Search

Semi-supervised self-training method can train an effective classifier by exploiting labeled and unlabeled samples. Recently, a self-training method based on density peaks of data (STDP) is proposed. However, it still suffers from some shortcomings to be addressed. For example, STDP is affected by cut-off distance $d_{c}$ . As a result, it is tricky for STDP to select an optimal parameter $d_{c}$ on each data set. Furthermore, STDP has a poor performance on data sets with some variations in density because of cut-off distance $d_{c}$ . In order to solve these problems, we present a new self-training method which connects unlabeled and labeled samples as vertexes of an optimum path forest to discover the underlying structure of feature space. Furthermore, the underlying structure of the feature space is used to guide the self-training method to train a classifier. Compared with STDP, our algorithm is free of parameters and can work better on data sets with some variations in density. Moreover, we are surprised to find that our algorithm also has some advantages in dealing with overlapping data sets. The experimental results on real data sets clearly demonstrate that our algorithm has better performance than some previous works in improving the performance of base classifiers of k-nearest neighbor, support vector machine and cart.

Sensor drift is one of the important problems affecting the stability of gas sensing and detection system. We proposed a novel memory mechanism, which can be combined with the traditional machine learning classifier to deal with and improve the sensor drift problem. Based on the memory mechanism, we established recognition model, which adopts support vector machine as base-classifier, to achieve better classification effect and eliminate the imbalance of data set. The Gas Sensor Array Drift Dataset from UCI Machine Learning Repository was selected for experimental verification. Through experiments, the method proposed in this paper can improve the classification effect in the short and medium term, postpone the negative effects brought by drift to a certain extent, and then extend the service life of the gas sensor.

Self-training and unsupervised pre-training have emerged as effective approaches to improve speech recognition systems using unlabeled data. However, it is not clear whether they learn similar patterns or if they can be effectively combined. In this paper, we show that pseudo-labeling and pre-training with wav2vec 2.0 are complementary in a variety of labeled data setups. Using just 10 minutes of labeled data from Libri-light as well as 53k hours of unlabeled data from LibriVox achieves word error rates (WER) of 2.8%/4.8% on the clean and other test sets of Librispeech – rivaling the best published systems trained on 960 hours of labeled data only a year ago. Training on all labeled data of Librispeech achieves WERs of 1.5%/3.1%.

Semi-supervised learning on class-imbalanced data, although a realistic problem, has been under studied. While existing semi-supervised learning (SSL) methods are known to perform poorly on minority classes, we find that they still generate high precision pseudo-labels on minority classes. By exploiting this property, in this work, we propose Class-Rebalancing Self-Training (CReST), a simple yet effective framework to improve existing SSL methods on class-imbalanced data. CReST iteratively retrains a baseline SSL model with a labeled set expanded by adding pseudo-labeled samples from an unlabeled set, where pseudo-labeled samples from minority classes are selected more frequently according to an estimated class distribution. We also propose a progressive distribution alignment to adaptively adjust the rebalancing strength dubbed CReST+. We show that CReST and CReST+ improve state-of-the-art SSL algorithms on various class-imbalanced datasets and consistently outperform other popular rebalancing methods. Code has been made available at https://github.com/google-research/crest., To appear in CVPR 2021. Code release: https://github.com/google-research/crest

Diabetic feet are a long-term effect of diabetes mellitus that are at risk of ulceration due to neuropathy and ischemia. Early ulcer stages show subtle changes hard to recognize by the human eye, especially on darker skin types. Acquired ulcers may become chronic for various reasons, requiring extensive documentation to monitor healing progression. For early stage detection and documentation support, object detection algorithms are a key technology for prevention and care improvement. However, attendant symptoms like malformed toenails, hyperkeratosis, and rhagades display challenges regarding faulty detections. The research at hand explores two disparate state-of-the-art detection frameworks: Detection Transformer (DETR) as representative of the novel transformer-based architectures for computer vision, and You Only Look Once v5 (YOLOv5) as an expedited PyTorch port of YOLOv4 with explicit mobile-focus. Both are compared on a recently released dataset for diabetic foot ulcer detection with images typical for common wound care documentation. In addition, effects of self-training for performance improvement are investigated. Achieved results outperform those of other state-of-the-art methods. These are discussed highlighting differences and potential for further optimization.

