Sequence Learning Research Articles

Web service embedding: Representing the invocation association between services with practical-valued vectors

Service representation using specified methods to express the service’s functionalities and non-functionalities in a machine-understandable format is crucial in service composition. In representing non-functionalities, existing approaches focus on QoS (Quality of Service) or sociability. The invocation association between services has not been considered in service representation. However, as the invocation association reflects the actual invocation situation on service execution, the inconsideration will increase the failure of service composition. Thus, finding an appropriate method to represent the invocation association is a significant issue in service composition. This paper proposes a novel service representation method called web service embedding (WSE) to generate a practical-valued vector representation using deep learning sequence models to address the challenge. First, we propose utilizing deep learning sequence models to capture the invocation association from service composition/invocation sequences. Next, the representation vectors of services can be generated using the pre-trained models by designing an unsupervised training solution. Then, two presented metrics, context invocation relevance, and semantic invocation relevance, are used to measure how well the embedding vectors of the models express the invocation association for clustering in service composition. Finally, various neural sequence architectures have been inspected to find suitable models for WSE. The experimental results show that our proposed WSE can successfully perform service representation for clustering based on the invocation association.

Kinesthetic vs. visual focus: No evidence for effects of practice modality in representation types after action imagery practice and action execution practice

Action-imagery practice (AIP) is assumed to result in partly different action representations than action-execution practice (AEP). The present study investigated whether focusing on either kinesthetic or visual aspects of a task during practice amplifies or diminishes such differences between AIP and AEP. In ten sessions, four groups, using either AIP or AEP with either kinesthetic or visual focus, practiced a twelve-element sequence in a unimanual serial reaction time task. Tests involved the practice sequence, a mirror sequence, and a different sequence, each performed with the practice and transfer hand. In AIP and AEP, in both hands, reaction times (RTs) were shorter in the practice sequence than in the different sequence, indicating effector-independent visual-spatial sequence representations. Further, RTs were shorter in the practice hand than in the transfer hand in the practice sequence (but not in the different sequence), indicating effector-dependent representations in AEP and AIP. Although the representation types did not differ, learning effects were stronger in AEP than in AIP. Thus, although to a lower extent than in AEP, effector-dependent representations can be acquired using AIP. Contrary to the expectations, the focus manipulation did not have an impact on the acquired representation types. Hence, modality instructions in AIP may not have such a strong impact as commonly assumed, at least in implicit sequence learning.

Modulating Visuomotor Sequence Learning by Repetitive Transcranial Magnetic Stimulation: What Do We Know So Far?

Predictive processes and numerous cognitive, motor, and social skills depend heavily on sequence learning. The visuomotor Serial Reaction Time Task (SRTT) can measure this fundamental cognitive process. To comprehend the neural underpinnings of the SRTT, non-invasive brain stimulation stands out as one of the most effective methodologies. Nevertheless, a systematic list of considerations for the design of such interventional studies is currently lacking. To address this gap, this review aimed to investigate whether repetitive transcranial magnetic stimulation (rTMS) is a viable method of modulating visuomotor sequence learning and to identify the factors that mediate its efficacy. We systematically analyzed the eligible records (n = 17) that attempted to modulate the performance of the SRTT with rTMS. The purpose of the analysis was to determine how the following factors affected SRTT performance: (1) stimulated brain regions, (2) rTMS protocols, (3) stimulated hemisphere, (4) timing of the stimulation, (5) SRTT sequence properties, and (6) other methodological features. The primary motor cortex (M1) and the dorsolateral prefrontal cortex (DLPFC) were found to be the most promising stimulation targets. Low-frequency protocols over M1 usually weaken performance, but the results are less consistent for the DLPFC. This review provides a comprehensive discussion about the behavioral effects of six factors that are crucial in designing future studies to modulate sequence learning with rTMS. Future studies may preferentially and synergistically combine functional neuroimaging with rTMS to adequately link the rTMS-induced network effects with behavioral findings, which are crucial to develop a unified cognitive model of visuomotor sequence learning.

