One important direction of Federated Foundation Models (FedFMs) is leveraging data from small client models to enhance the performance of a large server‑side foundation model. Existing methods based on model level or representation level knowledge transfer either require expensive local training or incur high communication costs and introduce unavoidable privacy risks. We reformulate this problem as a reinforcement learning style evaluation process and propose FedGRPO, a privacy preserving framework comprising two modules. The first module performs competence-based expert selection by building a lightweight confidence graph from auxiliary data to identify the most suitable clients for each question. The second module leverages the “Group Relative” concept from the Group Relative Policy Optimization (GRPO) framework by packaging each question together with its solution rationale into candidate policies, dispatching these policies to a selected subset of expert clients, and aggregating solely the resulting scalar reward signals via a federated group–relative loss function. By exchanging reward values instead of data or model updates, FedGRPO reduces privacy risk and communication overhead while enabling parallel evaluation across heterogeneous devices. Empirical results on diverse domain tasks demonstrate that FedGRPO achieves superior downstream accuracy and communication efficiency compared to conventional FedFMs baselines.

The decentralized nature of federated learning (FL) makes it difficult to verify the trustworthiness of participating clients, creating an opportunity for backdoor attacks. This paper addresses a general backdoor-resilient decentralized FL problem without any prior knowledge of the type of backdoor attacks or information about malicious clients. After an in-depth investigation of how backdoor attacks are conducted in FL, we introduce a multi-armed bandit-based knowledge distillation approach to help benign clients learn useful knowledge from other clients while rejecting potential backdoors hidden in shared updates. Unlike most previous works that rely on identifying and removing malicious updates—an approach limited to scenarios with fewer than 50% attackers—our knowledge distillation technique enables benign clients to reject backdoored knowledge while preserving useful information, maintaining effective defense even when malicious clients exceed 50% of the population. Additionally, to handle the various updates from clients with Non-IID dataset, a multi-armed bandit scheme is designed for each benign client to select the most appropriate teachers for knowledge distillation, resulting in high accuracy and fast convergence. Extensive experiments demonstrate that our multi-armed bandit-based knowledge distillation approach achieves high accuracy and general backdoor resilience. Comparisons with previous works show that our approach can reduce the attack success rate by 14.71%∼96.78% on average.

The growing resource demands of large-scale transformer models pose significant challenges for resource-constrained users, particularly in distributed environments. To address this issue, we propose a federated learning framework called Fed-Grow, which enables multiple participants to collaboratively learn a lightweight scaling operation that transfers knowledge from pretrained small models to a large transformer model. In Fed-Grow, we introduce the Dual-LiGO (Dual Linear Growth Operator) architecture, consisting of Local-LiGO and Global-LiGO components. Local-LiGO addresses model heterogeneity by adapting each participant's pre-trained model to a common intermediate form, while Global-LiGO facilitates knowledge sharing across participants without sharing local models or raw data, ensuring privacy preservation. This federated approach offers a scalable solution for growing large transformers in a distributed manner, where only the Global-LiGO is shared, significantly reducing communication overhead while maintaining comparable model performance under the same communication constraints. Experimental results demonstrate that Fed-Grow outperforms state-of-the-art methods in terms of accuracy and precision, while reducing the number of trainable parameters by 59.25% and communication costs by 73.01%. These improvements allow for higher efficiency in training large models in distributed environments, without sacrificing performance. To the best of our knowledge, Fed-Grow is the first method that enables cooperative transformer scaling in a distributed setting, making it a practical solution for resource-constrained users.

Recent research in federated large language models (LLMs) has primarily focused on enabling clients to fine-tune their locally deployed homogeneous LLMs collaboratively or on transferring knowledge from server-based LLMs to small language models (SLMs) at downstream clients. However, a significant gap remains in the simultaneous mutual enhancement of both the server’s LLM and clients’ SLMs. To bridge this gap, we propose FedMKT, a parameter-efficient federated mutual knowledge transfer framework for large and small language models. This framework is designed to adaptively transfer knowledge from the server’s LLM to clients’ SLMs while concurrently enhancing the LLM with clients’ unique domain insights. We facilitate token alignment using minimum edit distance (MinED) and then selective mutual knowledge transfer between client-side SLMs and a server-side LLM, aiming to collectively enhance their performance. Through extensive experiments across three distinct scenarios, we evaluate the effectiveness of FedMKT by utilizing diverse public LLMs and SLMs on a variety of NLP text generation tasks. Empirical results demonstrate that FedMKT simultaneously boosts the performance of both LLMs and SLMs. Our code has been contributed to the FATE open-source project and is now publicly accessible at https://github.com/FederatedAI/FATE-LLM/tree/main/python/fate_llm/algo/fedmkt