One commonly used clinical approach towards detecting melanomas recognises the existence of Ugly Duckling nevi, or skin lesions which look different from the other lesions on the same patient. An automatic method of detecting and analysing these lesions would help to standardize studies, compared with manual screening methods. However, it is difficult to obtain expertly-labelled images for ugly duckling lesions. We therefore propose to use self-supervised machine learning to automatically detect outlier lesions. We first automatically detect and extract all the lesions from a wide-field skin image, and calculate an embedding for each detected lesion in a patient image, based on automatically identified features. These embeddings are then used to calculate the L2 distances as a way to measure dissimilarity. Using this deep learning method, Ugly Ducklings are identified as outliers which should deserve more attention from the examining physician. We evaluate through comparison with dermatologists, and achieve a sensitivity rate of 72.1% and diagnostic accuracy of 94.2% on the held-out test set., Accepted at Sixth ISIC Skin Image Analysis Workshop @ CVPR 2021

The goal of semi-supervised object detection is to learn a detection model using only a few labeled data and large amounts of unlabeled data, thereby reducing the cost of data labeling. Although a few studies have proposed various self-training-based methods or consistency regularization-based methods, they ignore the discrepancies among the detection results in the same image that occur during different training iterations. Additionally, the predicted detection results vary among different detection models. In this paper, we propose an interactive form of self-training using mean teachers for semi-supervised object detection. Specifically, to alleviate the instability among the detection results in different iterations, we propose using nonmaximum suppression to fuse the detection results from different iterations. Simultaneously, we use multiple detection heads that predict pseudo labels for each other to provide complementary information. Furthermore, to avoid different detection heads collapsing to each other, we use a mean teacher model instead of the original detection model to predict the pseudo labels. Thus, the object detection model can be trained on both labeled and unlabeled data. Extensive experimental results verify the effectiveness of our proposed method.

Deep spectral clustering networks have shown their superiorities due to the integration of feature learning and cluster assignment, and the ability to deal with non-convex clusters. Nevertheless, deep spectral clustering is still an ill-posed problem. Specifically, the affinity learned by the most remarkable SpectralNet is not guaranteed to be consistent with local invariance and thus hurts the final clustering performance. In this paper, we propose a novel framework of Constrained Spectral Clustering Network (CSCN) by incorporating pairwise constraints and clustering oriented fine-tuning to deal with the ill-posedness. To the best of our knowledge, this is the first constrained deep spectral clustering method. Another advantage of CSCN over existing constrained deep clustering networks is that it propagates pairwise constraints throughout the entire dataset. In addition, we design a clustering oriented loss by self-training to simultaneously finetune feature representations and perform cluster assignments, which further improve the quality of clustering. Extensive experiments on benchmark datasets demonstrate that our approach outperforms the state-of-the-art clustering methods.

Automatic understanding of facial behavior is hampered by factors such as occlusion, illumination, non-frontal head pose, low image resolution, or limitations in labeled training data. The EmotioNet 2020 Challenge addresses these issues through a competition on recognizing facial action units on in-the-wild data. We propose to combine multi-task and self-training to make best use of the small manually / fully labeled and the large weakly / partially labeled training datasets provided by the challenge organizers. With our approach (and without using additional data) we achieve the second place in the 2020 challenge - with a performance gap of only 0.05% to the challenge winner and of 5.9% to the third place. On the 2018 challenge evaluation data our method outperforms all other known results.

Pupil detection is an indispensable step in eye-tracking, however, there are lots of difficulties in locating the pupil under low-resolution environments. This study proposes a self-training method of pupil detection for eye-tracking under low resolution. The system is applied to a typical application scenario as using a personal computer, a webcam with a resolution of $640 \times 480$ is used in our study. We first generate an initial pupil pattern by verifying the color intensity of the eye regions and estimate the gaze according to the matched pupil position in the further frames. After that, we utilize the computer users' actions of mouse clicks to calibrate the estimation result and annotate the position of the pupil to train the weight map of the pupil pattern. The experimental result shows that the proposed method could minimize the effect of bright spots, moreover, the precision of pupil detection and eye-tracking is increased. The proposed method will be potentially used in light-weight applications of human-computer interaction.

