Aug 2026· Neural Networks· Vol 205 Pt B, pp.
109503
· 0 citations· 62 references
Medicine
TL;DR
This work introduces a comprehensive framework featuring a dynamic category-scaling weight mechanism that hierarchically addresses imbalance at the sample and domain levels, and builds a progressively refined "Ideal Domain" through the adaptive integration of well-performing domains.
Abstract
Class imbalance presents a persistent bottleneck in machine learning, biasing models toward majority classes and degrading performance on rare events. Active learning partially alleviates this by focusing annotation efforts on underrepresented samples. However, this issue becomes more pronounced in multi-domain learning settings, where both intra-domain imbalances and shifts across domains complicate model training. Current multi-domain active learning techniques typically neglect the interaction of these dual imbalances, resulting in biased selection strategies and limited ability to generalize across domains. To tackle this, we introduce a comprehensive framework featuring a dynamic category-scaling weight mechanism that hierarchically addresses imbalance at the sample and domain levels. This mechanism assesses the relative difficulty of each class to guide effective sampling, while concurrently constructing a progressively refined "Ideal Domain" through the adaptive integration of well-performing domains. This ideal representation further supports a minimax optimization process to allocate annotation budgets both fairly and robustly. Extensive evaluations on various standard datasets confirm that our approach consistently surpasses leading methods in managing complex imbalanced scenarios across multiple domains.
Multiclass imbalance and concept drift often coexist in real-world data streams, and the challenge becomes more severe under limited labeling budgets, where existing online ensemble and active learning methods still struggle to preserve minority-class recognition while adapting to evolving concepts. To address this issue, we propose DMAE, a dual-memory active ensemble learning method for multiclass imbalanced concept-drifting data streams. DMAE integrates an ensemble classifier, a drift detector, an instance sliding window, a label sliding window, a long-term prototype memory, and an initialization training sequence into a unified online framework for prediction, querying, memory maintenance, and model updating. To improve label efficiency, we develop a variable-threshold uncertainty strategy based on a decomposable asymmetric margin-threshold matrix, which combines a global threshold with class-pair-level corrections to focus queries on genuinely ambiguous, minority-relevant regions while controlling annotation cost. We further design drift-strength-aware dual-memory initialization (SDMI) and a probationary soft-replacement strategy (PDSR), which adaptively balance recent information and long-term prototypes according to online-estimated drift strength, supporting both historical-pattern retention and rapid adaptation under abrupt, gradual, and mixed drifts. We also propose a composite sample-weighting formulation, PCN-Weight, which jointly models boundary difficulty, class-imbalance status, sample–prototype relations, and temporal decay to guide incremental training and strengthen minority-class and hard-region representations. Experiments on 15 synthetic data streams and 5 real-world imbalanced data streams show that DMAE achieves more stable overall performance than ten state-of-the-art ensemble baselines in Accuracy, Kappa, G-Mean, and Recall, while remaining robust under different labeling rates and key hyperparameter settings.
Meng Han, Yajie Xue, Yikai Li et al.· Journal of King Saud Univers...· 0 citations
Deep neural networks excel in various tasks but struggle to generalize across evolving data distributions, leading to significant performance degradation under domain shifts. Domain incremental learning (DIL) addresses this challenge by enabling models to continuously adapt while retaining prior knowledge. Among existing DIL approaches, the parameter-isolation paradigm achieves state-of-the-art performance. However, these methods often adopt a one-size-fits-all approach to adapt to new domains, resulting in either insufficient learning capacity or redundant parameters. In this work, we propose BPG, a unified framework that addresses both challenges through two complementary components: BPG-Adapter, which dynamically determines each domain's adapter hidden dimension based on domain-specific feature separability, and BPG-Inference, a soft domain mixture strategy that integrates multiple domain-specific models at test time, mitigating domain ID misselection. Experimental results on DomainNet, CDDB, and CORe50 demonstrate that BPG consistently outperforms uniform adapter-based approaches and hard domain selection strategies, achieving state-of-the-art average accuracy while reducing forgetting to as low as 0.22% on DomainNet.
