38-2.55, p<0.001), 6 (OR= 2.19, CI%1.62-2.97, p<0.001) and 9 months (OR=6.51, CI%4.83-8.83, p<0.001) and changing to a more aversive training category between 9-12 months (compared to changing to a less aversive/ maintenance of the same category; OR=1.66, CI%1.06-2.63, p=0.029). In the second model (X2=178.4, R2=0.35, p=<0.001, n=590) owner-reported problem behaviour at 12 months was associated with increased odds of owner-reported problem behaviour at all previous timepoints (OR=1.83, CI%1.20-2.79, p<0.005; OR= 2.29, CI%1.1.49-3.51, p<0.001; OR=6.98, CI%4.61-10.70, p<0.001 for 12/16 weeks, 6 and 9 months, respectively). In this model, higher odds of owner-reported problem behaviour at 12 months were additionally associated with being a first-time owner (OR=1.93; CI%1.12-3.08, p=0.006), not seeing the dog’s mother on acquisition (OR=2.03, CI%1.12-3.68, p=0.02) and owner Openness personality trait score (OR=1.05, CI%1.01-1.09, p=0.02). The perception of deteriorating dog behaviour during the first year of a dog’s life may motivate owners to apply training with aversive components, warranting further education into the importance of consistent rewards-based training.

Categories of training to improve empathy: A systematic review and meta-analysis.

Due to the vital role of empathy in promoting prosocial behaviors and nurturing social bonds, there is a growing interest in cultivating empathy. Yet, the effectiveness of existing training methods on empathy, especially on different dimensions of empathy (i.e., affective, cognitive, motivational, and behavioral empathy), varies tremendously, and the underlying causes for this heterogeneity remain insufficiently explored. To address this issue, we categorized various training methods into three distinct approaches based on the premise that empathy can be influenced by factors associated with the subject, the object, and their relationship. Respectively, these are Subject-Oriented, Object-Oriented, and Socially Oriented approaches. To examine the effects of training and sustainability of these approaches on different dimensions of empathy, we conducted a meta-analysis encompassing 110 eligible studies with 32, 44, 39, 39, and 91 samples for affective, cognitive, motivational, behavioral, and composite empathy, respectively. Results showed that trainings produced small and unsustainable effects on affective empathy, moderate and unsustainable effects on cognitive empathy, small and sustainable effects on motivational empathy, and moderate and sustainable effects on behavioral empathy. The effects of training on composite empathy were robust but decreased over time. Among the three training approaches, Socially Oriented approaches were the most effective in improving all dimensions of empathy. Subject-Oriented and Object-Oriented approaches were effective only in improving cognitive and composite empathy. Altogether, our study offers practical guidance for selecting appropriate training approaches and proposes theoretical principles for developing optimal training strategies in both basic research and clinical applications. (PsycInfo Database Record (c) 2024 APA, all rights reserved).

BRAVE: A cascaded generative model with sample attention for robust few shot image classification

Few-shot learning (FSL) confronts notable challenges due to the disparity between training and testing categories, leading to channel bias in neural networks and hindering accurate feature discernment. To address this, we introduce Biased-Reduction Attentive Network (BRAVE), an innovative model that incorporates a refined Vector Quantized Variational Autoencoder (VQ-VAE) backbone, enhanced with our Diverse Quantization (DQ) Module, for unbiased, fine-grained feature creation. Alongside, our Sample Attention (SA) Module is utilized for extracting discriminative features from these unbiased, fine-grained features. The DQ Module in BRAVE strategically integrates prior distribution regularization and stochastic masking with Gumbel sampling for balanced and diverse codebook engagement, while the SA Module leverages inter-sample dynamics for identifying critical features. This synergy effectively counters channel bias and boosts classification accuracy in FSL setups, surpassing current leading methods. Our approach represents a practical balance between preserving detailed features through the decoder and ensuring classification effectiveness, marking a significant advance in FSL. BRAVE’s implementation is accessible for community use and further exploration. Code and models available at https://github.com/ApocalypsezZ/BRAVE.