The goal of meta-learning is to learn a learning procedure to generate a learner from only a handful of labeled datapoints. However, the learning procedure is learned by a meta-learner from an enormous amount of few-shot tasks constructed from a large amount of labeled datapoints. From this point of view, few-shot learning is also depending on huge amount of labeled data. In this paper, we present a simple and efficient method for few-shot classification in a semi-supervised setting where only a small portion of training samples are labeled. We assign these unlabeled data pseudo-labels using a classifier trained with both labeled and unlabeled data. Once the pseudo-label obtained, we can run meta-learning over tasks constructed from labeled and pseudo-labeled data. We evaluate our method on miniImagenet and tieredImagenet benchmarks whose meta-training sets are split into unlabeled portion and labeled portion further in order to adapt to our framework. Our experimental results confirm that our semi-supervised meta-learning approach acquires a considerable performance gain over meta-learning with only labeled data and significantly outperforms previous state-of-the-art semi-supervised meta-learning methods.

In a data-driven world, semi-supervised learning methods are motivated by the availability of large unlabeled datasets than a small amount of labeled data. However, incorporating unlabeled data into learning does not guarantee an improvement in classification performance. In this paper, we present an approach based on a deep learning system to predict missing classes by integrating a model of semi-supervised learning which is the self-training. In order to evaluate its performance, we used a set of diabetes data and four performance measures: Precision, Recall, F-Measure and Area Under the ROC Curve (AUC).

Recent advances in domain adaptation show that deep self-training presents a powerful means for unsupervised domain adaptation. These methods often involve an iterative process of predicting on target domain and then taking the confident predictions as pseudo-labels for retraining. However, since pseudo-labels can be noisy, self-training can put overconfident label belief on wrong classes, leading to deviated solutions with propagated errors. To address the problem, we propose a confidence regularized self-training (CRST) framework, formulated as regularized self-training. Our method treats pseudo-labels as continuous latent variables jointly optimized via alternating optimization. We propose two types of confidence regularization: label regularization (LR) and model regularization (MR). CRST-LR generates soft pseudo-labels while CRST-MR encourages the smoothness on network output. Extensive experiments on image classification and semantic segmentation show that CRSTs outperform their non-regularized counterpart with state-of-the-art performance. The code and models of this work are available at https://github.com/yzou2/CRST., Comment: Accepted to ICCV 2019 (Oral)

In this article a real-time visual feedback system for strength self-training is proposed, with which users, especially beginners can understand comparison between exact motion and theirs. Using a camera-based motion capture and a normal LCD, the system does not require any additional devices to wear. Such the easy installation of the system benefits of the users who practice not only in a gym but also at his/her home. The performance and usefulness of the system are evaluated by a user test.

Self-training is a useful strategy for semi-supervised learning, leveraging raw texts for enhancing model performances. Traditional self-training methods depend on heuristics such as model confidence for instance selection, the manual adjustment of which can be expensive. In addition, characteristics of extra training sentences are not considered beyond the baseline method. To address these challenges, we propose a deep reinforcement learning method to learn self-training strategy automatically. Based on the neural representation of sentences and the hidden features from classifiers, a deep Q-network based model is designed to capture their linguistic characteristics and learn an optimal policy for instance selection automatically. Results show that our approach outperforms baseline self-training in terms of better performances and stability.

This paper discussed the self-training system for a hemiplegia person using power assist hand. The power assist hand is the rehabilitation equipment for hemiplegic disabilities using bellows actuators. We developed the self-training system which is composed of a power assist hand and a hand motion capture device (HMCD). The HMCD used here is a Leap Motion Sensor. This sensor is very useful device for getting the positions of finger joints. The system recognizes the flexion or extension motion of healthy fingers using the leap motion sensor, the power assist hand moves according to the motion of healthy hand. The feature of this system is the good usability that the user can use intuitively. The effectiveness and usability of our system are clarified by several experiments.

