This paper proposes a new unsupervised domain adaptation approach called Collaborative and Adversarial Network (CAN), which uses the domain-collaborative and domain-adversarial learning strategy for training the neural network. The domain-collaborative learning aims to learn domain-specific feature representation to preserve the discriminability for the target domain, while the domain adversarial learning aims to learn domain-invariant feature representation to reduce the domain distribution mismatch between the source and target domains. We show that these two learning strategies can be uniformly formulated as domain classifier learning with positive or negative weights on the losses. We then design a collaborative and adversarial training scheme, which automatically learns domain-specific representations from lower blocks in CNNs through collaborative learning and domain-invariant representations from higher blocks through adversarial learning. Moreover, to further enhance the discriminability in the target domain, we propose Self-Paced CAN (SPCAN), which progressively selects pseudo-labeled target samples for re-training the classifiers. We employ a self-paced learning strategy to select pseudo-labeled target samples in an easy-to-hard fashion. Comprehensive experiments on different benchmark datasets, Office-31, ImageCLEF-DA, and VISDA-2017 for the object recognition task, and UCF101-10 and HMDB51-10 for the video action recognition task, show our newly proposed approaches achieve the state-of-the-art performance, which clearly demonstrates the effectiveness of our proposed approaches for unsupervised domain adaptation.

Large Language Models (LLMs) such as OpenAI's GPT-4 and Meta's LLaMA offer a promising approach for scalable personality assessment from open-ended language. However, inferring personality traits remains challenging, and earlier work often relied on synthetic data or social media text lacking psychometric validity. We introduce a real-world benchmark of 555 semi-structured interviews with BFI-10 self-report scores for evaluating LLM-based personality inference. Three state-of-the-art LLMs (GPT-4.1 Mini, Meta-LLaMA, and DeepSeek) were tested using zero-shot prompting for BFI-10 item prediction and both zero-shot and chain-of-thought prompting for Big Five trait inference. All models showed high test-retest reliability, but construct validity was limited: correlations with ground-truth scores were weak (max Pearson's r = 0.27), interrater agreement was low (Cohen's κ< 0.10), and predictions were biased toward moderate or high trait levels. Chain-of-thought prompting and longer input context modestly improved distributional alignment, but not trait-level accuracy. These results underscore limitations in current LLM-based personality inference and highlight the need for evidence-based development for psychological applications.

Accurately annotating and controlling protein function from sequence data remains a major challenge, particularly within homologous families where annotated sequences are scarce and structural variation is minimal. We present a two-stage approach for semi-supervised functional annotation and conditional sequence generation in protein families using representation learning. First, we demonstrate that protein language models, pretrained on large and diverse sequence datasets and possibly finetuned via contrastive learning, provide embeddings that robustly capture fine-grained functional specificities, even with limited labeled data. Second, we use the inferred annotations to train a generative probabilistic model, an annotation-aware Restricted Boltzmann Machine, capable of producing synthetic sequences with prescribed functional labels. Across several protein families, we show that this approach achieves highly accurate annotation quality and supports the generation of functionally coherent sequences. Our findings underscore the power of combining self-supervised learning with light supervision to overcome data scarcity in protein function prediction and design.

Introduction:Motor imagery-based brain-computer interfaces (BCIs) are a technique for decoding and classifying the intention of motor execution, solely based on imagined (rather than executed) movements. Although deep learning techniques have increased the potential of BCIs, the complexity of decoding unilateral upper limb motor imagery remains challenging. To understand whether neurophysiological features, which are directly related to neural mechanisms of motor imagery, might influence classification accuracy, most studies have largely leveraged traditional machine learning frameworks, leaving deep learning-based techniques underexplored. Methods:In this work, three different deep learning models from the literature (EEGNet, FBCNet, NFEEG) and two common spatial pattern-based machine learning classifiers (SVM, LDA) were used to classify imagined right elbow flexion and extension from participants using electroencephalography data. From two recorded resting states (eyes-open, eyes-closed), absolute and relative alpha and beta power of the frontal, fronto-central and central electrodes were used to predict the accuracy of the different classifiers. Results:The prediction of classifier accuracies by neurophysiological features revealed negative correlations between the relative alpha band and classifier accuracies and positive correlations between the absolute and relative beta band and classifiers accuracies. Most ipsilateral EEG channels yielded significant correlations with classifier accuracies, especially for the machine learning classifier. Discussion:This pattern contrasts with previous findings from bilateral MI paradigms, where contralateral alpha and beta activity were more influential. These inverted correlations suggest task-specific neurophysiological mechanisms in unilateral MI, emphasizing the role of ipsilateral inhibition and attentional processes.

Federated learning (FL) is a recent technique that emerged to handle the vast amount of training data needed in machine learning algorithms while fulfilling data owners’ privacy challenges in such scenarios. Simultaneously, the field of quantum computing (QC), using quantum properties such as entanglement and superposition to perform computation, has experienced exponential growth, theoretically proving to be more efficient in specific machine learning tasks and creating the discipline known as quantum machine learning (QML). Thus, an emerging body of knowledge has started studying the combination of these two research agendas, giving rise to the field of quantum federated learning (QFL). In this review, we systematically classify the existing literature through a novel taxonomy, identify current trends and challenges, and highlight research gaps and future directions to support the continued development of this emerging field.

