A sequential multitask learning algorithm for pattern recognition

Due to technological advances, a massive amount of data is produced daily, presenting challenges for application areas where data needs to be labelled by a domain specialist or by expensive procedures, in order to be useful for supervised machine learning purposes. In order to select which data points will provide more information when labelled, one can make use of active learning methods. Active learning (AL) is a subfield of machine learning which addresses methods to build models with fewer, but more representative instances. Even though AL has been vastly studied, it has not been thoroughly investigated in hierarchical multi-label classification, a learning task where multiple class labels can be assigned to an instance and these labels are hierarchically structured. In this work, we provide a public framework containing baseline and state-of-the-art algorithms suitable for this task. Additionally, we also propose a new algorithm, namely Hierarchical Query-By-Committee (H-QBC), which is validated on datasets from different domains. Our results show that H-QBC is capable of providing superior predictive performance results compared to its competitors, while being computationally efficient and parameter free.

This paper presents a new sequential multi-task learning model with the following functions: one-pass incremental learning, task allocation, knowledge transfer, task consolidation, learning of multi-label data, and active learning. This model learns multi-label data with incomplete task information incrementally. When no task information is given, class labels are allocated to appropriate tasks based on prediction errors; thus, the task allocation sometimes fails especially at the early stage. To recover from the misallocation, the proposed model has a backup mechanism called task consolidation, which can modify the task allocation not only based on prediction errors but also based on task labels in training data (if given) and a heuristics on multi-label data. The experimental results demonstrate that the proposed model has good performance in both classification and task categorization.

Language Impairment (LI) is a developmental disorder that mainly manifests impaired language learning and processing. Evidence, largely from English-speaking population studies, has shown that children with LI compared to typically developing (TD) children have low scores in sequential learning tasks but similar performance in declarative learning tasks. According to the declarative/procedural model, LI children compensate for their deficiency in syntactic skills (i.e., deficits in the procedural memory system) by using the declarative memory system (indispensable for vocabulary acquisition). Although there are specific deficits in children with LI depending on the language they speak, it is assumed that this model can explain the shortcomings of such pathology regardless of the language spoken. In the current study, we compared the performance of fifteen school-aged Mexican Spanish-speaking children with LI and twenty TD children during sequential and declarative learning tasks and then analyzed the relationship between their performance in these tasks and their abilities in syntax and semantics. Children with LI displayed lower scores than normal children in the sequential learning task, but no differences were found in declarative learning performance with verbal or visual stimuli. No significant correlations were observed in children with LI between their performance in sequential learning and their abilities in semantics and no significant correlations were observed in TD children between their performance in sequential learning and their abilities in syntax. In contrast, for children with LI, a significant correlation between their performance in declarative learning and their abilities in semantics was observed and for the group of TD children a significant correlation between their performance in declarative learning and their abilities in syntax was observed. This study shows that Spanish-speaking children with LI display a pattern of learning impairment that supports the declarative/procedural model hypothesis. However, they display poor verbal declarative learning skills, probably due to low verbal working memory capacity.

Multi-label classification has many applications in the text categorization, biology and medical diagnosis, in which multiple class labels can be assigned to each training instance simultaneously. As it is often the case that there are relationships between the labels, extracting the existing relationships between the labels and taking advantage of them during the training or prediction phases can bring about significant improvements. In this paper, we have introduced positive, negative and hybrid relationships between the class labels for the first time and we have proposed a method to extract these relations for a multi-label classification task and consequently, to use them in order to improve the predictions made by a multi-label classifier. We have conducted extensive experiments to assess the effectiveness of the proposed method. The obtained results advocate the merits of the proposed method in improving the multi-label classification results.