Federated continual learning (FCL) allows each client to continually update its knowledge from task streams, enhancing the applicability of federated learning in real-world scenarios. However, FCL needs to address not only spatial data heterogeneity between clients but also temporal data heterogeneity between tasks. In this paper, empirical experiments demonstrate that such input-level heterogeneity significantly affects the model’s internal parameters and outputs, leading to severe spatial-temporal catastrophic forgetting of local and previous knowledge. To this end, we propose Federated Tail Anchor (FedTA) to mix trainable Tail Anchor with the frozen output features to adjust their position in the feature space, thereby overcoming parameter-forgetting and output-forgetting. Three novel components are also included: Input Enhancement for improving the performance of pre-trained models on downstream tasks; Selective Input Knowledge Fusion for fusion of heterogeneous local knowledge on the server; and Best Global Prototype Selection for finding the best anchor point for each class in the feature space. Extensive experiments demonstrate that FedTA not only outperforms existing FCL methods but also effectively preserves the relative positions of features. Code is available at: https://github.com/SkyOfBeginning/FedTA_CVPR2025.

Federated Learning (FL) is a promising approach for training machine learning models on decentralized data while preserving privacy. However, privacy risks, particularly Membership Inference Attacks (MIAs), which aim to determine whether a specific data point belongs to a target client’s training set, remain a significant concern. Existing methods for implementing MIAs in FL primarily analyze updates from the target client, focusing on metrics such as loss, gradient norm, and gradient difference. However, these methods fail to leverage updates from non-target clients, potentially underutilizing available information. In this paper, we first formulate a one-tailed likelihood-ratio hypothesis test based on the likelihood of updates from non-target clients. Building upon this formulation, we introduce a three-step Membership Inference Attack (MIA) method, called FedMIA, which follows the "all for one"—leveraging updates from all clients across multiple communication rounds to enhance MIA effectiveness. Both theoretical analysis and extensive experimental results demonstrate that FedMIA outperforms existing MIAs in both classification and generative tasks. Additionally, it can be integrated as an extension to existing methods and is robust against various defense strategies, Non-IID data, and different federated structures. Our code is available in https://github.com/Liar-Mask/FedMIA.

Decentralized Federated Learning (DFL) is a privacy-preserving framework that eliminates the need for a central server, enabling direct communication between clients to save communication resources. However, existing approaches do not adequately address the high data heterogeneity among communication neighbors and often employ inefficient neighbor selection strategies. We propose ProToDFL, a novel framework that leverages prototype learning for dynamic topology optimization in DFL. ProToDFL constructs communication topologies through selecting 1-hop and 2-hop neighbors based on the Kullback-Leibler divergence between prototype vectors, ensuring alignment with local data distributions. To prevent topology splitting during optimization, it develops a cut-edge detection mechanism. Additionally, ProToDFL incorporates a Stop-and-Wait strategy inspired by exponential back-off to minimize prototype vector transmissions. We provide theoretical convergence analysis of ProToDFL within a general non-convex framework for decentralized training. Experimental results on multiple real-world datasets show that ProToDFL significantly outperforms existing DFL methods, particularly in heterogeneous and sparse environments.

