Ziyi Wang, Yuxuan Lu, Yimeng Zhang, Pei Chen, Ziwei Dong, Jing Huang, Jiri Gesi, Xianfeng Tang, Chen Luo, Qun Liu, Yisi Sang, Hanqing Lu, Manling Li, Jin Lai, Dakuo Wang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Data selection bias has been a long-lasting challenge in the machine learning domain, especially in multi-stage recommendation systems, where the distribution of labeled items for model training is very different from that of the actual candidates during inference time. This distribution shift is even more prominent in the context of online advertising where the user base is diverse and the platform contains a wide range of contents. In this paper, we first investigate the data selection bias in the upper funnel (Ads Retrieval) of Pinterest's multi-cascade ads ranking system. We then conduct comprehensive experiments to assess the performance of various state-of-the-art methods, including transfer learning, adversarial learning, and unsupervised domain adaptation. Moreover, we further introduce some modifications into the unsupervised domain adaptation and evaluate the performance of different variants of this modified method. Our online A/B experiments show that the modified version of unsupervised domain adaptation (MUDA) could provide the largest improvements to the performance of Pinterest's advertisement ranking system compared with other methods and the one used in current production.

Efficient discovery and effective sharing of spectrum opportunities are the most challenging issues in a multi-channel cognitive radio network (CRN) with multiple secondary users (SUs). In this paper, we propose a novel spectrum sensing and access protocol in CRNs where SUs are allowed to sequentially sense two channels in a single time slot and are coordinated to access the potentially unused channels. The proposed protocol is formulated as a channel selection problem coupled with a channel assignment problem. We subsequently investigate the myopic sensing policy and propose a Markov chain-based greedy channel assignment scheme (MCGA) to maximize the expected total SU throughput. Moreover, we extend our proposal to the scenario with imperfect spectrum sensing and obtain the optimal sensing time to enhance the performance improvement by considering spectrum sensing and spectrum access in a joint manner. Finally, we evaluate the performance of our proposal in a saturated network. Numerical and simulation results demonstrate that compared to the existing work, our approach can achieve significant performance improvements in SU throughput.

Cooperative spectrum sensing has been proposed to significantly improve spectrum sensing accuracy by taking advantage of the cooperation among multiple secondary users (SUs). Most of existing work assumes that all SUs have the same SNR values of primary users' signal while the difference of SNR values among SUs, although it is very common in practice, is largely ignored. In this paper, we investigate two cooperative spectrum sensing scenarios where multiple geographically diverse SUs may have different SNR values. In the first scenario a cognitive radio network (CRN) with a single primary channel is considered. We aim to optimize the individual thresholds of SUs to maximize SU throughput subject to the constraint of missed detection probability. In the second scenario where there are multiple channels in a CRN, we jointly optimize the allocation of SUs to sense different channels and the individual detection thresholds of SUs. Our objective is to maximize total SU throughput over multiple channels while guaranteeing the missed detection probability below a given threshold. Simulation results demonstrate that our proposed schemes achieve significant improvements in SU throughput over the existing schemes.

Cooperative spectrum sensing has been proposed to significantly improve spectrum sensing accuracy by taking advantage of the cooperation among secondary users (SUs), but also this incurs some sensing cost. In this paper, we present a cooperative spectrum sensing model with consideration to spectrum sensing cost in distributed cognitive radio networks where each SU aims to maximize its utility. Under the scenario with selfish SUs, we formulate cooperative spectrum sensing as a non-cooperative game and obtain the mixed strategy Nash equilibrium of the formulated spectrum sensing game by deriving the sensing probabilities of SUs. Under the scenario with limited collaboration of SUs, we formulate cooperative spectrum sensing as a nonlinear optimization problem and derive the optimal sensing strategy of SUs by using our proposed Newton-Raphson based algorithm. Numerical results demonstrate that SUs with limited collaboration are able to achieve much better performance than the outcome of the Nash equilibrium and by choosing the optimal sensing strategy SUs are able to maximize their utility, which is an effective tradeoff between SU throughput and sensing cost.

In this paper we present a novel dynamic spectrum sensing and access model in cognitive radio networks. This model allows secondary users (SUs) to sequentially sense two channels in a single time slot and provides coordinated access of multiple SUs to the available channels. The presented access model is formulated as a channel assignment optimization problem which is shown to be NP-hard. We subsequently propose and analyze a Markov chain based greedy channel assignment scheme (MCGA) which allows for sequential sensing of two channels with a priority order per time slot. Finally, we analyze and evaluate the performance of our approach in a saturated network. Our analytical results, validated by simulation, indicate that compared to the existing work, our approach can achieve significant improvements in terms of SU throughput and MAC delay.

This paper studies optimal channel reservation in cooperative cognitive radio networks (CRNs) where secondary users (SUs) have access to the combined spectrum pool of cooperating CRNs. Motivated by SU high forced termination in cooperative CRNs, we propose two channel reservation schemes, Fixed Channel Placement Reservation (FCPR) and Dynamic Channel Placement Reservation (DCPR), and theoretically analyze their performances using the Markov chain approach. Our numerical results, validated by simulation, indicate that for a given number of reserved channels, the DCPR algorithm achieves better user experience by reducing the forced termination probability. Based on this analysis, we propose two enhanced reservation algorithms: Algorithm A maximizes the overall capacity of CRNs to enable network operators to increase their revenue; Algorithm B minimizes the user experience cost function to provide better services.