Learning Approach Research Articles

A pedagogy of uselessness: challenging solutionism and utility in the Anthropocene through architectural pedagogy

This article discusses an architectural pedagogy of uselessness in relation to the Anthropocene, explicitly building on pedagogical involvement in a design studio at a department for sustainable architecture. The suggestion of a pedagogy of uselessness aims at questioning the criteria and paradigms of utility and solutionism that tend to characterise the discourse on sustainable development, environmental governance, and resilience. To this end, a pedagogy of uselessness moves towards what we have called sustainability otherwise and suggests an orientation which resists the calculability of utility and solutionism. The article is organised in three parts: the first looks at four specific student works that, following the scope of the design studio, question the dominant discourse on sustainability through architecture while addressing territorial issues in several different geographies. The second part focuses on the conceptual references that informed the notion of sustainability otherwise: thinking of modernity as a bewitched time; terrain and landscape as conflict; and weaving as a notion to describe a methodological approach. The third and final part elaborates on the organisation and content of the design studio, in which the article suggests a teaching and learning approach that questions both strictly technical and formal approaches, and offers a provisional and negotiable definition for a pedagogy of uselessness.

Significance of plasma p-tau217 in predicting long-term dementia risk in older community residents: Insights from machine learning approaches.

Whether plasma biomarkers play roles in predicting incident dementia among the general population is worth exploring. A total of 1857 baseline dementia-free older adults with follow-ups up to 13.5 years were included from a community-based cohort. The Recursive Feature Elimination (RFE) algorithm aided in feature selection from 90 candidate predictors to construct logistic regression, naive Bayes, bagged trees, and random forest models. Area under the curve (AUC) was used to assess the model performance for predicting incident dementia. During the follow-up of 12,716 person-years, 207 participants developed dementia. Four predictive models, incorporated plasma p-tau217, age, and scores of MMSE, STICK, and AVLT, exhibited AUCs ranging from 0.79 to 0.96 in testing datasets. These models maintained robustness across various subgroups and sensitivity analyses. Plasma p-tau217 outperforms most traditional variables and may be used to preliminarilyscreen older individuals at high risk of dementia. Plasma p-tau217 showed comparable importance with age and cognitive tests in predicting incident dementia among community older adults. Machine learning models combining plasma p-tau217, age, and cognitive tests exhibited excellent performance in predicting incident dementia. The training models demonstrated robustness in subgroup and sensitivity analysis.

Improving thermodynamic stability of double perovskites with machine learning: The role of cation composition

Perovskite materials have exhibited promising potential in various fields, including solar cells, photodetectors, and light-emitting devices, owing to their exceptional optoelectronic properties and low-cost synthesis methods. In this work, the quest for thermodynamic stability in perovskite materials for optoelectronic applications is addressed through a machine learning (ML) approach. We also analyzed the importance of cations in the stability of double perovskites. It was found that the AUC (area under the curve) and accuracy of the LightGBM model in classification are 0.918 and 0.92, respectively, and the RMSE (root mean square error) and R2-score in regression are 0.119 and 0.872, respectively. Subsequently, the model was used to predict the properties of 10 new perovskites with high accuracy. Finally, the analysis of cationic sites revealed that the stability of the double perovskite is significantly affected by the elements occupying the A, B, and B’ sites. Our results provide an optimal machine learning approach for discerning the thermodynamic stability of double perovskites, which can be a useful guide for tuning the properties of perovskite materials.

Decoding the disciplines: creating mutual understanding in multiprofessional education

Decoding the disciplines as a pedagogical approach to overcome threshold concepts in student learning. The method focuses on a process of making discipline-specific thinking and acting more accessible to students in order to enhance their learning process. Students of any discipline have difficulties with discipline-specific contents, methods or ways of thinking or acting, the so-called bottlenecks. In addition, there are bottlenecks specific to each discipline, such as topics related to scientific work. Findings from a Scholarship of Teaching and Learning approach in a multi-professional group of students in an online seminar session during the corona pandemic are presented. A decoding process on how to find a theoretical approach to one’s research question was conducted. The implementation of the decoding method took place either in face-to-face decoding expert interviews or in a writing process using predetermined guiding questions. Students’ reflections on the use of the method were analyzed using normalization process theory to identify facilitating and inhibiting aspects for the face-to-face and written implementation of the method in interprofessional learning groups. The results show that students could experience a role change or a new role when the decoding topic changes. With the interview method in particular, students can take on the role of experts or novices depending on the topic and thus actively participate. Whether a role change happens or not always depends on the dynamics and the communicative exchange of the group. Decoding takes a lot of time, especially for reflective writing.