Cancer diseases have serious influence on people's live, but the-state-of-art machine learning approaches have the potential to decrease cancer death rates by formulating prevention strategies for cancer treatment. Some supervised learning methods have been used to give early warning of cancer successfully by using gene expression data, but the most prominent challenge is the insufficient labeled samples of biological data, especially in cancer datasets. It may cause the training model with over-fitting. Therefore, semi-supervised method approaches, such as self-training, are designed to improve the model performance by utilizing the unlabeled samples which are more than the labeled samples in some biological data. The high-quality pseudo-labeled sample is selected to train for improvement of model based curriculum learning paradigm. Of note, the self-training can meet above requirements. Next samples with high quality are pseudo-labeled by self-training. Afterward, put in the training dataset to increase the number of available labeled samples, and the cycle will continue until all unlabeled samples are annotated. The ensemble learning is a widely learning method, which can improve the model classification performance by combining multi-week classifier style. It can help self-training to learn the initial classifiers using the labeled samples. Therefore, in this study an ensemble self-training learning(ESTL) method is proposed to selects the unlabeled samples with high-quality more effectively and improves the robustness of the model. Compared with other tradition classification algorithms, the results of real cancer dataset showed that our proposed ESTL approach improve about 5% in classification performance.

This paper describes a prototype self-training system for tennis forehand shots that allows trainees to practice their motion forms by themselves. Our system uses a motion capture device to record a trainee's motion, and visualizes the differences between the features of the trainee's motion and the correct motion as performed by an expert. This system enables trainees to understand the errors in their motion and how to reduce or eliminate them. In this study, we classified the motion features and corresponding visualization methods using one-dimensional spatial, rotational, and temporal features based on the key sporting poses. We also developed a statistical model for the motion features, allowing the system to assess and prioritize all features of a trainee's motion. This research focuses on the motion of a tennis forehand shot and evaluates our prototype through several user experiments.

We created a self-training assessment support system utilizing a web camera for the self-study of injection techniques. In this study, we tested the effectiveness of utilizing this assessment support system for improving the skills of 19 second-year nursing students. The results showed that being able to objectively observe those techniques requiring correction led to nurses taking their errors seriously. In particular, the results suggested that an assessment support system that can capture blind spots is useful for the self-study of difficult techniques requiring fine hand movements such as injection.

In educational data mining, program-level student classification is a popular task focusing on predicting the final study status of a student at the end of the permitted period of study time with a program. In a practical context where the task is applied to predict students early in their learning path, two issues have to be addressed. The first issue is the completeness of data in both model construction and prediction of new data, while the latter refers to the number of labeled data in model construction. Nonetheless, a few works considered this practical context thoroughly. In order to manage these issues, we propose a self-training algorithm using $k$ -means and random forest models, named $k$ -meansTriForest. Our algorithm is regarded as a solution to the program-level student classification task in the aforementioned practical context. $k$ -means helps us form a cluster-based incomplete data imputation method. Random forests play an important role of base classifiers of an ensemble in our algorithm. Besides, the Tri-training styled unlabeled instance selection is defined so that our algorithm can remove the burden of probability threshold settings from its users. Moreover, in our algorithm, the diversity of the base classifiers can be preserved when the unlabeled data set can be exploited and the labeled data set can be enhanced over time. As a result, our algorithm outperforms the previous ones in the related works on three real educational data sets, in order to classify students into the groups with different final study status at three different points in time. In particular, our improvements in Accuracy are ranged from about 1.5% to about 7%. Such higher accuracy in an empirical study with many various incomplete data and unlabeled data percentages has confirmed our algorithm's effectiveness and robustness.

Self-training semi-supervised classification methodology is highly effective in alleviating the shortage of labeled data in classification tasks via an iterative self-training process. In this paper, we propose a self-training semi-supervised classification algorithm based on density peaks of data and differential evolution. The proposed algorithm consists of two main parts. First part is to use the underlying structure of data space, which is discovered based on density peaks of data, to help train a better classifier. Second part is to use the differential evolution to optimize the positioning of newly labeled data during the self-training process, where newly labeled data denotes the unlabeled data labeled by classifier during the self-training process and optimizing the positioning means optimally adjusting the attributes values of date. Experimental results on 12 benchmark datasets clearly demonstrate that the proposed algorithm is more effective than some previous works in improving the performance of base classifier of support vector machine or k-nearest neighbor.