Manipulating and measuring variation in deep neural network (DNN) representations of objects

We explore how DNNs can be used to develop a computational understanding of individual differences in high-level visual cognition given their ability to generate rich meaningful object representations informed by their architecture, experience, and training protocols. As a first step to quantifying individual differences in DNN representations, we systematically explored the robustness of a variety of representational similarity measures: Representational Similarity Analysis (RSA), Centered Kernel Alignment (CKA), and Projection-Weighted Canonical Correlation Analysis (PWCCA), with an eye to how these measures are used in cognitive science, cognitive neuroscience, and vision science. To manipulate object representations, we next created a large set of models varying in random initial weights and random training image order, training image frequencies, training category frequencies, and model size and architecture and measured the representational variation caused by each manipulation. We examined both small (All-CNN-C) and commonly-used large (VGG and ResNet) DNN architectures. To provide a comparison for the magnitude of representational differences, we established a baseline based on the representational variation caused by image-augmentation techniques used to train those DNNs. We found that variation in model randomization and model size never exceeded baseline. By contrast, differences in training image frequency and training category frequencies caused representational variation that exceeded baseline, with training category frequency manipulations exceeding baseline earlier in the networks. These findings provide insights into the magnitude of representational variations that can be expected with a range of manipulations and provide a springboard for further exploration of systematic model variations aimed at modeling individual differences in high-level visual cognition.

Combined deep-learning optimization predictive models for determining carbon dioxide solubility in ionic liquids

This study explores the development of predictive models for carbon dioxide (CO2) solubility in ionic liquids based on a compiled dataset of 10,116 experimentally measured data points involving four input variables: pressure (P), temperature (T), cation type, and anion type. The deep-learning (DL) predictive models evaluated are standalone and hybrid versions of convolutional neural network (CNN) and long short-term memory (LSTM) algorithms with cuckoo optimization algorithm (COA) and gradient-based optimization (GBO). The laboratory-measured data was separated into training and test categories, and each category was normalized separately to improve the performance of the deep learning algorithms. The Mahalanobis distance-based quantile method was utilized to identify any outliers in the training data. Once identified, the outlier data points were eliminated from the training dataset. The control parameters of the deep learning algorithms were optimized using COA to enhance their efficiency, and the algorithms were hybridized with optimization algorithms to further improve their performance. The resulting models were analyzed to assess their accuracy, degree of overfitting, and the importance of input features. The study found that using 80% of the data for training and 20% for testing results in more accurate and generalizable models. Using the outlier detection method on the training data led to 307 data points being eliminated as outliers. Developing CO2-solubility predictive model showed that, the CNNCOA model had the lowest RMSE and highest R2 among the developed models, indicating high generalizability for data unseen by the trained model. The analysis revealed that using optimization algorithms increased the CO2-solubility prediction performance of DL algorithms and reduced overfitting. T and cation type were the most and least important input features, respectively. Simultaneous changes in cation and anion type on CO2-solubility predictions displayed no systematic pattern. For increases in T, CO2 solubility typically decreased, whereas for increases in P CO2 solubility always increased but at variable rates. The results of this study can be used to develop accurate and generalizable CO2-solubility predictive models for various applications.

Saudi teachers’ perceptions on pedagogical affordances of digital applications in teaching students with learning disabilities

ABSTRACT Creating a supporting learning environment for students with learning disabilities has been the concern of most teachers. However, digital applications provide numerous pedagogical affordances for teaching students with learning disabilities. This quantitative study examined teachers’ perspectives on the pedagogical affordances of digital applications and challenges when teaching students with learning disabilities in primary schools in Riyadh, Saudi Arabia. Data were collected from 280 teachers of students with learning disabilities through a questionnaire using a random strategy. The results showed that despite teachers’ awareness of the pedagogical affordances of digital applications in developing the academic skills of students with learning disabilities, the use of digital applications was intermediate. Teachers identified a lack of digital tools, time pressure and teachers’ workload as hindering the use of digital applications in their teaching. Findings also reveal statistically significant differences in perceptions for males and females, holding a master’s degree or above qualification, experience category (1–5 years), and training category (1–5 training sessions). These findings highlight the need for ongoing professional development and support that focuses more on real-classroom practices and resources, and how teachers can use digital tools to meet the various learning needs of students with learning disabilities.

A Nova Lei de Licitações e Contratos: as perspectivas dos Tribunais de Contas do Nordeste

Purchases in public administration follow legal procedures, and managers need to be aware of the financial and budgetary aspects and the requirements of the New Law on Bidding and Contracts. Law nº 14.133/2021 is currently being adapted, and the Courts of Auditors are responsible for overseeing the use of public funds by governments. In this context, this study aims to describe the main topics discussed in the Court of Auditors of the states of the Brazilian Northeast regarding the New Law on Bidding and Contracts. The documentary, descriptive, predominantly qualitative research, with data from the period 2021 to 2023, was carried out based on official materials obtained from the transparency website of the courts of audit. For the proposed analysis, in light of the text of the new law, the courts' perspectives were grouped into seven central and related themes (categories): governance, planning, training, innovative Information and Communication Technology (ICT) solutions, risk plan, sustainability and compliance. Based on the highest average occurrences of the categories surveyed, the results indicated that the Courts of Audit of Rio Grande do Norte and Bahia had the best performances in perspectives on the central points of the new law on bidding and contracts. In addition, the categories of planning, training and risk plan were the most discussed by the courts in the Northeast of Brazil, while the topic of sustainability, despite its relevance, was little addressed by the courts in the period analyzed.

