Peidong Wang, Zhiming Ma, Xin Dai, YongKang Liu, Shi Feng, Xiaocui Yang, Wenxing Hu, Zhihao Wang, Mingjun Pan, Li Yuan, Daling Wang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

As a crucial data mining problem, domain adaptation (DA) under dataset shift offers an effective paradigm for transferring knowledge in changing environments (i.e., shifting data distributions). Recent theoretical advances show that learning conditional information is crucial for successful DA, and empirical models with provable performance have been developed under discrete priors. However, for many real-world complex tasks with continuous label spaces, e.g., regression, conditional information is generally intractable, and shift correction models remain unexplored due to the high-dimensional nature of continuous variables. To deal with these challenges, we propose an information-theoretic learning principle called bi-level conditional dependence (BiCD) for continuous shift setting. BiCD characterizes the conditional information of learned representations from two aspects: conditional domain dependence and conditional label dependence, where our results prove that bi-level dependence learning is sufficient to obtain representations with desired properties and to minimize the generalization error. Moreover, BiCD also admits appealing properties for practical modeling: 1) an equivalent formulation with marginal dependence is derived to avoid the explicit computation of high-dimensional conditional dependence in practical modeling; 2) a kernel dependence-based variant is developed to serve as a guaranteed numerical proxy for the vanilla BiCD; 3) a graph Laplacian understanding is provided to justify the learning mechanism of BiCD. Empirically, BiCD is evaluated on standard DA regression datasets, where the significance of the proposed dependence learning principle is validated and SOTA performance is consistently achieved.

Reinforcement Learning (RL) has emerged as a powerful tool for neural combinatorial optimization, enabling models to learn heuristics that solve complex problems without requiring expert knowledge. Despite significant progress, existing RL approaches face challenges such as diminishing reward signals and inefficient exploration in vast combinatorial action spaces, leading to inefficiency. In this paper, we propose **Preference Optimization**, a novel method that transforms quantitative reward signals into qualitative preference signals via statistical comparison modeling, emphasizing the superiority among sampled solutions. Methodologically, by reparameterizing the reward function in terms of policy and utilizing preference models, we formulate an entropy-regularized RL objective that aligns the policy directly with preferences while avoiding intractable computations. Furthermore, we integrate local search techniques into the fine-tuning rather than post-process to generate high-quality preference pairs, helping the policy escape local optima. Empirical results on various benchmarks, such as the Traveling Salesman Problem (TSP), the Capacitated Vehicle Routing Problem (CVRP) and the Flexible Flow Shop Problem (FFSP), demonstrate that our method significantly outperforms existing RL algorithms, achieving superior convergence efficiency and solution quality.

When it comes to deep-learning-based methods for despeckling synthetic aperture radar (SAR) images, maintaining spatial structure, ensuring process stability, and achieving fast inference are still major challenges. To address these issues, a novel method named scattering guided image despeckling (SGID), which is based on the wavelet—transform—conditioned diffusion model for SAR despeckling, has been proposed. First, the core component of this approach is the adaptive scattering extraction module (ASEM), which is designed to preserve the spatial structure of images effectively. It achieves the preservation of spatial structure by leveraging SAR imaging principles to adaptively extract scattering points, thereby retaining critical spatial information. Second, considering that speckle noise in SAR images is mainly concentrated in the high-frequency region, this study has designed a high-frequency despeckling module (HFDM). This module adopts a direction-aware adaptive weight allocation strategy, aiming to precisely regulate the weight allocation of high-frequency components in all the directions during the processing, so as to maximize the coherence and consistency of the high-frequency detail structures in the images. Finally, the entire processing workflow is conducted in the frequency domain, which contributes significantly to the improvement of inference speed. Experimental results demonstrate that the proposed method achieves state-of-the-art (SOTA) performance in both spatial structure preservation and despeckling quality. Besides, its inference speed is 0.19 s, which is 67 times faster than that of comparable diffusion-model-based methods. The code for this study will be publicly available athttps://github.com/SihaoDong/SGID.