Shenglai Zeng, Tianqi Zheng, Chuan Tian, Dante Everaert, Yau-Shian Wang, Yupin Huang, Michael J. Morais, Rohit Patki, Jinjin Tian, Xinnan Dai, Kai Guo, Monica Xiao Cheng, Hui Liu. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Task offloading among low-earth orbit (LEO) satellites with on-board computing (OBC) is important for real-time applications. However, OBC is constrained by the battery capacity of LEO, which fluctuates with orbital dynamics and available solar power. This paper addresses the problem of energy sustainability and timeliness in LEO-OBC systems by proposing a sustainable OBC-LEO framework that combines parallel offloading strategies with dynamic energy management. This problem is formulated as a Markov decision process aiming to minimize the overall delay while satisfying the LEO satellite energy constraints and achieving a high task success rate. To balance immediate computational demands and long-term energy stability, a Lyapunov optimization-based dynamic parallel offloading (LODPO) algorithm is designed to make decisions dynamically within each time slot, integrated with subtask allocation based on a low-cost (SABLC) algorithm that dynamically adjusts task allocations. Finally, simulation results demonstrate that the LODPO framework achieves a significant reduction in execution delay, incurring only 34.0% of the delay cost of binary offloading. Most critically, it ensures exceptional reliability, with a task drop rate that is only 8.5% of that seen in binary offloading and 12.0% of that in the DQN-based approach. This ensures high responsiveness and dependability for mission-critical, delay-sensitive applications.

Haoyu Wang, Ruirui Li, Haoming Jiang, Jinjin Tian, Zhengyang Wang, Chen Luo, Xianfeng Tang, Monica Xiao Cheng, Tuo Zhao, Jing Gao. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

We tackle the challenge of predicting models' Out-of-Distribution (OOD) performance using in-distribution (ID) measurements without requiring OOD data. Existing evaluations with ``Effective robustness'', which use ID accuracy as an indicator of OOD accuracy, encounter limitations when models are trained with diverse supervision and distributions, such as class labels (*Vision Models, VMs, on ImageNet*) and textual descriptions (*Visual-Language Models, VLMs, on LAION*). VLMs often generalize better to OOD data than VMs despite having similar or lower ID performance. To improve the prediction of models' OOD performance from ID measurements, we introduce the *Lowest Common Ancestor (LCA)-on-the-Line* framework. This approach revisits the established concept of LCA distance, which measures the hierarchical distance between labels and predictions within a predefined class hierarchy, such as WordNet. We assess 75 models using ImageNet as the ID dataset and five significantly shifted OOD variants, uncovering a strong linear correlation between ID LCA distance and OOD top-1 accuracy. Our method provides a compelling alternative for understanding why VLMs tend to generalize better. Additionally, we propose a technique to construct a taxonomic hierarchy on any dataset using $K$-means clustering, demonstrating that LCA distance is robust to the constructed taxonomic hierarchy. Moreover, we demonstrate that aligning model predictions with class taxonomies, through soft labels or prompt engineering, can enhance model generalization. Open source code in our [Project Page](https://elvishelvis.github.io/papers/lca/).

Wireless-based federated learning (FL), as an emerging distributed learning approach, has been widely studied for 6G systems. When the paradigm shifts from terrestrial to non-terrestrial networks (NTN), FL may need to address several open challengings, e.g., limited service time of low earth orbit (LEO) satellites and time-efficient uploading and aggregation for massive devices. In this work, we exploit the synergy of LEO and FL for future integrated 6G-satellite systems by taking advantage of ubiquitous wireless access provided by LEO and appealing characteristics of collaborative training and data privacy preservation in FL. The studied LEO-FL framework may need to improve multi-metric performance in practice. Different from most FL works, we simultaneously improve the communication-training efficiency and local training accuracy from a multi-objective optimization (MOO) perspective. To solve the problem, we propose a decomposition, deep reinforcement learning and transfer learning based MOO algorithm for FL (DRT-FL), aiming at adapting to the dynamic satellite-terrestrial environments, achieving efficient uploading and aggregation, and approaching Pareto optimal sets. Compared to the state-of-the-art MOO algorithms, the effectiveness of the proposed LEO-FL framework and DRT-FL algorithm are assessed on MNIST and CIFAR-IO datasets.

A lot of recent advances in unsupervised feature learning are based on designing features which are invariant under semantic data augmentations. A common way to do this is contrastive learning, which uses positive and negative samples. Some recent works however have shown promising results for non-contrastive learning, which does not require negative samples. However, the non-contrastive losses have obvious “collapsed” minima, in which the encoders output a constant feature embedding, independent of the input. A folk conjecture is that so long as these collapsed solutions are avoided, the produced feature representations should be good. In our paper, we cast doubt on this story: we show through theoretical results and controlled experiments that even on simple data models, non-contrastive losses have a preponderance of non-collapsed bad minima. Moreover, we show that the training process does not avoid these minima. Code for this work can be found at https://github.com/ashwinipokle/contrastive_landscape.

MOTIVATION: Gene-gene co-expression networks (GCN) are of biological interest for the useful information they provide for understanding gene-gene interactions. The advent of single cell RNA-sequencing allows us to examine more subtle gene co-expression occurring within a cell type. Many imputation and denoising methods have been developed to deal with the technical challenges observed in single cell data; meanwhile, several simulators have been developed for benchmarking and assessing these methods. Most of these simulators, however, either do not incorporate gene co-expression or generate co-expression in an inconvenient manner. RESULTS: Therefore, with the focus on gene co-expression, we propose a new simulator, ESCO, which adopts the idea of the copula to impose gene co-expression, while preserving the highlights of available simulators, which perform well for simulation of gene expression marginally. Using ESCO, we assess the performance of imputation methods on GCN recovery and find that imputation generally helps GCN recovery when the data are not too sparse, and the ensemble imputation method works best among leading methods. In contrast, imputation fails to help in the presence of an excessive fraction of zero counts, where simple data aggregating methods are a better choice. These findings are further verified with mouse and human brain cell data. AVAILABILITY AND IMPLEMENTATION: The ESCO implementation is available as R package ESCO. Users can either download the development version via github (https://github.com/JINJINT/ESCO) or the archived version via Zenodo (https://zenodo.org/record/4455890). SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Major internet companies routinely perform tens of thousands of A/B tests each year. Such large-scale sequential experimentation has resulted in a recent spurt of new algorithms that can provably control the false discovery rate (FDR) in a fully online fashion. However, current state-of-the-art adaptive algorithms can suffer from a significant loss in power if null p-values are conservative (stochastically larger than the uniform distribution), a situation that occurs frequently in practice. In this work, we introduce a new adaptive discarding method called ADDIS that provably controls the FDR and achieves the best of both worlds: it enjoys appreciable power increase over all existing methods if nulls are conservative (the practical case), and rarely loses power if nulls are exactly uniformly distributed (the ideal case). We provide several practical insights on robust choices of tuning parameters, and extend the idea to asynchronous and offline settings as well.