Multimodal Religiously Hateful Social Media Memes Classification Based on Textual and Image Data

Multimodal hateful social media meme detection is an important and challenging problem in the vision-language domain. Recent studies show high accuracy for such multimodal tasks due to datasets that provide better joint multimodal embedding to narrow the semantic gap. Religiously hateful meme detection is not extensively explored among published datasets. While there is a need for higher accuracy on religiously hateful memes, deep learning–based models often suffer from inductive bias. This issue is addressed in this work with the following contributions. First, a religiously hateful memes dataset is created and published publicly to advance hateful religious memes detection research. Over 2000 meme images are collected with their corresponding text. The proposed approach compares and fine-tunes VisualBERT pre-trained on the Conceptual Caption (CC) dataset for the downstream classification task. We also extend the dataset with the Facebook hateful memes dataset. We extract visual features using ResNeXT-152 Aggregated Residual Transformations–based Masked Regions with Convolutional Neural Networks (R-CNN) and Bidirectional Encoder Representations from Transformers (BERT) uncased for textual encoding for the early fusion model. We use the primary evaluation metric of an Area Under the Operator Characters Curve (AUROC) to measure model separability. Results show that the proposed approach has a higher AUROC score of 78%, proving the model’s higher separability performance and an accuracy of 70%. It shows comparatively superior performance considering dataset size and against ensemble-based machine learning approaches.

Contrastive machine learning reveals Parkinson’s disease specific features associated with disease severity and progression

Parkinson’s disease (PD) exhibits heterogeneity in terms of symptoms and prognosis, likely due to diverse neuroanatomical alterations. This study employs a contrastive deep learning approach to analyze Magnetic Resonance Imaging (MRI) data from 932 PD patients and 366 controls, aiming to disentangle PD-specific neuroanatomical alterations. The results reveal that these neuroanatomical alterations in PD are correlated with individual differences in dopamine transporter binding deficit, neurodegeneration biomarkers, and clinical severity and progression. The correlation with clinical severity is verified in an external cohort. Notably, certain proteins in the cerebrospinal fluid are strongly associated with PD-specific features, particularly those involved in the immune function. The most notable neuroanatomical alterations are observed in both subcortical and temporal regions. Our findings provide deeper insights into the patterns of brain atrophy in PD and potential underlying molecular mechanisms, paving the way for earlier patient stratification and the development of treatments to slow down neurodegeneration.

Assessing the impact of jigsaw technique for cooperative learning in undergraduate medical education: merits, challenges, and forward prospects

BackgroundJigsaw method is a structured cooperative-learning technique that lays the groundwork towards achieving collective competence, which forms the core of effective clinical practice. It promotes deep learning and effectively enhances team-work among students, hence creating a more inclusive environment.ObjectivePresent study was designed to introduce jigsaw model of cooperative learning to early-year undergraduate medical students, measure its effectiveness on their academic performance, and evaluate the perspectives of both students and faculty members regarding the same.MethodsIt was a mixed method research, involving eighty second-year undergraduate medical students. The jigsaw cooperative learning approach was introduced in two themes within neurosciences module. Students were divided into two equal groups, with one group experiencing typical small-group discussions (SGDs) in first theme and other group exposed to jigsaw approach. The groups were then reversed for second theme. Following the activity, an assessment comprising multiple-choice-questions was conducted to evaluate the impact of jigsaw technique on students’ academic performance, with scores from both groups compared. Student perspectives were gathered through self-designed and validated questionnaire, while faculty perceptions were obtained through focus group discussions. Quantitative data were analyzed using SPSS v22, while thematic analysis was performed for qualitative data.ResultsThe students of jigsaw group displayed significantly higher median assessment score percentage compared to control group (p = 0.003). Moreover, a significantly greater number of students achieved scores ≥ 60% in jigsaw group compared to control group (p = 0.006). The questionnaire responses indicated a favorable perception of this technique among students, in terms of acceptance, positive interdependence, improvement of interpersonal skills, and comparison with typical SGDs. This technique was also well-perceived within the educational context by faculty members.ConclusionThe jigsaw method is associated with higher levels of academic performance among students when compared to typical small-group discussion. The students and faculty perceived this technique to be an effective cooperative learning strategy in terms of enhanced student engagement, active participation, and a sense of inclusivity.