Self-training plays an important role in sports exercise. However, if not under the instruction of a coach, it would be ineffective for most amateurs or inexperienced players to exercise on their own. Therefore, establishing computerassisted training systems for sports exercise is a recently emerging topic. In this paper, we propose a billiard self-training system, which aims at improving billiard players’ performance by utilizing intelligent glasses as a wearable camera and displayer. The proposed system is able to automatically analyze user-captured images of the billiard table from multiple views and display the ball configurations on a virtual top-view table. Enriched visual presentation can be provided to give the practitioner a further sight into the game. The experiments conducted on sixteen sets of different ball configurations show promising results.

The Semi-Supervised Support Vector Machine (S3VM) solves a non-convex, Mixed-Integer Program (MIP). Due to difficulty in solving the problem, convex approximations have typically been used. However, existing approaches suffer from poor scalability and struggle on certain datasets, compared to graph based counterparts. The poor predictive performance suggests that for some datasets, convex approximations may not be a sufficiently accurate approximation to the problem. We present a self-training approach with self-adapting regularization parameters for S3VM formulations. At each iteration, the regularization parameters are adapted to better reflect label confidence, class proportion, and to gradually include more unlabeled points. We show that updating the S3VM framework iteratively in this fashion, the sequence of SVM subproblems can be solved very efficiently and the solution generated by this sequence yields superior performance compared to leading SSL methods.

Semi-supervised classification methods try to improve a supervised learned classifier with the help of unlabeled data. In many cases one assumes a certain structure on the data, as for example the manifold assumption, the smoothness assumption or the cluster assumption. Self-training is a method that does not need any assumptions on the data itself. The idea is to use the supervised trained classifier to label the unlabeled points and to enlarge this way the training data. This paper aims to show that a self-training approach with soft-labeling is preferable in many cases in terms of expected loss (risk) minimization. The main idea is to use a soft-labeling to minimize the risk on labeled and unlabeled data together, in which the hard-labeled self-training is an extreme case.

The biggest issue is painful when pregnant women are in normal spontaneous delivery, although prenatal education has taught pregnant women Lamaze breathing to relax. There may be a lack of exercise and low participation, so it is difficult to execute the breathing method. This proposal used smart devices to develop App for Lamaze breathing self-training application, in order to enhance the proficiency and effectiveness. When pregnant women are in labor, it will be useful to reduce pain.

Teaching a complex discipline like power electronics is an open question and the best educational approach or the objectives to be achieved at the end of a power electronic course are still in debate. This paper presents a visual didactic approach chosen by the Authors in the last years in their respective courses in power electronics. It also discuss the objectives and the skills required to the students at the end of the course: in particular, some meta-cognition objectives recently fully entered into the education world in general. Such meta-cognition goals are now discussed for the particular case of power electronics courses. The Switching Functions (SFs) generalized approach is then discussed as an extensive and useful tool that can be integrated in many different simulation programming environment, encouraging students to use it in many different analysis situations both for examining the behavior of a power converter and to introduce them in the design of firing circuit and power system control strategies.

The feasibility of using video-games for self-training of the upper extremity and balance following stroke was assessed. Specifically we aimed to compare the (1) training time, (2) satisfaction and (3) effectiveness of a self-training program using video-games compared to a traditional self-training program. A single-blind randomized controlled trial with four assessments; two before and two following the 5-week self training intervention was conducted. The self-training intervention included playing video-games or performing traditional exercises for one hour a day, 5 times a week for 5 weeks. To date 12 participants completed the intervention. High levels of compliance and satisfaction with the intervention were found for both groups. Higher perceived exertion was reported for playing the video-games while standing which also improved standing balance. An overall improvement in upper extremity measures was seen. These preliminary findings are encouraging and we are still recruiting participants into the study.