A dual-branch siamese spatial-spectral transformer attention network for Hyperspectral Image Change Detection

The convolutional neural networks have recently gained widespread attention in Hyperspectral Image Change Detection (HSI-CD) due to their outstanding feature extraction ability. However, limited by the inherent network backbones, the convolutional neural networks (CNNs) fail to mine the sequence attributes and model the intricate relationships of spectral signatures. In contrast, transformers are proficient at learning sequence information owing to the powerful self-attention mechanisms. The two backbone structures exhibit complementary spatial and spectral feature extraction strengths, respectively. Inspired by this, we propose a dual-branch siamese spatial–spectral transformer attention network (DBS3TAN) for HSI-CD. The main idea is to fully exploit the advantages of CNNs and transformers for spatial and spectral feature extraction. More importantly, we devise the two key modules, i.e., the spatial attention module and the spatial–spectral transformer module. The former utilizes depthwise separable convolutions and attention mechanisms to emphasize the features of dual-temporal HSIs from the spatial perspective. The latter focuses on the sequence attributes of spectral signatures and mines the spatial characteristics from adjacent pixels. We employ the weighted contrastive loss function to separate the changed and unchanged pixels more reliably and set the random weight factors to balance the contributions of the two branches. Finally, the threshold values judgment is used to obtain the ultimate detection maps. We conduct extensive experiments to evaluate the DBS3TAN on three HSI datasets, demonstrating its superior performances than compared methods qualitatively and quantitatively. The source code will be available at https://github.com/zhangyiyan001/DBS3TAN.

Ensemble learning-based approach for automatic classification of termite mushrooms.

Termite mushrooms are edible fungi that provide significant economic, nutritional, and medicinal value. However, identifying these mushroom species based on morphology and traditional knowledge is ineffective due to their short development time and seasonal nature. This study proposes a novel method for classifying termite mushroom species. The method utilizes Gradient Boosting machine learning techniques and sequence encoding on the Internal Transcribed Spacer (ITS) gene dataset to construct a machine learning model for identifying termite mushroom species. The model is trained using ITS sequences obtained from the National Center for Biotechnology Information (NCBI) and the Barcode of Life Data Systems (BOLD). Ensemble learning techniques are applied to classify termite mushroom species. The proposed model achieves good results on the test dataset, with an accuracy of 0.91 and an average AUCROC value of 0.99. To validate the model, eight ITS sequences collected from termite mushroom samples in An Linh commune, Phu Giao district, Binh Duong province, Vietnam were used as the test data. The results show consistent species identification with predictions from the NCBI BLAST software. The results of species identification were consistent with the NCBI BLAST prediction software. This machine-learning model shows promise as an automatic solution for classifying termite mushroom species. It can help researchers better understand the local growth of these termite mushrooms and develop conservation plans for this rare and valuable plant resource.

Physics-constrained sequence learning with attention mechanism for multi-horizon production forecasting

Production forecasting of oil and gas energy is a crucial task for reservoir management as it provides a fundamental basis for optimizing development plans and determining investment decisions. Recently, machine learning-based production forecasting methods are proliferating due to their powerful non-linear fitting capability. However, most existing methods lack direct consideration of the engineering physical processes, which may lead to production forecasts inconsistent with engineering science. In this paper, we propose a novel physics-constrained sequence learning method with attention mechanism for multi-horizon production forecasting. The proposed method is built upon on a backbone sequence-to-sequence neural network composed of gated recurrent unit (GRU) cells, which aims to learn the underlying decline behavior from historical production. An attention mechanism is introduced on the backbone network to improve the information utilization of historical production. Furthermore, known engineering parameters are incorporated into the decoder as physical constraints, enabling the network to take into consideration artificially controllable variables in addition to natural decline tendency. The proposed network is trained with a scheduled sampling strategy, which randomly allows the GRU decoder to access the true production from the previous step, thereby improving the robustness of the model to prediction errors. The proposed model is evaluated on the Volve public dataset. Experimental results validate the significant advantage of our method over ten baselines in multi-horizon production forecasting. In addition, ablation study from multiple aspects verifies the contributions of key components of our model. Our code is released at https://github.com/changji95/PAPF for academic use.