Optimisation of recovery policies in the era of supply chain disruptions: a system dynamics and reinforcement learning approach

Incidents like the COVID-19 pandemic or military conflicts disrupted global supply chains, causing long-lasting shortages in multiple sectors. This so-called ripple effect denotes the propagation of disruptions to further elements of the supply chain. Due to the severity of the impact that the ripple effect has on revenues, service levels, and reputation among supply chain entities, it is essential to understand the related implications. Given the unpredictable nature of disrupting events, this study emphasises the value of a reactive development of effective recovery policies on an operational level. In this article, a system dynamics model for a supply chain is used as framework to investigate the ripple effect. Based on this model, recovery policies are generated using reinforcement learning (RL), which represents a novel approach in this context. As main findings, the experimental results demonstrate the applicability of the proposed approach in mitigating the ripple effect based on secondary data from a major aerospace and defence supply chain and furthermore, the results indicate a broad applicability of the approach without the need for complete information about the disruption characteristics and supply chain entities. With further refinement and real-world implementation, the presented approach provides the potential to enhance supply chain resilience in practice.

Automated remote sleep monitoring needs uncertainty quantification.

Wearable electroencephalography devices emerge as a cost-effective and ergonomic alternative to gold-standard polysomnography, paving the way for better health monitoring and sleep disorder screening. Machine learning allows to automate sleep stage classification, but trust and reliability issues have hampered its adoption in clinical applications. Estimating uncertainty is a crucial factor in enhancing reliability by identifying regions of heightened and diminished confidence. In this study, we used an uncertainty-centred machine learning pipeline, U-PASS, to automate sleep staging in a challenging real-world dataset of single-channel electroencephalography and accelerometry collected with a wearable device from an elderly population. We were able to effectively limit the uncertainty of our machine learning model and to reliably inform clinical experts of which predictions were uncertain to improve the machine learning model's reliability. This increased the five-stage sleep-scoring accuracy of a state-of-the-art machine learning model from 63.9% to 71.2% on our dataset. Remarkably, the machine learning approach outperformed the human expert in interpreting these wearable data. Manual review by sleep specialists, without specific training for sleep staging on wearable electroencephalography, proved ineffective. The clinical utility of this automated remote monitoring system was also demonstrated, establishing a strong correlation between the predicted sleep parameters and the reference polysomnography parameters, and reproducing known correlations with the apnea-hypopnea index. In essence, this work presents a promising avenue to revolutionize remote patient care through the power of machine learning by the use of an automated data-processing pipeline enhanced with uncertainty estimation.

Simple, Efficient, and Scalable Structure-Aware Adapter Boosts Protein Language Models.

Fine-tuning pretrained protein language models (PLMs) has emerged as a prominent strategy for enhancing downstream prediction tasks, often outperforming traditional supervised learning approaches. As a widely applied powerful technique in natural language processing, employing parameter-efficient fine-tuning techniques could potentially enhance the performance of PLMs. However, the direct transfer to life science tasks is nontrivial due to the different training strategies and data forms. To address this gap, we introduce SES-Adapter, a simple, efficient, and scalable adapter method for enhancing the representation learning of PLMs. SES-Adapter incorporates PLM embeddings with structural sequence embeddings to create structure-aware representations. We show that the proposed method is compatible with different PLM architectures and across diverse tasks. Extensive evaluations are conducted on 2 types of folding structures with notable quality differences, 9 state-of-the-art baselines, and 9 benchmark data sets across distinct downstream tasks. Results show that compared to vanilla PLMs, SES-Adapter improves downstream task performance by a maximum of 11% and an average of 3%, with significantly accelerated convergence speed by a maximum of 1034% and an average of 362%, the training efficiency is also improved by approximately 2 times. Moreover, positive optimization is observed even with low-quality predicted structures. The source code for SES-Adapter is available at https://github.com/tyang816/SES-Adapter.

Active learning approaches in molecule pKi prediction.