Continual Learning (CL) primarily aims to retain knowledge to prevent catastrophic forgetting and transfer knowledge to facilitate learning new tasks. Unlike traditional methods, we propose a novel perspective: CL not only needs to prevent forgetting, but also requires intentional forgetting.This arises from existing CL methods ignoring biases in real-world data, leading the model to learn spurious correlations that transfer and amplify across tasks. From feature extraction and prediction results, we find that data biases simultaneously reduce CL's ability to retain and transfer knowledge. To address this, we propose ErrorEraser, a universal plugin that removes erroneous memories caused by biases in CL, enhancing performance in both new and old tasks. ErrorEraser consists of two modules: Error Identification and Error Erasure. The former learns the probability density distribution of task data in the feature space without prior knowledge, enabling accurate identification of potentially biased samples. The latter ensures only erroneous knowledge is erased by shifting the decision space of representative outlier samples. Additionally, an incremental feature distribution learning strategy is designed to reduce the resource overhead during error identification in downstream tasks. Extensive experimental results show that ErrorEraser significantly mitigates the negative impact of data biases, achieving higher accuracy and lower forgetting rates across three types of CL methods. The code is available at https://github.com/diadai/ErrorEraser.

We investigate the challenge of generating OverpassQL from natural language in the Text-to-OverpassQL task and explore the data in the existing OverpassNL dataset. To address the structural mismatch between natural language and OverpassQL, we propose a task decomposition-based multi-step prompting approach that generates auxiliary information to help align natural language with OverpassQL structures, thereby enhancing model performance. Furthermore, we introduce a Key-Value Correction Module specifically targeting key-value pair matching difficulties in Text-to-OverpassQL tasks, designed to rectify potential syntactic errors and key-value mismatches in generated queries. Our experiments on GPT-3.5 Turbo and GPT-4 demonstrate absolute performance gains of$\mathbf{1. 4 \%}$and 0.6 % respectively. Under retrieval-augmented setting ablation, we achieve a more significant 3.5 % improvement with GPT-3.5 Turbo. Experimental results confirm that our method consistently improves performance across various models and configurations, particularly showing enhanced effectiveness in medium and small-scale models.

In this paper, we explore an important yet previously neglected question: Do context aggregation patterns across Language Models (LMs) share commonalities? While some works have investigated context aggregation or attention weights in LMs, they typically focus on individual models or attention heads, lacking a systematic analysis across multiple LMs to explore their commonalities. In contrast, we focus on the commonalities among LMs, which can deepen our understanding of LMs and even facilitate cross-model knowledge transfer. In this work, we introduce the Order-Level Attention (OLA) derived from the order-wise decomposition of Attention Rollout and reveal that the OLA at the same order across LMs exhibits significant similarities. Furthermore, we discover an implicit mapping between OLA and syntactic knowledge. Based on these two findings, we propose the Transferable OLA Adapter (TOA), a training-free cross-LM adapter transfer method. Specifically, we treat the OLA as a unified syntactic feature representation and train an adapter that takes OLA as input. Due to the similarities in OLA across LMs, the adapter generalizes to unseen LMs without requiring any parameter updates. Extensive experiments demonstrate that TOA's cross-LM generalization effectively enhances the performance of unseen LMs. Code is available at \url{https://github.com/jinglin-liang/OLAS}.

Federated learning (FL) enables multiple clients to collaboratively learn a shared model without sharing their individual data. Concerns about utility, privacy, and training efficiency in FL have garnered significant research attention. Differential privacy has emerged as a prevalent technique in FL, safeguarding the privacy of individual user data while impacting utility and training efficiency. Within Differential Privacy Federated Learning (DPFL), previous studies have primarily focused on the utility-privacy trade-off, neglecting training efficiency, which is crucial for timely completion. Moreover, differential privacy achieves privacy by introducing controlled randomness (noise) on selected clients in each communication round. Previous work has mainly examined the impact of noise level ($\sigma$) and communication rounds ($T$) on the privacy-utility dynamic, overlooking other influential factors like the sample ratio ($q$, the proportion of selected clients). This paper systematically formulates an efficiency-constrained utility-privacy bi-objective optimization problem in DPFL, focusing on$\sigma$,$T$, and$q$. We provide a comprehensive theoretical analysis, yielding analytical solutions for the Pareto front. Extensive empirical experiments verify the validity and efficacy of our analysis, offering valuable guidance for low-cost parameter design in DPFL.

