Mobile agents rely on Large Language Models (LLMs) to plan and execute tasks on smartphone user interfaces (UIs). While cloud-based LLMs achieve high task accuracy, they require uploading the full UI state at every step, exposing unnecessary and often irrelevant information. In contrast, local LLMs avoid UI uploads but suffer from limited capacity, resulting in lower task success rates. We propose $\textbf{CORE}$, a $\textbf{CO}$llaborative framework that combines the strengths of cloud and local LLMs to $\textbf{R}$educe UI $\textbf{E}$xposure, while maintaining task accuracy for mobile agents. CORE comprises three key components: (1) $\textbf{Layout-aware block partitioning}$, which groups semantically related UI elements based on the XML screen hierarchy; (2) $\textbf{Co-planning}$, where local and cloud LLMs collaboratively identify the current sub-task; and (3) $\textbf{Co-decision-making}$, where the local LLM ranks relevant UI blocks, and the cloud LLM selects specific UI elements within the top-ranked block. CORE further introduces a multi-round accumulation mechanism to mitigate local misjudgment or limited context. Experiments across diverse mobile apps and tasks show that CORE reduces UI exposure by up to 55.6\% while maintaining task success rates slightly below cloud-only agents, effectively mitigating unnecessary privacy exposure to the cloud. The code is available at https://github.com/Entropy-Fighter/CORE.

Visual AutoRegressive (VAR) modeling employs a next-scale decoding paradigm that progresses from coarse structures to fine details. While enhancing fidelity and scalability, this approach challenges two fundamental assumptions of conventional dynamic inference: semantic stability (intermediate outputs approximating final results) and monotonic locality (smooth representation evolution across layers), which renders existing dynamic inference methods ineffective for VAR models. To address this challenge, we propose FreqExit, a unified training framework that enables dynamic inference in VAR without altering its architecture or compromising output quality. FreqExit is based on a key insight: high-frequency details are crucial for perceptual quality and tend to emerge only in later decoding stages. Leveraging this insight, we design targeted mechanisms that guide the model to learn more effectively through frequency-aware supervision. The proposed framework consists of three components: (1) a curriculum-based supervision strategy with progressive layer dropout and early exit loss; (2) a wavelet-domain high-frequency consistency loss that aligns spectral content across different generation steps; and (3) a lightweight self-supervised frequency-gated module that guides adaptive learning of both structural and detailed spectral components. On ImageNet 256×256, FreqExit achieves up to 2× speedup with only minor degradation, and delivers 1.3× acceleration without perceptible quality loss. This enables runtime-adaptive acceleration within a unified model, offering a favorable trade-off between efficiency and fidelity for for practical and flexible deployment.

Many billion-scale large language models (LLMs) have been released for resource-constraint mobile devices to provide local LLM inference service when cloud-based powerful LLMs are not available. However, the capabilities of current on-device LLMs still lag behind those of cloud-based LLMs, and how to effectively and efficiently enhance on-device LLM inference becomes a practical requirement. We thus propose to collect the user's historical interactions with the cloud-based LLM and build an external datastore on the mobile device for enhancement using nearest neighbors search. Nevertheless, the full datastore improves the quality of token generation at the unacceptable expense of much slower generation speed. To balance performance and efficiency, we propose to select an optimal subset of the full datastore within the given size limit, the optimization objective of which is proven to be submodular. We further design an offline algorithm, which selects the subset after the construction of the full datastore, as well as an online algorithm, which performs selection over the stream and can be flexibly scheduled. We theoretically analyze the performance guarantee and the time complexity of the offline and the online designs to demonstrate effectiveness and scalability. We finally take three ChatGPT related dialogue datasets and four different on-device LLMs for evaluation. Evaluation results show that the proposed designs significantly enhance LLM performance in terms of perplexity while maintaining fast token generation speed. Practical overhead testing on the smartphone reveal the efficiency of on-device datastore subset selection from memory usage and computation overhead.

Federated learning is a new distributed machine learning framework, where numerous heterogeneous clients collaboratively train a model without sharing training data. In this work, we consider a practical and ubiquitous issue when deploying federated learning in mobile environments: intermittent client availability, where the set of eligible clients may change during the training process. Such intermittent client availability would seriously deteriorate the performance of the classical federated averaging algorithm (FedAvg). Thus, we propose a simple distributed nonconvex optimization algorithm, called federated latest averaging (FedLaAvg), which leverages the latest gradients of all clients, even when the clients are not available, to jointly update the global model in each iteration. Our theoretical analysis shows that FedLaAvg achieves guaranteed convergence and a sublinear speedup with respect to the total number of clients. We implement FedLaAvg along with several baselines and evaluate them over the benchmarking MNIST and Sentiment140 data sets. The evaluation results demonstrate that FedLaAvg achieves more stable training than FedAvg in both convex and nonconvex settings and reaches a sublinear speedup. Source code and online supplement are available at the IJOC GitHub site ( http://dx.doi.org/10.1287/ijoc.2022.0057.cd , https://github.com/INFORMSJoC/2022.0057 ). History: Accepted by Ram Ramesh, Area Editor for Data Science & Machine Leaning. Funding: This work was supported by the National Key R&D Program of China [Grant 2022ZD0119100], the National Natural Science Foundation of China (NSFC) [Grants 61972252, 61972254, 62072303, 62025204, 62132018, 62202296, and 62202297], the Alibaba Innovation Research (AIR) Program, and the Tencent Rhino Bird Key Research Project. Supplemental Material: The software that supports the findings of this study is available within the paper and its Supplemental Information ( https://pubsonline.informs.org/doi/suppl/10.1287/ijoc.2022.0057 ) as well as from the IJOC GitHub software repository ( https://github.com/INFORMSJoC/2022.0057 ). The complete IJOC Software and Data Repository is available at https://informsjoc.github.io/ .