This paper aims to analyze user's emotion automatically by analyzing Twitter using "data without sentiment labels", not only "data with sentiment labels", to increase accuracy of sentiment analysis through an improved Self-Training, one of Semi-supervised learning techniques. Self-Training has a weak point that a classification mistake can reinforce itself. Self-Training iteratively modifies the model based on the output of the model. Thus, if the model generates wrong output, the model can be wrongly modified. For alleviate this weak point, we propose a competitive Self-Training technique. We create three models based on the output of the model and choose the best. Three models are created by binary mixture perspectives: the threshold, the same number, and the maximum number for updates. We repeat step that creating model and choosing a best model highest to get F-measure. Finally, we can improve the performance of sentiment analysis model.

This paper presents a domain adaptive action recognition approach, which utilizes labeled training videos taken under one environment (source domain) to train an action classifier for the videos taken under another environment (target domain), so that the cost for preparing training data can be greatly alleviated. Our proposed approach jointly utilizes self-training and feature selecting to gradually select these training data and feature dimensions that contribute to the training in target domain. With the proposed approach, classifiers for videos in new environments can be learned efficiently without extra labeling efforts. The superiority of our approach has been confirmed by multiple benchmark dataset.

Ruling line removal is an important pre-processing step in document image processing. Several algorithms have been proposed for this task. However, it is important to be able to take full advantage of the existing algorithms by adapting them to the specific properties of a document image collection. In this paper, a system is presented, appropriate for fine-tuning the parameters of ruling line removal algorithms or appropriately adapt them to a specific document image collection, in order to improve the results. The application of our method to an existed line removal algorithms is presented.

Self-training plays an important role in sports exercise. However, if not under the instruction of a coach, improper training postures can cause serious harm to muscles and ligaments of the body. Hence, the development of computer-assisted self-training systems for sports exercise is a recently emerging research topic. In this paper, we propose a Yoga self-training system, entitled YogaST, which aims at instructing the user/practitioner to perform the asana (Yoga posture) correctly and preventing injury caused by improper postures. Involving professional Yoga training knowledge, YogaST analyzes the practitioner's posture from both front and side views using two Kinects with perpendicular viewing directions and assists him/her in rectifying bad postures. The contour, skeleton, and feature axes of the human body are extracted as posture representation. Then, YogaST analyzes the practitioner's posture and presents visualized instruction for posture rectification so that the practitioner can easily understand how to adjust his/her posture.

Sufficient training with feedback was important for nursing students to learn the techniques. In view of this, we studied the method for measuring and evaluating the performance of nursing students in order to develop a self-training system. Focusing on the training of transferring a patient from a bed to a wheelchair, we defined seven evaluation items related with the postures. In addition, evaluation indexes of each item were determined. Then, we established a prototype system based on two Kinect range cameras. Using the system, first, we recognized the body parts and joints through the color of the markers attached on the bodies. After that, the body joints' spatial locations and body parts' inclination angles were measured via the combination of color and depth information in order to calculate the indexes. We applied Bayes minimum error decision to classify nursing students' performance of each items as correct or incorrect. Ten inexperienced nursing students and five experienced nurses were asked to transfer patient from a bed to a wheelchair at least twice. Every time the patient was transferred, the nursing teacher evaluated the trainee's performance. In addition, proposed system measured and recorded the data. The significant difference between correct and incorrect performance of each item was observed through the determined indexes (P

The purpose of this paper was to develop a support system to help nursing trainees learn the skill of bed-making. The system consisted of three steps: measurement, evaluation, and feedback. This paper focused on the first two issues. First, the evaluation points of bed-making were determined. Next, a measurement system was constructed using color and distance information provided by Kinects and image processing. The system extracted specific segments of whole images depicting trainees, a bed, and necessary equipment. Finally, the procedure used by trainees in making beds was quantitatively evaluated using thresholds that were determined in advance by observing teachers and students involved in bed-making. Compared to the evaluation of nursing teachers, the accuracy of the evaluation system was as much as 70%.