Multi-label learning deals with training examples each represented by a single instance while associated with multiple class labels. Due to the exponential number of possible label sets to be considered by the predictive model, it is commonly assumed that label correlations should be well exploited to design an effective multi-label learning approach. On the other hand, class-imbalance stands as an intrinsic property of multi-label data which significantly affects the generalization performance of the multi-label predictive model. For each class label, the number of training examples with positive labeling assignment is generally much less than those with negative labeling assignment. To deal with the class-imbalance issue for multi-label learning, a simple yet effective class-imbalance aware learning strategy called cross-coupling aggregation (COCOA) is proposed in this article. Specifically, COCOA works by leveraging the exploitation of label correlations as well as the exploration of class-imbalance simultaneously. For each class label, a number of multiclass imbalance learners are induced by randomly coupling with other labels, whose predictions on the unseen instance are aggregated to determine the corresponding labeling relevancy. Extensive experiments on 18 benchmark datasets clearly validate the effectiveness of COCOA against state-of-the-art multi-label learning approaches especially in terms of imbalance-specific evaluation metrics.

Many real-world data have multiple class labels known as multi-label data, where the labels are correlated with each other, and as such, they are not independent. Since these data are usually high-dimensional, and the current multi-label feature selection methods have not been precise enough, then a new feature selection method is necessary. In this paper, for the first time, we have modeled the problem of multi-label feature selection to a bipartite graph matching process. The proposed method constructs a bipartite graph of features (as the left vertices) and labels (as the right vertices), called Feature-Label Graph (FLG), where each feature is connected to the set of labels, where the weight of the edge between each feature and label is equal to their correlation. Then, the Hungarian algorithm estimates the best matching in FLG. The selected features in each matching are sorted by weighted correlation distance and added to the ranking vector. To select the discriminative features, the proposed method considers both the redundancy of features and the relevancy of each feature to the class labels. The results indicate the superiority of the proposed method against the other methods in classification measures.

Active semi-supervised learning with multiple complementary information

We propose a rapid adaptive learning framework for streaming object detection, called EER-ASSL. The method combines the expected error reduction (EER) dependent rollback learning and the active semi-supervised learning (ASSL) for a rapid adaptive CNN detector. Most CNN object detectors are built on the assumption of static data distribution. However, images are often noisy and biased, and the data distribution is imbalanced in a real world environment. The proposed method consists of collaborative sampling and EER-ASSL. The EER-ASSL utilizes the active learning (AL) and rollback based semi-supervised learning (SSL). The AL allows us to select more informative and representative samples measuring uncertainty and diversity. The SSL divides the selected streaming image samples into the bins and each bin repeatedly transfers the discriminative knowledge of the EER and CNN models to the next bin until convergence and incorporation with the EER rollback learning algorithm is achieved. The EER models provide a rapid short-term myopic adaptation and the CNN models an incremental long-term performance improvement. EER-ASSL can overcome noisy and biased labels in varying data distribution. Extensive experiments shows that EER-ASSL obtained 70.9 mAP compared to state-of-the-art technology such as Faster RCNN, SSD300, and YOLOv2.

Gathering labeled data is one of the most time-consuming and expensive tasks in supervised machine learning. In practical applications, there are usually quite limited labeled training samples but abundant unlabeled data that is easy to collect. Semi-supervised learning and active learning are two important techniques for learning a discriminative classification model when labeled data is scarce. However, unlabeled data with significant noises and outliers cannot be well exploited and usually worsen the performance of semi-supervised learning and the performance of active learning also needs a powerful initial classifier learned from the quite limited labeled training data. In order to solve the above issues, in this paper we proposed a novel model of semi-supervised dictionary active learning (SSDAL), which aims to integrate semi-supervised learning and active learning to effectively use all the training data. In particular, two criterions based on estimated class possibility are designed to select the unlabeled data with confident class estimation for semi-supervised learning and the informative unlabeled data for active learning, respectively. Extensive experiments are conducted to show the superior performance of our method in classification applications, e.g., handwritten digit recognition, face recognition and large-scale image classification.