Lithium-ion battery state of health estimation using a hybrid model based on a convolutional neural network and bidirectional gated recurrent unit

This paper proposes a real-time state of health (SOH) estimation model based on a deep learning (DL) framework. The proposed model is a combination of two different architectures; a one-dimensional convolutional neural network (1D-CNN) and a bidirectional gated recurrent unit (BiGRU). The hybrid CNN-BiGRU uses the 1D CNN layers to extract pertinent features from input data and then relies on the BiGRU layers for sequence learning in both directions. To account for all SOH indicators, the proposed approach uses the current, voltage, and temperature measurements, which are readily obtainable from the electric vehicle's battery management system (BMS). This prevents the complex and time-consuming feature extraction used in most related papers. Since the hyperparameters have a significant impact on the performance of neural network models, a Bayesian Optimization (BO) technique based on the Gaussian Process (GP) was considered to tune the CNN-BiGRU model hyperparameters. Accordingly, the objective function was able to converge to a low Mean Squared Error (MSE) of 1.2×10−5 in just 19 iterations. Afterward, to verify the accuracy of the optimized model, a Lithium-ion battery dataset with several discharge profiles provided by the National Aeronautics and Space Administration (NASA) was used. The obtained results demonstrated the accuracy and robustness of the proposed method compared to other commonly used models. The CNN-BiGRU model yielded a Mean Absolute Error (MAE) of 2.080% and a root-mean-square error (RMSE) of 2.516% in the case of the battery set #C, referring to a set with 70 cycles already used at 24 °C. Additionally, the End of life (EOL) indicator error of zero cycles for the same data.

Brain plasticity underlying sleep-dependent motor consolidation after motor imagery.

Motor imagery can, similarly to physical practice, improve motor performance through experience-based plasticity. Using magnetoencephalography, we investigated changes in brain activity associated with offline consolidation of motor sequence learning through physical practice or motor imagery. After an initial training session with either physical practice or motor imagery, participants underwent overnight consolidation. As control condition, participants underwent wake-related consolidation after training with motor imagery. Behavioral analyses revealed that overnight consolidation of motor learning through motor imagery outperformed wake-related consolidation (95% CI [0.02, 0.07], P < 0.001, RP2 = 0.05). As regions of interest, we selected the generators of event-related synchronization/desynchronization of alpha (8-12Hz) and beta (15-30Hz) oscillations, which predicted the level of performance on the motor sequence. This yielded a primary sensorimotor-premotor network for alpha oscillations and a cortico-cerebellar network for beta oscillations. The alpha network exhibited increased neural desynchronization after overnight consolidation compared to wake-related consolidation. By contrast, the beta network exhibited an increase in neural synchronization after wake-related consolidation compared to overnight consolidation. We provide the first evidence of parallel brain plasticity underlying behavioral changes associated with sleep-dependent consolidation of motor skill learning through motor imagery and physical practice.

Reliability of on-line visual feedback influences learning of continuous motor task of healthy young adults.

A continuous task was used to determine how the reliability of on-line visual feedback during acquisition impacts motor learning. Participants performed a right hand pointing task of a repeated sequence with a visual cursor that was either reliable, moderately unreliable, or largely unreliable. Delayed retention tests were administered 24 h later, as well as intermanual transfer tests (performed with the left hand). A visuospatial transfer test was performed with the same targets' sequence (same visuospatial configuration) while a motor transfer test was performed with the visual mirror of the targets' sequence (same motor patterns). Results showed that pointing was slower and long-term learning disrupted in the largely unreliable visual cursor condition, compared with the reliable and moderately unreliable conditions. Also, analysis of transfers revealed classically better performance on visuospatial transfer than on motor transfer for the reliable condition. However, here we first show that such difference disappears when the cursor was moderately or largely unreliable. Interestingly, these results indicated a difference in the type of sequence coding, depending on the reliability of the on-line visual feedback. This recourse to mixed coding opens up interesting perspectives, as it is known to promote better learning of motor sequences.