Exploitative training versus explorative training: the effect of strategic human capital investment on organizational performance

PurposeNotwithstanding its significance as a form of strategic human capital investment, employee training has not yielded consistent conclusions among scholars regarding its impact on organizational performance. Some studies deem it effective, while others regard it as ineffective. We contend that distinct types of training impact various facets of firm performance.Design/methodology/approachIn this study, we categorize employee training as either exploitative or explorative. Specifically, we examine their impact on two aspects of organizational performance: short-term performance and long-term competence, using a quasi-experimental setting and a difference-in-differences (DID) method.FindingsWe find that exploitative training is more effective in improving firms’ short-term performance (e.g. firms’ sales revenue), while explorative training is more effective in enhancing firms’ long-term competence (e.g. firms’ innovation output).Originality/valueThe findings of this study illuminate concrete benefits of training for practitioners, suggesting that firms can strategically select employee training category to maximize returns on their investment in strategic human capital based on their strategic orientations.

Discrimination sensitivity of visual shapes sharpens in autistic adults but only after explicit category learning

BackgroundCategorization and its influence on perceptual discrimination are essential processes to organize information efficiently. Individuals with Autism Spectrum Condition (ASC) are suggested to display enhanced discrimination on the one hand, but also to experience difficulties with generalization and ignoring irrelevant differences on the other, which underlie categorization. Studies on categorization and discrimination in ASC have mainly focused on one process at a time, however, and typically only used either behavioral or neural measures in isolation. Here, we aim to investigate the interrelationships between these perceptual processes using novel stimuli sampled from a well-controlled artificial stimulus space. In addition, we complement standard behavioral psychophysical tasks with frequency-tagging EEG (FT-EEG) to obtain a direct, non-task related neural index of discrimination and categorization.MethodsThe study was completed by 38 adults with ASC and 38 matched neurotypical (NT) individuals. First, we assessed baseline discrimination sensitivity by administering FT-EEG measures and a complementary behavioral task. Second, participants were trained to categorize the stimuli into two groups. Finally, participants again completed the neural and behavioral discrimination sensitivity measures.ResultsBefore training, NT participants immediately revealed a categorical tuning of discrimination, unlike ASC participants who showed largely similar discrimination sensitivity across the stimuli. During training, both autistic and non-autistic participants were able to categorize the stimuli into two groups. However, in the initial training phase, ASC participants were less accurate and showed more variability, as compared to their non-autistic peers. After training, ASC participants showed significantly enhanced neural and behavioral discrimination sensitivity across the category boundary. Behavioral indices of a reduced categorical processing and perception were related to the presence of more severe autistic traits. Bayesian analyses confirmed overall results.LimitationsData-collection occurred during the COVID-19 pandemic.ConclusionsOur behavioral and neural findings indicate that adults with and without ASC are able to categorize highly similar stimuli. However, while categorical tuning of discrimination sensitivity was spontaneously present in the NT group, it only emerged in the autistic group after explicit categorization training. Additionally, during training, adults with autism were slower at category learning. Finally, this multi-level approach sheds light on the mechanisms underlying sensory and information processing issues in ASC.

Early Detection of Phishing Sites with Enhanced Neural Network Models

Phishing is a digital crime committed with the aim of obtaining personal data by creating a link or website that resembles the original. This form of cyber attack is caused by a notification in a text message, email, or phone call. A common anti-phishing countermeasure technique is to perform early detection of potentially phishing sites, primarily according to the source code features, which are required to traverse web page content, as well as third parties that slow down the process of clarifying phishing URLs. Although the latest technology has long been used in phishing early detection, there is still a need for manual feature engineering that is important and reliable enough to detect emerging phishing offenses. One of these involves training a neural network (NN) using a dataset of known phishing URLs and legitimate URLs. The research was conducted using 200 data, Data were separated into training and testing categories. Training was done using 100 and 120 data. Training results on 100 data and 160 data had lower iterations and errors on the tanh activation function compared to the logistic activation function. The number of iterations that occur in logistic activation is as many as 400 iterations, while when using the tanh activation function only 175 iterations are needed.