Qiang Wang, Songlin Dong, Shaokun Wang et al.· 0 citations
Active Domain Adaptation (ADA) enhances transfer learning by selecting the most informative samples from the target domain for annotation. The source-free variant, ASFDA, operates using only a pre-trained source model and unlabeled target data. However, existing approaches typically assume that distribution shift arises solely from changes in the feature or label distribution, thereby neglecting concept shift (the divergence of $p(y|x)$ across domains). In this paper, we propose ADA-CS, a plug-and-play module compatible with any ADA or ASFDA framework. Without requiring access to source data, ADA-CS isolates concept shift from other forms of shifts. The concept shift severity is quantified to prioritize samples exhibiting the largest discrepancy in $p(y|x)$ for annotation. We further introduce a CSS metric to quantify the Concept Shift Severity across domains, revealing that non-negligible concept shift exists in many transfer tasks. Extensive experiments on four vision benchmarks (Office-31, Digits, DomainNet, and Office-Home RSUT) and one large-scale tabular dataset (USAccident) show that ADA-CS improves performance when combined with a wide range of active learning, ADA, and ASFDA strategies. Moreover, the rapid decline of CSS during adaptation provides direct evidence of our approach's effectiveness in identifying and correcting concept shift.
Overall, this dissertation provides a unified investigation into data imbalance, data quality, and data scarcity-three core bottlenecks of modern deep learning-and proposes principled solutions that improve robustness, interpretability, and efficiency across both CV and NLP domains.
Classifying cybersecurity vulnerabilities using the Common Weakness Enumeration (CWE) taxonomy is challenging due to extreme class imbalance and strong hierarchical dependencies among weakness categories. Although oversampling techniques such as Synthetic Minority Oversampling Technique (SMOTE) and Adaptive Synthetic Sampling (ADASYN) are widely adopted to mitigate class imbalance, their effectiveness for hierarchical CWE text classification remains largely unexplored. This paper proposes a Hierarchy-Aware RoBERTa framework that explicitly incorporates CWE structural information through learnable parent-class embeddings, preserving taxonomic consistency. Our experiments demonstrate that synthetic interpolation in high-dimensional embedding spaces violates the inherent parent-child constraints of the CWE hierarchy, offering only marginal benefits for classical ML models while consistently degrading deep learning architectures. Evaluated on a CWE Research Concept dataset, the proposed model achieves a weighted F1-score of 0.76 without data augmentation, outperforming all baselines with notable gains on minority classes, including the Class category whose F1-score improves from 0.49 to 0.60 over the BERT baseline. Our results suggest that hierarchyaware representation learning is a more principled alternative to oversampling for structured vulnerability classification.
Bipin Chhetri, D. Giri, Avishek Kadel et al.· Annual International Compute...· 0 citations
Data streams in the real world often display severe class imbalance, with a critical under-representation of minority-class instances, resulting in biased and unreliable classification results from conventional deep learning frameworks. We introduce a new Imbalance-Resilient Deep Classification Framework that is defined for robust and stable analytics deployed on dynamic non-stationary data streams. The proposed framework combines adaptive resampling strategies, cost-sensitive learning mechanisms, and deep neural architectures to address the adverse impact of imbalanced class distributions. To cope with this, we propose a new dynamic threshold calibration module which gradually updates decision boundaries according to the changing characteristics of data streams in order to ensure relatively high classification accuracy overtime. The framework also includes an ensemble-driven feature extraction pipeline, enabling to learn discriminative representation for minority classes without compromising majority-class accuracy. The experiments conducted on benchmark and real-world imbalanced streaming datasets show that the proposed approach outperforms existing state-of-the-art algorithms, both in G-Mean, AUC-ROC and F1-Score metrics. Our findings validate the convenience, scalability and broad applicability of our framework in important applications like fraud detection, medical diagnosis, and network intrusion where imbalance constantly arises as a prominent issue.
Soma Sekhar Gaddipati, T. Lakshmi, Nithya Krishnan et al.· 2026 International Conferenc...· 0 citations