Generalization of procedural motor sequence learning after a single practice trial

When humans begin learning new motor skills, they typically display early rapid performance improvements. It is not well understood how knowledge acquired during this early skill learning period generalizes to new, related skills. Here, we addressed this question by investigating factors influencing generalization of early learning from a skill A to a different, but related skill B. Early skill generalization was tested over four experiments (N = 2095). Subjects successively learned two related motor sequence skills (skills A and B) over different practice schedules. Skill A and B sequences shared ordinal (i.e., matching keypress locations), transitional (i.e., ordered keypress pairs), parsing rule (i.e., distinct sequence events like repeated keypresses that can be used as a breakpoint for segmenting the sequence into smaller units) structures, or possessed no structure similarities. Results showed generalization for shared parsing rule structure between skills A and B after only a single 10-second practice trial of skill A. Manipulating the initial practice exposure to skill A (1 to 12 trials) and inter-practice rest interval (0–30 s) between skills A and B had no impact on parsing rule structure generalization. Furthermore, this generalization was not explained by stronger sensorimotor mapping between individual keypress actions and their symbolic representations. In contrast, learning from skill A did not generalize to skill B during early learning when the sequences shared only ordinal or transitional structure features. These results document sequence structure that can be very rapidly generalized during initial learning to facilitate generalization of skill.

Down regulation of Cathepsin W is associated with poor prognosis in pancreatic cancer

Pancreatic ductal adenocarcinoma (PDAC) is associated with a very poor prognosis. Therefore, there has been a focus on identifying new biomarkers for its early diagnosis and the prediction of patient survival. Genome-wide RNA and microRNA sequencing, bioinformatics and Machine Learning approaches to identify differentially expressed genes (DEGs), followed by validation in an additional cohort of PDAC patients has been undertaken. To identify DEGs, genome RNA sequencing and clinical data from pancreatic cancer patients were extracted from The Cancer Genome Atlas Database (TCGA). We used Kaplan–Meier analysis of survival curves was used to assess prognostic biomarkers. Ensemble learning, Random Forest (RF), Max Voting, Adaboost, Gradient boosting machines (GBM), and Extreme Gradient Boosting (XGB) techniques were used, and Gradient boosting machines (GBM) were selected with 100% accuracy for analysis. Moreover, protein–protein interaction (PPI), molecular pathways, concomitant expression of DEGs, and correlations between DEGs and clinical data were analyzed. We have evaluated candidate genes, miRNAs, and a combination of these obtained from machine learning algorithms and survival analysis. The results of Machine learning identified 23 genes with negative regulation, five genes with positive regulation, seven microRNAs with negative regulation, and 20 microRNAs with positive regulation in PDAC. Key genes BMF, FRMD4A, ADAP2, PPP1R17, and CACNG3 had the highest coefficient in the advanced stages of the disease. In addition, the survival analysis showed decreased expression of hsa.miR.642a, hsa.mir.363, CD22, BTNL9, and CTSW and overexpression of hsa.miR.153.1, hsa.miR.539, hsa.miR.412 reduced survival rate. CTSW was identified as a novel genetic marker and this was validated using RT-PCR. Machine learning algorithms may be used to Identify key dysregulated genes/miRNAs involved in the disease pathogenesis can be used to detect patients in earlier stages. Our data also demonstrated the prognostic and diagnostic value of CTSW in PDAC.

Targeted memory reactivation during post-learning sleep does not enhance motor memory consolidation in older adults.

Targeted memory reactivation (TMR) during sleep enhances memory consolidation in young adults by modulating electrophysiological markers of neuroplasticity. Interestingly, older adults exhibit deficits in motor memory consolidation, an impairment that has been linked to age-related degradations in the same sleep features sensitive to TMR. We hypothesised that TMR would enhance consolidation in older adults via the modulation of these markers. A total of 17 older participants were trained on a motor task involving two auditory-cued sequences. During a post-learning nap, two auditory cues were played: one associated to a learned (i.e., reactivated) sequence and one control. Performance during two delayed re-tests did not differ between reactivated and non-reactivated sequences. Moreover, both associated and control sounds modulated brain responses, yet there were no consistent differences between the auditory cue types. Our results collectively demonstrate that older adults do not benefit from specific reactivation of a motor memory trace by an associated auditory cue during post-learning sleep. Based on previous research, it is possible that auditory stimulation during post-learning sleep could have boosted motor memory consolidation in a non-specific manner.

A deep learning sequence model based on self-attention and convolution for wind power prediction

Renewable energy has garnered significant attention recently due to its sustainable nature and minimal environmental footprint. Among various sources, wind energy emerges as one of the most promising. However, its inherently unpredictable and irregular characteristics pose challenges to forecasting wind power generation. This study introduces a wind power prediction model that employs self-attention to capture long-range relationships and convolutional layers to understand the local temporal dynamics within time-series data. Unlike traditional deep learning sequence models, such as the recurrent neural network (RNN), long short-term memory (LSTM), and gated recurrent unit (GRU), our method adeptly integrates both global and local insights. We validate the model’s efficacy through experiments on three datasets. The results consistently show our model’s superior performance over alternative methods. Further, we conduct comprehensive experiments to analyze our proposed model.