USING THE «METAPHORIZATION» TECHNOLOGY IN MEDIA LITERACY TRAINING

The article is devoted to the analysis of interactive and reflective technology and its use in the process of conducting training on media literacy and critical thinking. The main approaches to the definition of the concept of «metaphorization» in the format of training and scientific categories of thinking are considered. Metaphorization in training is the use of metaphors as a powerful learning tool that promotes deeper understanding and assimilation of the material. This article discusses the role of metaphors in the training process, identifies their main functions and mechanisms of influence on participants. The authors emphasize the ability of metaphors to promote creativity, deeper comprehension of the training material, and more effective communication between the trainer and participants. Considerable attention is paid to the analysis of specific examples of metaphors that have been successfully used in trainings, as well as to the methods of creating them and adapting them to different learning objectives and audiences. The influence of cultural factors on the perception of metaphors is also considered, which requires trainers to have a deeper understanding of the cultural context of their participants. The peculiarities of approaches to the definition of the concept of metaphorization in education are analyzed. It is argued that since each person is a carrier of a special semantic space peculiar only to him or her, many elements of which are metaphorical, it is possible to distinguish a set of individual metaphors in relation to the problem or phenomena under study. On this basis, the author's original technology «metaphorization in training» was created, which has been repeatedly used by the author in working with media trainers in the process of conducting media literacy trainings. The article provides examples of creative projects of the training participants and the author's comments on the use of the metaphorization method for various aspects of media literacy. The author's generalized conclusions on each group work give a systematic view of both the process of media literacy development in our country during Ukraine's liberation war against russia and reveal the main opportunities for using this technology in the process of conducting trainings. The conclusions of the article emphasize the importance of integrating metaphors into the structure of training programs as an effective means of improving the quality of training and motivating participants.

A study on training needs of chilli growers in Andhra Pradesh

The present study assesses the training needs of chilli growers of Andhra Pradesh. An ex-post facto research design was adopted for the study. Two districts, Guntur and Prakasam, were purposefully selected for the study, as they cover the highest area of chilli cultivation. Two mandals were selected from each district purposively based on the highest area and from each mandal, 5 villages were selected. Thirteen respondents from each village were selected randomly. Thus, a total sample size of 260 was selected for the study. After a thorough review of the literature, focal group discussions with chilli growers and consultation with experts, 7 different categories of training areas were identified. The data was collected using the personal interview method, using a pre-tested, well-prepared interview schedule and ranked based on mean score. The study revealed that among all the different categories of training needs, the respondents perceived crop protection and crop production as the most important training needs, with mean scores of 2.324 and 2.159 and were ranked first and second. Training needs pertaining to marketing, harvesting and post-harvest handling, processing and value-addition, nursery management and mechanization occupied the next 3rd to 7th ranks respectively. Hence, the extension personnel, scientists and policymakers shall keep the results of this study in view regarding what content needs to be focused on while preparing training programs for chilli farmers.

A study on problems and suggestions of chilli growers in Andhra Pradesh, India

The present study assesses the training needs of chilli growers of Andhra Pradesh. An ex-post facto research design was adopted for the study. Two districts, Guntur and Prakasam, were purposefully selected for the study, as they cover the highest area of chilli cultivation. Two mandals were selected from each district purposively based on the highest area and from each mandal, 5 villages were selected. Thirteen respondents from each village were selected randomly. Thus, a total sample size of 260 was selected for the study. After a thorough review of the literature, focal group discussions with chilli growers and consultation with experts, 7 different categories of training areas were identified. The data was collected using the personal interview method, using a pre-tested, well-prepared interview schedule and ranked based on mean score. The study revealed that among all the different categories of training needs, the respondents perceived crop protection and crop production as the most important training needs, with mean scores of 2.324 and 2.159 and were ranked first and second. Training needs pertaining to marketing, harvesting and post-harvest handling, processing and value-addition, nursery management and mechanization occupied the next 3rd to 7th ranks respectively. Hence, the extension personnel, scientists and policymakers shall keep the results of this study in view regarding what content needs to be focused on while preparing training programs for chilli farmers.

What and how to train for strategic crisis management: A systematic literature review

AbstractThis article provides a systematic literature review of existing strategic crisis management training research. The review explores practices on which strategic crisis management skills to prioritize and how to train these skills to be prepared when a crisis strikes. Our research question was: How is strategic crisis management trained, and is there a best practice? In all, we identified 538 articles from 3 databases and read the abstracts. From this, we selected 41 articles based on the defined inclusion criteria and read them as a whole. In the end, eight articles were a part of this analysis. A thematic analysis based on the categorization of repetitions of concepts was used to analyse the data. The analysis resulted in two main categories: what to train, which focuses on the outcome and skills, and how to train, which focuses on the process and pedagogy. The outcome (what to train) categories were (1) related to working together, (2) related to understanding the situation, (3) related to making adequate decisions in complex contexts, and (4) related to practicalities. The process (how to train) categories were (1) training methods and (2) learning theories. A model considering three main strategic crisis management training approaches, outcome‐focused training, process‐focused training, and learning‐focused training, was developed. The latter was concluded to be the desired option, as it is a balanced approach of both outcome and process focus.