In pattern recognition supervised learning algorithms has been used to model the relationships among the characteristics of a group of objects (or patterns) and their class labels. This model is build based on a set of training images that have been labelled by an expert in a specific problem domain. In general, the labelling process consists in defining the object's class label as well as demarcating the precise part of the image in which it is located the object of interest. In many computer vision applications, for example in automatic visual inspection, the labelling process may be a laborious task which requires an extensive work, it could even become impractical when a large number of training images are needed to train the pattern recognition system. This limitation has led to the development of new learning algorithms that allow some ambiguity or "weakness" in the way class labels are assigned. These algorithms are known as weakly supervised learning algorithms. In this paper we present an evaluation of different weakly supervised learning paradigms grouped according to the representation of the object of interest (in one or several characteristic vectors) associated to its class label (single or multiple label). We perform different experiments in the automatic visual inspection context, using a benchmark dataset of texture images with artificial defects. We test different weakly supervised learning algorithms. Results show that representing an object using multiple instances, presents better results in the identification of defective texture image.

Health of people with obesity is a global concern. We developed an explainable sequential deep learning model using nationally representative physical fitness data to predict people with obesity and to identify the most influential predictors. We analyzed data from 204,334 adults collected between 2010 and 2023. A sequential neural network model was trained and evaluated using stratified 5-fold cross-validation. Model performance was assessed using accuracy, precision, recall, F1-score, and ROC-AUC, as well as MAE, MSE, and R². Explainability was examined using SHAP and LIME analyses to rank feature importance and interpret decision patterns. Across five stratified folds, the model achieved an overall accuracy of 0.87-0.88 (p < 0.001 vs. random). Fold 4 showed optimal performance (TN = 1,462; FN = 184; FP = 249; TP = 1,554), yielding an accuracy of 0.873 (precision = 0.866, recall = 0.855, F1 = 0.876, ROC-AUC = 0.95) and stabilizing at 20 epochs. For this model, MAE was 0.122, MSE was 0.041, and R² was 0.833, with an average prediction error of 0.171 for the first 50 samples. SHAP identified 20-m shuttle run estimated VO₂max (importance = 0.339), gender (0.2481), and relative grip strength (0.135) as the top predictors. LIME (intercept = 0.511, predicted=0.668, R² = 0.995) indicated that low estimated VO₂max ( < 28.71 ml/kg/min) and low relative grip strength ( < 38.17%) substantially increased the probability of obesity classification, particularly among females. This explainable sequential deep learning model accurately predicts people with obesity using physical fitness variables and highlights the critical role of cardiorespiratory fitness in obesity risk assessment and management.