By combining self-training method of the semi-supervised learning with two-dimensional principal component analysis (2DPCA), a semi-supervised learning based face recognition method is proposed. On the basis of two-dimensional principal component analysis, few labeled samples are used to obtain classifier. Then unlabeled samples are classified by the classifier. And according to the self-training method of semi-supervised learning, the face samples with the highest confidence are added to the training set in order to increase the number of face samples in training set. Experimental results on ORL and Yale face database show the effectiveness of the presented method.

The task of finding transcription start sites (TSSs) can be modeled as a classification problem. Semi-Supervised Support Vector Machines (S3VMs) are an appealing method for using unlabeled data in classification. Based incorporation prior biological knowledge for recognizing TSSs, propose a Self-Training S3VMs (ST-S3VMs) algorithm. ST-S3VM builds a SVM classifier based small amounts of labeled data and large amounts of unlabeled data, incorporates prior biological knowledge by engineering an appropriate kernel function with a self-training algorithm The algorithm has been implemented and tested on previously published data. Our experimental results on real nucleotide sequences data show that our method improve the prediction accuracy greatly and our method performs significantly better than ESTSCAN and SVMs with Salzberg kernel.

Document Image Binarization techniques have been studied for many years, and many practical binarization techniques have been developed and applied successfully on commercial document analysis systems. However, the current state-of-the-art methods, fail to produce good binarization results for many badly degraded document images. In this paper, we propose a self-training learning framework for document image binarization. Based on reported binarization methods, the proposed framework first divides document image pixels into three categories, namely, foreground pixels, background pixels and uncertain pixels. A classifier is then trained by learning from the document image pixels in the foreground and background categories. Finally, the uncertain pixels are classified using the learned pixel classifier. Extensive experiments have been conducted over the dataset that is used in the recent Document Image Binarization Contest(DIBCO) 2009. Experimental results show that our proposed framework significantly improves the performance of reported document image binarization methods.

Self-training learning is one of the most important semi-supervised learning paradigms in which a learner keeps on classifying the unlabeled examples and retaining the most confident examples to the training set. With the increasing training set, it is possible to enhance the classification performance on unseen data. However, sometimes the classifier misclassifies some unlabeled examples and keeps them in the training set, which worse the classification performance. In this paper, we present a novel method based on local consistency to eliminate the noises. According the manifold assumption, an unlabeled example expects to join the training set if its label given by classifier should be consistent with the local neighborhood in the training set on the manifold. We test the new method on several data sets from synthetic and real-world data from UCI, the empirical result indicates the proposed approach is effective and reliable.

we have developed a biofeedback-based breath training environment in which the user can easily relax his body and mind only by following the induction. The environment includes body relaxation video guide and breathing induction program. The body relaxation video guide helps relax the body with simple stretch motions in a seating posture. The breathing induction program is operated in two modes: biofeedback training mode and self training mode. In the biofeedback training mode, a visual breathing induction is provided and each breath following the induction is assessed by processing the change of pulse rates measured from photo-plethysmography. In self-training mode, voice and visual induction is provided to breathe consciously in a natural posture without any physical restrictions such as sensor contact and visual induction eye tracing. The environment is also equipped with a breathing test function. The breathing test evaluates current breathing habit to recommend an optimal breathing guide.

This paper proposes a self-training algorithm for load balancing in cluster computing. It uses load information including CPU load, memory usage and network traffic to decide the load of each node and combines this information with properties of each job, including CPU bound, memory bound and I/O bound features, that are extracted from the previous runs of these jobs. The proposed algorithm is compared to another algorithm that only uses the load information of each node for the purpose of load balancing. The performance evaluation results show that the proposed algorithm performs well.

Use the algorithm of generating variable-radius detectors to generate detectors. Analyze different effects on detection results by choosing different radii. Test samples need to compare with all detectors to detect intrusions. When the number of detectors is large, the detection speed is slow. Cluster near detectors into an area. The test samples which fall in this area only compare with the detectors in this area not all detectors, so the times of comparing are reduced and the detection speed is increased. For the test sample falling in some area of non-self space which is not covered by any detector, comparing the minimum distance between this test sample and all self training samples with the minimum distance between this test sample and all detectors can judge whether this test sample is non-self It is proved by experiments that this method can increase the detection speed as well as detection rate.