During the early stages of drug design, identifying compounds with suitable bioactivities is crucial. Given the vast array of potential drug databases, it's feasible to assay only a limited subset of candidates. The optimal method for selecting the candidates, aiming to minimize the overall number of assays, involves an active learning (AL) approach. In this work, we benchmarked a range of AL strategies with two main objectives: (1) to identify a strategy that ensures high model performance and (2) to select molecules with desired properties using minimal assays. To evaluate the different AL strategies, we employed the simulated AL workflow based on "virtual" experiments. These experiments leveraged ChEMBL datasets, which come with known biological activity values for the molecules. Furthermore, for classification tasks, we proposed the hybrid selection strategy that unified both exploration and exploitation AL strategies into a single acquisition function, defined by parameters n and c. We have also shown that popular minimal margin and maximal variance selection approaches for exploration selection correspond to minimization of the hybrid acquisition function with n=1 and 2 respectively. The balance between the exploration and exploitation strategies can be adjusted using a coefficient (c), making the optimal strategy selection straightforward. The primary strength of the hybrid selection method lies in its adaptability; it offers the flexibility to adjust the criteria for molecule selection based on the specific task by modifying the value of the contribution coefficient. Our analysis revealed that, in regression tasks, AL strategies didn't succeed at ensuring high model performance, however, they were successful in selecting molecules with desired properties using minimal number of tests. In analogous experiments in classification tasks, exploration strategy and the hybrid selection function with a constant c<1 (for n=1) and c≤0.2 (for n=2) were effective in achieving the goal of constructing a high-performance predictive model using minimal data. When searching for molecules with desired properties, exploitation, and the hybrid function with c≥1 (n=1) and c≥0.7 (n=2) demonstrated efficiency identifying molecules in fewer iterations compared to random selection method. Notably, when the hybrid function was set to an intermediate coefficient value (c=0.7), it successfully addressed both tasks simultaneously.

Development of Flipped Classroom Learning Design Assisted by Digital Storytelling Media in the Photography Course for Visual Communication Design Students in The Bachelor's Program at STIE Asia Malang

This study aims to develop a flipped classroom instructional design assisted by digital storytelling media for the Photography course for undergraduate students of Visual Communication Design at STIE Asia Malang. The approach of project-based learning integrates the use of technology and active learning methods to enhance the effectiveness of learning. The Research and Development design used the instructional design proposed by Degeng &amp; Degeng (2018). It divided into three phases of instructional design formation, namely 1) analysis of learning conditions, 2) development of learning strategies, 3) development of outcome measurement procedures. The results of the study show that this instructional design is effective in increasing student engagement and active participation. Students are more involved in the learning process, more motivated, and better able to apply the knowledge they have gained in real projects. Additionally, their learning outcomes significantly improved in both conceptual understanding and practical application. The efficiency of this design is also evident from the optimization of learning time, where students can utilize out-of-class time to study the materials, while in-class time is used for more interactive discussions and collaborative projects.

Fusion of machine learning and blockchain-based privacy-preserving approach for healthcare data in the Internet of Things

Fusion of machine learning and blockchain-based privacy-preserving approach for healthcare data in the Internet of Things

Unraveling climate trends in the mediterranean: a hybrid machine learning and statistical approach

AbstractThis study presents a comprehensive spatiotemporal analysis of sea surface temperatures (SST) and surface air temperatures (TAS) across 15 Mediterranean coastal stations, leveraging centennial-scale data to analyze regional climate dynamics. The modeling framework integrates three sequential phases: data preprocessing, statistical analysis, and advanced machine learning techniques, creating a robust analytical pipeline. The data preprocessing phase harmonizes diverse datasets, addresses missing values, and applies transformations to ensure analytical consistency. The statistical modeling employs the Pettitt test for change point detection and linear trend analysis to unveil underlying patterns. The machine learning phase utilizes K-means clustering for climate regime classification and implements tailored Convolutional Neural Networks (CNNs) for cluster-specific future climate anomaly projections. Results unveil a marked anthropogenic climate signal, with contemporary observations consistently surpassing historical baselines. Breakpoint analyses and linear trend assessments reveal heterogeneous climatic shifts, with pronounced warming in the northern Mediterranean. Notably, Nice and Ajaccio exhibit the highest SST increases (0.0119 and 0.0113 °C/decade, respectively), contrasting with more modest trends in Alexandria (0.0052 °C/decade) and Antalya (0.0047 °C/decade) in the eastern Mediterranean. The application of clustering and CNN projections provides granular insights into differential warming trajectories. By 2050, cooler northwestern Mediterranean zones are projected to experience dramatic SST anomalies of approximately 3 °C above the average, with corresponding TAS increases of 2.5 °C. In contrast, warmer eastern and southern regions display more subdued warming patterns, with projected SST and TAS increases of 1.5–2.5 °C by mid-century. This research’s importance is highlighted by its potential to inform tailored adaptation strategies and contribute to the theoretical understanding of climate dynamics, advancing climate modeling and analysis efforts.