Research on online and offline double-loop teaching of college students’ mental health education based on cluster analysis algorithm

Abstract In this paper, a knowledge difficulty clustering algorithm (MIBKPC) based on multidimensional time-series data and learning path networks is proposed. The algorithm integrates the cyclic learning sequence, forgetting behavior and system interaction degree formed by learners learning some of the more difficult knowledge points, and constructs a computational model for the difficulty similarity of knowledge points based on system interaction behavior so as to better portray the difficulty similarity of knowledge points based on system interaction behavior. Then a knowledge point difficulty clustering algorithm (MFSKPC) based on multidimensional time-series data and maximum frequent subgraphs is proposed. The algorithm extracts the maximum frequent subgraphs of the atlas based on the learner’s directed learning path atlas using the gSpan-based maximum frequent subgraph mining algorithm and portrays the difficulty similarity of knowledge points based on system interaction behaviors by combining the system interaction degree and the maximum frequent subgraphs. In the accuracy comparison, the clustering accuracy of the MFSKPC algorithm for advanced learners was higher than that of the MIBKPC algorithm for both K=3 and K =5 conditions by 7.89% and 8.29%, respectively. In the comparative analysis of double-loop psychological instruction, the experimental class improved its pre and post-test scores by an average of 2.38 points, while the control group improved by an average of only 1.88 points. The experiment showed that the double-loop teaching based on the MFSKPC algorithm was more effective.

Timed Sequence Task: A New Paradigm to Study Motor Learning and Flexibility in Mice.

Motor learning and flexibility allow animals to perform routine actions efficiently while keeping them flexible. A number of paradigms are used to test cognitive flexibility, but not many of them focus specifically on the learning of complex motor sequences and their flexibility. While many tests use operant or touchscreen boxes that offer high throughput and reproducibility, the motor actions themselves are mostly simple presses of a designated lever. To focus more on motor actions during the operant task and to probe the flexibility of these well trained actions, we developed a new operant paradigm for mice, the "timed sequence task." The task requires mice to learn a sequence of lever presses that have to be emitted in precisely defined time limits. After training, the required pressing sequence and/or timing of individual presses is modified to test the ability of mice to alter their previously trained motor actions. We provide a code for the new protocol that can be used and adapted to common types of operant boxes. In addition, we provide a set of scripts that allow automatic extraction and analysis of numerous parameters recorded during each session. We demonstrate that the analysis of multiple performance parameters is necessary for detailed insight into the behavior of animals during the task. We validate our paradigm in an experiment using the valproate model of autism as a model of cognitive inflexibility. We show that the valproate mice show superior performance at specific stages of the task, paradoxically because of their propensity to more stereotypic behavior.

STTRE: A Spatio-Temporal Transformer with Relative Embeddings for multivariate time series forecasting

The prevalence of multivariate time series data across several disciplines fosters a demand and, subsequently, significant growth in the research and advancement of multivariate time series analysis. Drawing inspiration from a popular natural language processing model, the Transformer, we propose the Spatio-Temporal Transformer with Relative Embeddings (STTRE) to address multivariate time series forecasting. This work primarily focuses on developing a Transformer-based framework that can fully exploit the spatio-temporal nature of a multivariate time series by incorporating several of the Transformer’s key components, but with augmentations that allow them to excel in multivariate time series forecasting. Current Transformer-based models for multivariate time series often neglect the data’s spatial component(s) and utilize absolute position embeddings as their only means to detect the data’s temporal component(s), which we show is flawed for time series applications. The lack of emphasis on fully exploiting the spatio-temporality of the data can incur subpar results in terms of accuracy. We redesign relative position representations, which we rename to relative embeddings, to unveil a new method for detecting latent spatial, temporal, and spatio-temporal dependencies more effectively than previous Transformer-based models. We couple these relative embeddings with a restructuring of the Transformer’s primary sequence learning mechanism, multi-head attention, in a way that allows for full utilization of relative embeddings, thus achieving up to a 24% improvement in accuracy over other state-of-the-art multivariate time series models on a comprehensive selection of publicly available multivariate time series forecasting datasets.