ACRE is a robotic training device for rehabilitation of the upper limbs. ACRE stimulates patients to train their affected arm more frequently and repetitively, via so called game based therapy. A prototype of ACRE (ACRE2) was tested in a user trial with patients who had recently suffered a stroke. Therapists concluded that the ACRE can be useful in addition to the conventional training. After the first user trail specific technical improvements were suggested and realized in a new version of the robot (ACRE3) to increase safety, comfort and autonomous self-training. Patients and therapists experienced training with the ACRE arm pleasant and beneficial for recovery. Eventually, robot-mediated therapy may be applied instead of or in addition to one-to-one therapy and therapist-to-group therapy and may lead to a better rehabilitation. More research is needed especially on the cost-effectiveness of the ACRE in rehabilitation practice. © 2007 IEEE.

Neural networks have been under examination for decades in many areas of research, with varying degrees of success and acceptance. Key goals of computer learning, rapid problem solution, and automatic adaptation have been elusive at best. This paper1,2 summarizes efforts at NASA's Marshall Space Flight Center harnessing such technology to autonomous space vehicle docking for the purpose of evaluating applicability to future missions.

The educational necessity of the person consisting of two components is analyzed: information and learning. There are main principles of the self-training theory, which is based on subject-objective relations during study-information interaction and study-cognitive activity.

The author introduces a series of novel approaches to backpropagation: (1) the use of logic forms (classical, modal, and nonmonotonic) as training tools; (2) the construction of new nets through the responses of logically trained nets (weight sets); (3) the use of N2 as a reset mechanism for impermissibly slow or false responses by subnets; and (4) the retraining of failing subnets by the logically trained nets. A biologically plausible basis for the system is offered. >

We present a new self-training method for adjusting the coefficients of a transversal equalizer with T-spaced taps in a multilevel partial response class-IV (PRIV) system. Self-training equalization from distorted random data signals is inherently more difficult to achieve for partial-response systems than for full-response systems. An equalizer with T-spaced taps is sufficient to obtain equalized output signals for arbitrary sampling phase. Following Sato (1975), in the described self-training equalization algorithm we first transform the equalizer output into full-response form, then compute a pseudo-error signal, and finally translate the pseudo-error signal into an error signal for the desired partial-response equalizer output. This error signal is used to adjust the equalizer coefficients according to the LMS algorithm. The new method differs from the Sato algorithm in two ways. First, the channel inversion for obtaining full-response signals is accomplished exactly by mixed linear feedback and decision feedback equalization, whereas in the case of the Sato algorithm the inversion is only achieved approximately. Secondly, for the derivation of pseudo-error signals more knowledge of the statistical properties of ideal, but noisy full-response signals is exploited. We present simulation results illustrating the superior convergence properties of the new self-training method.

The unsupervised learning of an autonomous mobile robot is a real research topic. It permits an artificial system to interact successfully with its environment and to avoid obstacles. This paper presents an intelligent control architecture which integrates self-training methods and is able to operate in complex, unknown environments in order to achieve its target. Our approach is based on reactive obstacle avoidance. The intelligent model integrates different neural networks and permits on-line learning. The results of experiments are discussed.

The purpose of this article is to present an original pedagogical product for teaching microwaves. It is a multimedia self-training product recommended to technicians and engineers working in conventional electronics and wishing to acquire advanced knowledge in microwaves in connection with their working structure. This self-training package first module contains 12 lessons. It is composed of two media: - a conventional aid: the paper document - a computer aid : the interactive educational software.

A self-training visual inspection system using a connectionist classifier is presented. The system is composed of a control unit, a signal-processing unit, and a connectionist classifier. The control unit both generates the training set and performs the function of teacher to the classifier. The second unit compresses the two-dimensional image into a one-dimensional signal. Potential flaws extracted from the one-dimensional signal are sent to the classifier. The classifier used in this work is a standard multilayer connectionist neural network that uses backpropagation for learning. The system is applied to two inspection tasks involving two-dimensional surfaces characterized by a known intensity distribution. Diagnostics for evaluating the classifier are presented, along with an evaluation of the classifier's performance. >