Integrated machine learning and geospatial analysis enhanced gully erosion susceptibility modeling in the Erer watershed in Eastern Ethiopia

Land degradation from gully erosion poses a significant threat to the Erer watershed in Eastern Ethiopia, particularly due to agricultural activities and resource exploitation. Identifying erosion-prone areas and underlying factors using advanced machine learning algorithms (MLAs) and geospatial analysis is crucial for addressing this problem and prioritizing adaptive and mitigating strategies. However, previous studies have not leveraged machine learning (ML) and GIS-based approaches to generate susceptibility maps identifying these areas and conditioning factors, hindering sustainable watershed management solutions. This study aimed to predict gully erosion susceptibility (GES) and identify underlying areas and factors in the Erer watershed. Four ML models, namely, XGBoost, random forest (RF), support vector machine (SVM), and artificial neural network (ANN), were integrated with geospatial analysis using 22 geoenvironmental predictors and 1,200 inventory points (70% used for training and 30% for testing). Model performance and robustness were validated through the area under the curve (AUC), accuracy, precision, sensitivity, specificity, kappa coefficient, F1 score, and logarithmic loss. The relative slope position is most influential, with 100% importance in SVM and RF and 95% importance in XGBoost, while annual rainfall (AR) dominated ANN (100% importance). Notably, XGBoost demonstrated robustness and superior prediction/mapping, achieving an AUC of 0.97, 91% accuracy, 92% precision, and 81% kappa while maintaining a low logloss (0.0394). However, SVM excelled in classifying gully resistant/susceptible areas (97% sensitivity, 98% specificity, and 91% F1 score). The ANN model predicted the most areas with very high gully susceptibility (13.74%), followed by the SVM (11.69%), XGBoost (10.65%), and RF (7.85%) models, while XGBoost identified the most areas with very low susceptibility (70.19%). The ensemble technique was employed to further enhance GES modeling, and it outperformed the individual models, achieving an AUC of 0.99, 93.5% accuracy, 92.5% precision, 97.5% sensitivity, 95.4% specificity, 85.8% kappa, and 94.9% F1 score. This technique also classified the GES of the watershed as 36.48% very low, 26.51% low, 16.24% moderate, 11.55% high, and 9.22% very high. Furthermore, district-level analyses revealed the most susceptible areas, including the Babile, Fedis, Harar, and Meyumuluke districts, with high GES areas of 32.4%, 21.3%, 14.3%, and 13.6%, respectively. This study offers robust and flexible ML models with comprehensive validation metrics to enhance GES modeling and identify gully prone areas and factors, thereby supporting decision-making for sustainable watershed conservation and land degradation prevention.

Comparison of Machine Learning Approaches for Prediction of the Equivalent Alkane Carbon Number for Microemulsions Based on Molecular Properties.

The chemical properties of oils are vital in the design of microemulsion systems. The hydrophilic-lipophilic difference equation used to predict microemulsions' phase behavior expresses the oils' physiochemical properties as the equivalent alkane carbon number (EACN). The experimental determination of EACN requires knowledge of the temperature dependence of the microemulsion system and the effects of different surfactant concentrations. Thus, the experimental determination is time-intensive and tedious, requiring days to months for proper separations. Furthermore, the experiments require high purity of chemicals because microemulsions are sensitive to impurities. Our work focuses on the quick and reliable predictions of the EACN with machine learning (ML) models. Due to the immaturity of ML chemical predictions, we compare three graph neural networks (GNNs) and a gradient-boosted tree algorithm, known as XGBoost. The GNNs use the molecular structures represented as simplified molecular-input line-entry system (SMILES) codes for the initial input, which allows us to assess whether geometry optimization is necessary for reliable results. The XGBoost model also begins with the SMILES representations of the molecules but uses molecular descriptors instead of geometry optimizations. The best model tested (crystal graph convolutional neural network with Merck molecular force field-94) has an error of 1.15 EACN units of the true EACN for unknown data with the errors skewed toward zero and an R2 score of 0.9.