Mobile livestreaming has revolutionized the online shopping paradigm, enabling streamers to promote products to consumers with an immersive and interactive experience. To guide consumers to the livestreams that involve their interested products, it is necessary to have a good understanding of livestreaming contents with low latency, and the key task is to accurately recognize the products being promoted by the streamers. However, the mainstream cloud-based service framework is challenged by the high concurrency of service requests, the high overhead of multimodal recognition, and the requirement of low response latency. To break the bottleneck, we propose a new device-cloud collaborative learning framework, where each streamer’s mobile device holds a unimodal recognition model that can process most of frames and also uploads the extracted unimodal features to facilitate the cloud-side multimodal recognition of the remaining few frames. In addition, the on-device unimodal model is incrementally trained over the samples constructed by leveraging the streamers’ manual labeling behaviors, thereby adapting to the heterogeneous and dynamic livestreaming contents of different streamers. Nevertheless, the device-side personalized unimodal features are misaligned in feature space and cannot be directly fused into the cloud-side multimodal model. We thus design a pluggable prompt generation module to transform the personalized unimodal features into prompt embeddings, instructing the multimodal backbone network in feature fusion. Both offline and online evaluation results reveal the effectiveness and efficiency of our design as well as its consistent advantage over existing baselines.

Cloud-based learning is currently the mainstream in both academia and industry. However, the global data distribution, as a mixture of all the users' data distributions, for training a global model may deviate from each user's local distribution for inference, making the global model non-optimal for each individual user. To mitigate distribution discrepancy, on-device training over local data for model personalization is a potential solution, but suffers from serious overfitting. In this work, we propose a new device-cloud collaborative learning framework under the paradigm of domain adaption, called MPDA, to break the dilemmas of purely cloud-based learning and on-device training. From the perspective of a certain user, the general idea of MPDA is to retrieve some similar data from the cloud's global pool, which functions as large-scale source domains, to augment the user's local data as the target domain. The key principle of choosing which outside data depends on whether the model trained over these data can generalize well over the local data. We theoretically analyze that MPDA can reduce distribution discrepancy and overfitting risk. We also extensively evaluate over the public MovieLens 20M and Amazon Electronics datasets, as well as an industrial dataset collected from Mobile Taobao over a period of 30 days. We finally build a device-tunnel-cloud system pipeline, deploy MPDA in the icon area of Mobile Taobao for click-through rate prediction, and conduct online A/B testing. Both offline and online results demonstrate that MPDA outperforms the baselines of cloud-based learning and on-device training only over local data, from multiple offline and online metrics.

We focus on federated learning in practical recommender systems and natural language processing scenarios. The global model for federated optimization typically contains a large and sparse embedding layer, while each client’s local data tend to interact with part of features, updating only a small submodel with the feature-related embedding vectors. We identify a new and important issue that distinct data features normally involve different numbers of clients, generating the differentiation of hot and cold features. We further reveal that the classical federated averaging algorithm (FedAvg) or its variants, which randomly selects clients to participate and uniformly averages their submodel updates, will be severely slowed down, because different parameters of the global model are optimized at different speeds. More specifically, the model parameters related to hot (resp., cold) features will be updated quickly (resp., slowly). We thus propose federated submodel averaging (FedSubAvg), which introduces the number of feature-related clients as the metric of feature heat to correct the aggregation of submodel updates. We prove that due to the dispersion of feature heat, the global objective is ill-conditioned, and FedSubAvg works as a suitable diagonal preconditioner. We also rigorously analyze FedSubAvg’s convergence rate to stationary points. We finally evaluate FedSubAvg over several public and industrial datasets. The evaluation results demonstrate that FedSubAvg significantly outperforms FedAvg and its variants.

Federated learning allows mobile clients to jointly train a global model without sending their private data to a central server. Extensive works have studied the performance guarantee of the global model, however, it is still unclear how each individual client influences the collaborative training process. In this work, we defined a new notion, called {\em Fed-Influence}, to quantify this influence over the model parameters, and proposed an effective and efficient algorithm to estimate this metric. In particular, our design satisfies several desirable properties: (1) it requires neither retraining nor retracing, adding only linear computational overhead to clients and the server; (2) it strictly maintains the tenets of federated learning, without revealing any client's local private data; and (3) it works well on both convex and non-convex loss functions, and does not require the final model to be optimal. Empirical results on a synthetic dataset and the FEMNIST dataset demonstrate that our estimation method can approximate Fed-Influence with small bias. Further, we show an application of Fed-Influence in model debugging.