Multilingual Large Language Models(MLLMs) demonstrate strong generalization across languages, yet they remain prone to hallucinations, especially in low-resource languages, due to training data imbalances. These hallucinations, which include inaccurate or fabricated outputs, are particularly problematic in domain-specific generation tasks (Chataigner et al., 2024). To address this challenge, we propose CCL-XCoT(Curriculum-based Contrastive Learning-based Cross-lingual Chain-of-Thought), a two-stage fine-tuning framework for mitigating hallucination in MLLMs. Our approach first enhances cross-lingual semantic alignment through curriculum-based contrastive learning combined with next-token prediction during continued pre-training. Building on this foundation, we then introduce a cross-lingual Chain-of-Thought (XCoT) prompting strategy during instruction fine-tuning, which guides the model to reason in a high-resource language before generating answers in the target low-resource language. Experimental results show that CCL-XCoT reduces hallucination rates by up to 62

IntroductionMachine learning courses usually focus on getting students prepared to apply various models in real-world settings, but much less attention is given to teaching students the various techniques to explain a model's decision-making process. This gap is particularly concerning given the increasing deployment of AI systems in high-stakes domains where interpretability is crucial for trust, regulatory compliance, and ethical decision-making. Despite the growing importance of explainable AI (XAI) in professional practice, systematic pedagogical approaches for teaching these techniques remain underdeveloped.MethodIn an attempt to fill this gap, we provide a pedagogical perspective on how to structure a course to better impart knowledge to students and researchers in machine learning about when and how to implement various explainability techniques. We developed a comprehensive XAI course, focused on the conceptual characteristics of the different explanation types. Moreover, the course featured four structured workbooks focused on implementation, culminating in a final project requiring students to apply multiple XAI techniques to convince stakeholders about model decisions.ResultsCourse evaluation using a modified Course Experience Questionnaire (CEQ) from 16 MSc students revealed high perceived quality (CEQ score of 12,050) and strong subjective ratings regarding students' ability to analyze, design, apply, and evaluate XAI outcomes. All students successfully completed the course, with 89% of them demonstrating confidence in multi-perspective model analysis.DiscussionThe survey results demonstrated that interactive tutorials and practical workbooks were crucial for translating XAI theory into practical skills. Students particularly valued the balance between theoretical concepts and hands-on implementation, though evaluation of XAI outputs remained the most challenging aspect, suggesting future courses should include more structured interpretation exercises and analysis templates.

This paper presents a comprehensive evaluation of the capabilities of Large Language Models (LLMs) in metaphor interpretation across multiple datasets, tasks, and prompt configurations. Although metaphor processing has gained significant attention in Natural Language Processing (NLP), previous research has been limited to single-dataset evaluations and specific task settings, often using artificially constructed data through lexical replacement. We address these limitations by conducting extensive experiments using diverse publicly available datasets with inference and metaphor annotations, focusing on Natural Language Inference (NLI) and Question Answering (QA) tasks. The results indicate that LLMs' performance is more influenced by features like lexical overlap and sentence length than by metaphorical content, demonstrating that any alleged emergent abilities of LLMs to understand metaphorical language are the result of a combination of surface-level features, in-context learning, and linguistic knowledge. This work provides critical insights into the current capabilities and limitations of LLMs in processing figurative language, highlighting the need for more realistic evaluation frameworks in metaphor interpretation tasks. Data and code are publicly available.

Accurately predicting syngas composition is essential for optimizing energy production and ensuring environmental sustainability. Despite the growing use of machine learning techniques in this field, publicly available datasets remain limited, and existing datasets contain relatively few samples. To bridge this gap, we generated a comprehensive dataset of 3748 samples under controlled laboratory conditions and publicly shared it on Kaggle (https://www.kaggle.com/datasets/miracnurciner/gasification-dataset. This study aims to identify the most successful machine learning model for predicting H2 and CH4 gas concentrations by evaluating nine models: Random Forest (RF), Linear Regression (LR), Decision Tree (DT), Support Vector Regression (Linear and RBF), K-Nearest Neighbors (KNN), Gradient Boosting Regressor (GBR), XGBoost, CatBoost, and LightGBM. Model performance was assessed using multiple metrics, including the coefficient of determination (R2), root mean squared error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE), and explained variance score (EVS). The Friedman test was applied to evaluate the statistical significance of performance differences among the models. The results show that the KNN model achieved the highest predictive performance for both H2 (R2 = 0.987, RMSE = 1.253) and CH4 (R2 = 0.979, RMSE = 0.920). Friedman test shows that the performance differences between the models are statistically significant (p < 0.001). By integrating Shapley Additive Explanations (SHAP) into the model, the contribution of each feature to the prediction results is clarified. SHAP analysis highlights that temperature and time are the main features affecting H2 and CH4 gas. This study highlights the potential of machine learning techniques for biomass gas prediction and advocates for integrating Explainable AI (XAI) methods, establishing a robust foundation for future research. Furthermore, by providing a large, publicly available dataset, this research significantly advances studies in syngas composition prediction.

The complexity and interconnectivity of entities involved in money laundering demand investigative reasoning over graph-structured data. This paper explores the use of large language models (LLMs) as reasoning engines over localized subgraphs extracted from a financial knowledge graph. We propose a lightweight pipeline that retrieves k-hop neighborhoods around entities of interest, serializes them into structured text, and prompts an LLM via few-shot in-context learning to assess suspiciousness and generate justifications. Using synthetic anti-money laundering (AML) scenarios that reflect common laundering behaviors, we show that LLMs can emulate analyst-style logic, highlight red flags, and provide coherent explanations. While this study is exploratory, it illustrates the potential of LLM-based graph reasoning in AML and lays groundwork for explainable, language-driven financial crime analytics.