Community of Learning as a World-Building Environment

AbstractIn this article, we present what we believe to be the fundamental problem of production systems in Colombia: the domination relationships inherited from colonial structures. We introduce the theoretical foundations of the Community of Learning approach, which we have been developing over the past decades as an option to overcome this issue. We argue that, in the contemporary Latin American context, these relationships are intertwined with the training of individuals for integration into the productive system and its close association with technological tools. This results in the establishment of a domineering sociotechnical system linked to coloniality which some authors associate with modernity. The theoretical foundation of the learning communities approach stems from a philosophical tradition that seeks to examine from a systemic point of view the relationships of domination that produce the alienation of the subject. Our proposal is to establish a pedagogical space for transcending relations of domination, where individuals can explore their identities through interactions with others. In this approach, we aim to foster spaces for collective world-building by affirming one’s own subjectivity within interactions with the environment. We view the technical dimension as inherently linked to these interactions. We term this approach the Community of Learning model, wherein we collectively shape the world through practices that are seen as meaningful actions transforming relationships into learning processes. This approach can be regarded as a means to transcend the domineering aspects of technological practices while concurrently engaging in learning and world-building endeavors. We present, as an example, the results obtained in several cases applying this approach and some proposals that can contribute to effective practices when implementing the Community of Learning approach.

Multistage transfer learning for medical images

Deep learning is revolutionizing various domains and significantly impacting medical image analysis. Despite notable progress, numerous challenges remain, necessitating the refinement of deep learning algorithms for optimal performance in medical image analysis. This paper explores the growing demand for precise and robust medical image analysis by focusing on an advanced deep learning technique, multistage transfer learning. Over the past decade, multistage transfer learning has emerged as a pivotal strategy, particularly in overcoming challenges associated with limited medical data and model generalization. However, the absence of well-compiled literature capturing this development remains a notable gap in the field. This exhaustive investigation endeavors to address this gap by providing a foundational understanding of how multistage transfer learning approaches confront the unique challenges posed by insufficient medical image datasets. The paper offers a detailed analysis of various multistage transfer learning types, architectures, methodologies, and strategies deployed in medical image analysis. Additionally, it delves into intrinsic challenges within this framework, providing a comprehensive overview of the current state while outlining potential directions for advancing methodologies in future research. This paper underscores the transformative potential of multistage transfer learning in medical image analysis, providing valuable guidance to researchers and healthcare professionals.

Peer-assisted learning augmented by peer counseling to foster academic and personal development in flipped classroom

AbstractIn recent decades, under the influence of technological advancements, evolving student demographics, and shifting societal demands, the landscape of higher education has undergone significant transformations. Among numerous learning approaches, Flipped Learning (FL) has gained particular prominence due to its ability to enhance students’ performance through active engagement and self-directed learning. Another teaching and learning linked to indicators of students’ academic achievements is Peer-Assisted Learning (PAL). However, previous research on PAL remains scarce. To fill this gap in the literature, the effects of the integration of Peer Counseling (PC) activity into PAL on both academic and personal development within the flipped classroom setting over time were investigated in the present study. The study participants were counseling students (n = 91), including 60 students in the PAL-integrated flipped classroom and 31 in traditional classroom used as the control group. The results revealed that, over a 14-week period, PAL-integrated flipped classroom led to improvement in students’ Solution-Focused Thinking (SFT) skills, self-efficacy, and effective counselor characteristics, with medium to large effect sizes. The results also indicated the positive effect of PC on students’ counseling performance and achievement scores. These findings offer important implications for creating a collaborative learning environment. The paper concludes with a discussion of challenges typically associated with the implementation of flipped classroom in higher education.

Learning to Allocate Time-Bound and Dynamic Tasks to Multiple Robots Using Covariant Attention Neural Networks

Abstract In various applications of multi-robotics in disaster response, warehouse management, and manufacturing, tasks that are known a priori and tasks added during run time need to be assigned efficiently and without conflicts to robots in the team. This multi-robot task allocation (MRTA) process presents itself as a combinatorial optimization (CO) problem that is usually challenging to be solved in meaningful timescales using typical (mixed)integer (non)linear programming tools. Building on a growing body of work in using graph reinforcement learning to learn search heuristics for such complex CO problems, this paper presents a new graph neural network architecture called the covariant attention mechanism (CAM). CAM can not only generalize but also scale to larger problems than that encountered in training, and handle dynamic tasks. This architecture combines the concept of covariant compositional networks used here to embed the local structures in task graphs, with a context module that encodes the robots’ states. The encoded information is passed onto a decoder designed using multi-head attention mechanism. When applied to a class of MRTA problems with time deadlines, robot ferry range constraints, and multi-trip settings, CAM surpasses a state-of-the-art graph learning approach based on the attention mechanism, as well as a feasible random-walk baseline across various generalizability and scalability tests. Performance of CAM is also found to be at par with a high-performing non-learning baseline called BiG-MRTA, while noting up to a 70-fold improvement in decision-making efficiency over this baseline.