adaptive representation aggregation

Multi-source predictive aggregation often degrades coverage probability—e.g., from the nominal level (1-alpha) down to (1-2alpha)—under standard conformal prediction. Method: This paper proposes a weighted p-value-based adaptive fusion framework that enables continuous interpolation of coverage guarantees between (1-alpha) and (1-2alpha), breaking the rigidity of fixed-bound approaches. It establishes a data-dependent, finite-sample valid theory for weighted aggregation and provides a general construction scheme. Crucially, it integrates weight learning directly into the conformal inference framework, ensuring both coverage controllability and practical utility during aggregation. Contribution/Results: Compared to standard conformal prediction, the method significantly improves prediction set compactness on both synthetic and real-world datasets, reducing coverage error by over 30% while rigorously maintaining the target (1-alpha) coverage guarantee.

This work addresses the challenge of effectively aggregating statistical evidence under unknown dependence structures by proposing a unified framework grounded in permutation invariance. The approach constructs exchangeable data units, aggregates statistics within transformed datasets, and calibrates results across transformations, accommodating single-batch, sequential, and two-stage strategies. By integrating group invariance, exchangeability modeling, sequential alpha-spending, and a decoupling of standardization from calibration, the method achieves high power and adaptivity in finite samples, substantially outperforming traditional calibration techniques such as Bonferroni correction. Empirical evaluations demonstrate that the framework guarantees valid inference under arbitrary dependence structures in tasks including nonparametric testing and conformal prediction, while supporting data-driven aggregation rules and early rejection mechanisms.

This work addresses the optimal adaptive aggregation of source and target domain samples in transfer learning to minimize the target risk. To overcome the limitation of existing methods—namely, their inability to uniformly handle diverse distribution divergence measures—we propose a unified weak/strong transfer modulus framework. This is the first approach that automatically adapts to multiple divergence classes—including Wasserstein distance and integral probability metrics (IPMs)—and characterizes their statistical limits. By integrating confidence-set reduction, modulus upper-bound derivation, and adaptive weighted estimation, we achieve near-optimal convergence rates even when the transfer modulus is unknown. Theoretical analysis further reveals that, under causal modeling assumptions, the framework yields provable generalization gains beyond standard transfer bounds. Extensive experiments demonstrate significant improvements in cross-domain classification and regression performance.

Conformal prediction for multi-class classification often suffers from inefficiency and overly large prediction sets due to reliance on a single scoring function. To address this, we propose a weighted ensemble of multiple scoring functions within the conformal prediction framework. Our method learns data-driven weights via joint optimization grounded in empirical risk minimization, integrating Vapnik–Chervonenkis (VC) theory with convex optimization. Crucially, we establish, for the first time, a theoretical connection between weighted score aggregation and VC subgraph classes—thereby enabling provably optimal multi-score fusion. Under strict coverage guarantees (e.g., 90%), our approach significantly reduces prediction set size, achieving an average reduction of 12.6% across multiple benchmark datasets. It consistently outperforms state-of-the-art single-score conformal methods in both efficiency and predictive performance.

Fuzzy rule models offer strong interpretability but suffer from poor scalability and susceptibility to overfitting in complex tasks and large-scale data scenarios. To address these limitations, this paper proposes a novel ensemble framework integrating gradient boosting with fuzzy rule-based base learners. We introduce a dynamic control factor that adaptively adjusts the weights of fuzzy base models in each boosting iteration, simultaneously serving as a regularizer and performance optimizer. Additionally, we design a validation-set-driven, sample-level correction mechanism to enhance generalization and ensemble diversity. Experimental results demonstrate that our approach significantly mitigates overfitting, reduces rule complexity (e.g., fewer rules and shorter antecedents), and preserves high model interpretability and maintainability. The method thus provides a practical pathway for deploying interpretable AI in complex industrial applications.

This work addresses the trade-off between computational cost and accuracy in neural network ensembles, where existing ensemble methods are computationally expensive while conventional weight aggregation techniques often sacrifice performance. To bridge this gap, the authors propose a “partial fusion” framework that formulates weight aggregation as a generalized pruning process. By leveraging partial optimal transport, the method matches and fuses the most similar neurons across models, allowing for neuron deletion, isolation, or linear combination. A similarity metric at the neuron level enables a controllable balance between computational overhead and model accuracy. Experiments demonstrate that partial fusion significantly reduces computational costs while preserving accuracy close to that of full ensembles; furthermore, its single-model variant outperforms traditional pruning approaches.

This work addresses the challenge of simultaneously achieving calibration, low regret, and multi-accuracy in online learning under arbitrarily time-varying data distributions—a setting where existing methods struggle to balance these competing objectives. The authors propose a novel local adaptive mechanism that integrates a multi-objective optimization framework with adaptive online learning algorithms. Without requiring explicit definitions of local targets, their approach dynamically optimizes performance over contiguous subintervals, thereby circumventing the limitations of traditional global worst-case analyses. Empirical evaluations on energy forecasting and algorithmic fairness benchmarks demonstrate that the method significantly outperforms current state-of-the-art techniques, delivering unbiased predictions for subpopulations while maintaining robust multi-objective performance under distributional shifts.

This work proposes an enhanced local learning framework to address the limitations of the original Forward-Forward algorithm in stability, robustness, and generalization. By integrating multi-scale goodness aggregation, layer-wise adaptive thresholds, adaptive curriculum-guided hard negative mining, and a warm-up cosine annealing learning rate schedule, the training dynamics are effectively optimized. The proposed method significantly improves performance while preserving biological plausibility and memory efficiency, achieving accuracy gains of up to 1.45% on MNIST and 1.50% on Fashion-MNIST without introducing noticeable computational overhead.

This study addresses the low rule recall inherent in existing Feature-Guided Analysis (FGA) methods for interpreting deep neural networks, which limits their practical applicability. To overcome this limitation, the work introduces ensemble learning into FGA for the first time and proposes a scalable rule aggregation mechanism that combines multiple neuron activation rules according to three distinct criteria. This approach significantly improves recall while maintaining high precision. Experimental results on the MNIST and LSC datasets demonstrate that the proposed method increases training recall by 28.51% and 33.15%, respectively, and boosts test recall by 25.76% and 30.81%, with a negligible drop in test accuracy of less than 1%. These findings illustrate a flexible and effective trade-off between precision and recall in model interpretation.

Conventional static core-set selection fails to adapt to the heterogeneous requirements across different training stages. Method: This paper proposes a dynamic multi-objective adaptive core-set selection framework that dynamically switches sampling strategies according to training progression—emphasizing class balance in early stages, feature diversity in mid-stages, and prediction uncertainty in late stages—thereby enabling the first training-process-aware, multi-objective co-optimization. Contribution/Results: We theoretically establish a (1−1/e)-approximation guarantee. By integrating submodular optimization, active learning, and representation analysis, our method achieves O(n log n) computational efficiency. Empirically, it attains full-dataset accuracy on multiple benchmarks while significantly reducing memory overhead. Moreover, it is the first work to quantitatively characterize the dynamic evolution of data utility throughout training.