Sequential Learning by Touch, Vision, and Audition Christopher M. Conway (cmc82@cornell.edu) Morten H. Christiansen (mhc27@cornell.edu) Department of Psychology Cornell University Ithaca, NY 14853, USA Abstract We investigated the extent to which touch, vision, and audition are similar in the ways they mediate the processing of statistical regularities within sequential input. While previous research has examined statistical/sequential learning in the visual and auditory domains, few researchers have conducted rigorous comparisons across sensory modalities; in particular, the sense of touch has been virtually ignored in such research. Our data reveal commonalities between the ways in which these three modalities afford the learning of sequential information. However, the data also suggest that in terms of sequential learning, audition is superior to the other two senses. We discuss these findings in terms of whether statistical/sequential learning is likely to consist of a single, unitary mechanism or multiple, modality-constrained ones. Introduction The acquisition of statistical/sequential information from the environment appears to be involved in many learning situations, ranging from speech segmentation (Saffran, Newport, & Aslin, 1996), to learning orthographic regularities of written words (Pacton, Perruchet, Fayol, & Cleeremans, 2001) to processing visual scenes (Fiser & Aslin, 2002). However, previous research, focusing exclusively on visual and auditory domains, has failed to investigate whether such learning can occur via touch. Perhaps more importantly, few studies have attempted directly to compare sequential learning as it occurs across the various sensory modalities. There are important reasons to pursue such avenues of study. First, a common assumption is that statistical/sequential learning is a broad, domain- general ability (e.g., Kirkham, Slemmer, & Johnson, 2002). But in order to adequately assess this hypothesis, systematic experimentation across the modalities is necessary. If differences exist between sequential learning in the various senses, it may reflect the operation of multiple mechanisms, rather than a single process. Second, in regards to the touch modality in particular, prior research has generally focused on low- level perception; discovering that the sense of touch can accommodate complex sequential learning may have important implications for tactile communication systems. This paper describes three experiments conducted with the aim to assess sequential learning in three sensory modalities: touch, vision, and audition. Experiment 1 provides the first direct evidence for a fairly complex tactile sequential learning capability. Experiment 2 provides a visual analogue of Experiment 1 and suggests commonalities between visual and tactile sequential learning. Finally, Experiment 3 assesses the auditory domain, revealing an auditory advantage for sequential processing. We conclude by discussing these results in relation to basic issues of cognitive and neural organization—namely, to what extent sequential learning consists of a single or multiple mechanisms. Sequential Learning We define sequential learning as an ability to encode and represent the order of discrete elements occurring in a sequence (Conway & Christiansen, 2001). Importantly, we consider a crucial aspect of sequential learning to be the acquisition of statistical regularities occurring among sequence elements. Artificial grammar learning (AGL; Reber, 1967) is a widely used paradigm for studying such sequential learning 1 . AGL experiments typically use finite-state grammars to generate the stimuli; in such grammars, a transition from one state to the next produces an element of the sequence. For example, in the grammar of Figure 1, the path begins at the left-most node, labeled S1. The next transition can lead to either S 2 or S3. Every time a number is encountered in the transition between states, it is added as the next element of the sequence, producing a sequence corresponding to the rules of the grammar. For example, by passing through the nodes S1, S2, S2, S4, S3, S5, the “legal” sequence 4-1-3-5-2 is generated. During a training phase, participants typically are exposed to a subset of legal sequences—often under the guise of a “memory experiment” or some other such task—with the intent that they will incidentally encode structural aspects of the stimuli. Next, they are tested on whether they can classify novel sequences as In the typical AGL task, the stimulus elements are presented simultaneously (e.g., letter strings)—rather than sequentially (i.e., one element at a time). We consider even the former case to be a sequential learning task because scanning strings of letters generally occurs in a left-to-right, sequential manner. However, our aim here is to create a truly sequential learning environment using temporally-distributed input.

Sequential dictionary learning algorithms have been successfully applied to a number of image processing problems. In a number of these problems however, the data used for dictionary learning are structured matrices with notions of smoothness in the column direction. This prior information which can be traduced as a smoothness constraint on the learned dictionary atoms has not been included in existing dictionary learning algorithms. In this paper, we remedy to this situation by proposing a regularized sequential dictionary learning algorithm. The proposed algorithm differs from the existing ones in their dictionary update stage. The proposed algorithm generates smooth dictionary atoms via the solution of a regularized rank-one matrix approximation problem where regularization is introduced via penalization in the dictionary update stage. Experimental results on synthetic and real data illustrating the performance of the proposed algorithm are provided.

Multi-label classification predicts the existence or non existence of labels of an instance which is correlated with multiple class labels. An efficient multi-label classification system uses a pre-clustered training set for learning a group of classifiers arranged as hierarchical tree to predict the labels of a particular example in the test set. The higher levels of the trees contain relevant labels with more distinguishable label ideas and the labels are transferred into the levels that is lower in position of the hierarchical tree that are difficult to distinguish. The admissible labels are then aggregated from the top level to the bottom level of the hierarchical tree to predict the labels of particular example in the test set. The expirimental study points out the improvement of An efficient system in terms of evaluation measures.

Training with noisy class labels impairs neural networks’ generalization performance. In this context, mixup is a popular regularization technique to improve training robustness by making memorizing false class labels more difficult. However, mixup neglects that multiple annotators, e.g., crowdworkers, typically provide class labels. Therefore, we propose an extension of mixup, which handles multiple class labels per instance while considering which class label originates from which annotator. Integrated into our multi-annotator classification framework annot-mix, it performs superiorly to eleven (mostly state-of-the-art) approaches in an evaluation study with eleven datasets comprising noisy class labels from either human or simulated annotators. Our code is publicly available through our GitHub repository at https://github.com/ies-research/multi-annotator-machine-learning/tree/annot-mix.