Investigating lexical categorization in reading based on joint diagnostic and training approaches for language learners

Efficient reading is essential for societal participation, so reading proficiency is a central educational goal. Here, we use an individualized diagnostics and training framework to investigate processes in visual word recognition and evaluate its usefulness for detecting training responders. We (i) motivated a training procedure based on the Lexical Categorization Model (LCM) to introduce the framework. The LCM describes pre-lexical orthographic processing implemented in the left-ventral occipital cortex and is vital to reading. German language learners trained their lexical categorization abilities while we monitored reading speed change. In three studies, most language learners increased their reading skills. Next, we (ii) estimated, for each word, the LCM-based features and assessed each reader’s lexical categorization capabilities. Finally, we (iii) explored machine learning procedures to find the optimal feature selection and regression model to predict the benefit of the lexical categorization training for each individual. The best-performing pipeline increased reading speed from 23% in the unselected group to 43% in the machine-selected group. This selection process strongly depended on parameters associated with the LCM. Thus, training in lexical categorization can increase reading skills, and accurate computational descriptions of brain functions that allow the motivation of a training procedure combined with machine learning can be powerful for individualized reading training procedures.

Asymmetric Mutual Alignment for Unsupervised Zero-Shot Sketch-Based Image Retrieval

In recent years, many methods have been proposed to address the zero-shot sketch-based image retrieval (ZS-SBIR) task, which is a practical problem in many applications. However, in real-world scenarios, on the one hand, we can not obtain training data with the same distribution as the test data, and on the other hand, the labels of training data are not available as usual. To tackle this issue, we focus on a new problem, namely unsupervised zero-shot sketch-based image retrieval (UZS-SBIR), where the available training data does not have labels while the training and testing categories are not overlapping. In this paper, we introduce a new asymmetric mutual alignment method (AMA) including a self-distillation module and a cross-modality mutual alignment module. First, we conduct self-distillation to extract the feature embeddings from unlabeled data. Due to the lack of available information in an unsupervised manner, we employ the cross-modality mutual alignment module to further excavate underlying intra-modality and inter-modality relationships from unlabeled data, and take full advantage of these correlations to align the feature embeddings in image and sketch domains. Meanwhile, the feature representations are enhanced by the intra-modality clustering relations, leading to better generalization ability to unseen classes. Moreover, we conduct an asymmetric strategy to update the teacher and student networks, respectively. Extensive experimental results on several benchmark datasets demonstrate the superiority of our method.

Target Detection and Segmentation Technology for Zero-shot Learning

Zero-shot learning (ZSL) in the field of computer vision refers to enabling the model to recognize and understand categories that have not been encountered during the training phase. It is particularly critical for object detection and segmentation tasks, because these tasks require the model to have good generalization capabilities to unknown categories. Object detection requires the model to determine the location of the object, while segmentation further requires the precise demarcation of the object's boundaries. In ZSL research, knowledge representation and transfer are core issues. Researchers have tried to use semantic attributes as a knowledge bridge to connect categories seen during the training phase and categories not seen during the testing phase. These attributes may be color, shape, etc., but this method requires accurate attribute annotation, which is often not easy to achieve in practice. Therefore, researchers have begun to explore the use of non-visual information such as knowledge maps and text descriptions to enrich the recognition capabilities of models, but this also introduces the challenge of information integration and alignment. At present, ZSL has made certain progress in target detection and segmentation tasks, but there is still a significant gap compared with traditional supervised learning. This is mainly due to the limited ability of ZSL models to generalize to new categories. To this end, researchers have begun to explore combining ZSL with other technologies, such as generative adversarial networks (GANs) and reinforcement learning, to enhance the model's detection and segmentation capabilities for new categories. Future research needs to focus on several aspects. The first is how to design a more effective knowledge representation and transfer mechanism so that the model can better utilize existing knowledge. The second step is to develop new algorithms to improve the performance of ZSL in complex environments. In addition, research should focus on how to reduce the dependence on computing resources so that the ZSL method can run effectively in resource-limited environments. In summary, the research on target detection and segmentation technology of zero-shot learning is a cutting-edge topic in the field of computer vision. Despite the challenges, with the deepening of research, we expect these technologies to contribute to improving the generalization ability and intelligence level of computer vision systems.