A Deep Learning Framework With Visualisation for Uncovering Students’ Learning Progression and Learning Bottlenecks

Educational process mining aims (EPM) to help teachers understand the overall learning process of their students. Although deep learning models have shown promising results in many domains, the event log dataset in many online courses may not be large enough for deep learning models to approximate the probability distribution of students’ learning sequence due to a lack of participants. This study proposes a deep learning framework to help uncover the learning progression of learners. It aims to produce a graphical representation of the overall educational process from event logs. Our framework adopts the Smith–Waterman algorithm from the bioinformatics field to evaluate general learning sequences generated from deep learning models. Using our framework, we compare the performance of a deep learning model with the modified cross-attention layer and a model without modification and find that the modified model outperforms the other. The contribution of this framework is that it enables the use of neural architecture search techniques to uncover students’ general learning sequence irrespective of the dataset’s size. The framework also helps educators identify education materials that present as learning bottlenecks, enabling them to improve the materials and their respective layout order, thus facilitating student learning.

Environment and taxonomy shape the genomic signature of prokaryotic extremophiles

This study provides comprehensive quantitative evidence suggesting that adaptations to extreme temperatures and pH imprint a discernible environmental component in the genomic signature of microbial extremophiles. Both supervised and unsupervised machine learning algorithms were used to analyze genomic signatures, each computed as the k-mer frequency vector of a 500 kbp DNA fragment arbitrarily selected to represent a genome. Computational experiments classified/clustered genomic signatures extracted from a curated dataset of sim 700 extremophile (temperature, pH) bacteria and archaea genomes, at multiple scales of analysis, 1le k le 6. The supervised learning resulted in high accuracies for taxonomic classifications at 2le k le 6, and medium to medium-high accuracies for environment category classifications of the same datasets at 3le k le 6. For k=3, our findings were largely consistent with amino acid compositional biases and codon usage patterns in coding regions, previously attributed to extreme environment adaptations. The unsupervised learning of unlabelled sequences identified several exemplars of hyperthermophilic organisms with large similarities in their genomic signatures, in spite of belonging to different domains in the Tree of Life.

Stimulus-locked auditory information facilitates real-time visuo-motor sequence learning

Prior research investigating whether and how multisensory information facilitates skill learning is quite mixed; whereas some research points to congruent information improving learning, other work suggests that people become reliant on the redundant information, such that its removal ultimately detracts from the ability to perform a unisensory task. We examined this question using the Serial Interception Sequence Learning (SISL) task, a visuo-motor paradigm in which participants implicitly learn a sequence embedded in noise. We investigated whether adding auditory information in different ways would enhance real time sequence learning and whether any benefits of multisensory learning would persist with visual-only testing. Auditory information was used either as feedback on the visuo-motor task (Experiments 1 and 2) or was presented synchronously with visual information during learning (Experiment 3). Robust sequence-specific performance advantages occurred across conditions and experiments; however, auditory information enhanced real-time performance only when it was synchronized with visual information. Participants were significantly more accurate, faster, and more precise with stimulus-locked auditory information during training. Notably, these benefits did not generalize to the visual-only context, suggesting that the benefits of stimulus-locked auditory information are primarily useful only when the perceptual information is present.

Auditory learning of recurrent tone sequences is present in the newborn's brain

The seemingly effortless ability of our auditory system to rapidly detect new events in a dynamic environment is crucial for survival. Whether the underlying brain processes are innate is unknown. To answer this question, electroencephalography was recorded while regularly patterned (REG) versus random (RAND) tone sequences were presented to sleeping neonates. Regular relative to random sequences elicited differential neural responses after only a single repetition of the pattern indicating the existence of an innate capacity of the auditory system to detect auditory sequential regularities. We show that the newborn auditory system accumulates evidence only somewhat longer than the minimum amount determined by the ideal Bayesian observer model (the prediction from a variable-order Markov chain model) before detecting a repeating pattern. Thus, newborns can quickly form representations for regular features of the sound input, preparing the way for learning the contingencies of the environment.