Federated Learning (FL) has emerged as a promising approach for collaborative model training without sharing private data. However, privacy concerns regarding information exchanged during FL have received significant research attention.Gradient Inversion Attacks (GIAs)have been proposed to reconstruct the private data retained by local clients from the exchanged gradients. While recovering private data, the data dimensions and the model complexity increase, which thwart data reconstruction by GIAs. Existing methods adopt prior knowledge about private data to overcome those challenges. In this article, we first observe that GIAs with gradients from a single iteration fail to reconstruct private data due to insufficient dimensions of leaked gradients, complex model architectures, and invalid gradient information. We investigate a Temporal Gradient Inversion Attack with a Robust Optimization framework, called TGIAs-RO, which recovers private data without any prior knowledge by leveraging multiple temporal gradients. To eliminate the negative impacts of outliers, e.g., invalid gradients for collaborative optimization, robust statistics are proposed. Theoretical guarantees on the recovery performance and robustness of TGIAs-RO against invalid gradients are also provided. Extensive empirical results on MNIST, CIFAR10, ImageNet and Reuters 21578 datasets show that the proposed TGIAs-RO with 10 temporal gradients improves reconstruction performance compared to state-of-the-art methods, even for large batch sizes (up to 128), complex models like ResNet18, and large datasets like ImageNet (224× 224pixels). Furthermore, the proposed attack method inspires further exploration of privacy-preserving methods in the context of FL.

Federated learning (FL) is a distributed machine learning paradigm allowing multiple clients to collaboratively train a global model without sharing their local data. However, FL entails exposing the model to various participants. This poses a risk of unauthorized model distribution or resale by the malicious client, compromising the intellectual property rights of the FL group. To deter such misbehavior, it is essential to establish a mechanism for verifying the ownership of the model and as well tracing its origin to the leaker among the FL participants. In this paper, we present FedTracker, the first FL model protection framework that provides both ownership verification and traceability. FedTracker adopts a bi-level protection scheme consisting of global watermark mechanism and local fingerprint mechanism. The former authenticates the ownership of the global model, while the latter identifies which client the model is derived from. FedTracker leverages Continual Learning (CL) principles to embed the watermark in a way that preserves the utility of the FL model on both primitive task and watermark task. FedTracker also devises a novel metric to better discriminate different fingerprints. Experimental results show FedTracker is effective in ownership verification, traceability, and maintains good fidelity and robustness against various watermark removal attacks.

Foundation Models (FMs) such as GPT-4 encoded with vast knowledge and powerful emergent abilities have achieved remarkable success in various natural language processing and computer vision tasks. Grounding FMs by adapting them to domain-specific tasks or augmenting them with domain-specific knowledge enables us to exploit the full potential of FMs. However, grounding FMs faces several challenges, stemming primarily from constrained computing resources, data privacy, model heterogeneity, and model ownership. Federated Transfer Learning (FTL), the combination of federated learning and transfer learning, provides promising solutions to address these challenges. Recently, the need for grounding FMs leveraging FTL, coined FTL-FM, has arisen strongly in both academia and industry. Motivated by the strong growth in FTL-FM research and the potential impact of FTL-FM on industrial applications, we propose an FTL-FM framework that formulates problems of grounding FMs in the federated learning setting, construct a detailed taxonomy based on the FTL-FM framework to categorize state-of-the-art FTL-FM works, and comprehensively overview FTL-FM works based on the proposed taxonomy. We also establish correspondence between FTL-FM and conventional phases of adapting FM so that FM practitioners can align their research works with FTL-FM. In addition, we overview advanced efficiency-improving and privacy-preserving techniques because efficiency and privacy are critical concerns in FTL-FM. Last, we discuss opportunities and future research directions of FTL-FM.

Federated Foundation Models (FedFMs) represent a distributed learning paradigm that fuses general competences of foundation models as well as privacy-preserving capabilities of federated learning. This combination allows the large foundation models and the small local domain models at the remote clients to learn from each other in a teacher-student learning setting. This paper provides a comprehensive summary of the ten challenging problems inherent in FedFMs, encompassing foundational theory, utilization of private data, continual learning, unlearning, Non-IID and graph data, bidirectional knowledge transfer, incentive mechanism design, game mechanism design, model watermarking, and efficiency. The ten challenging problems manifest in five pivotal aspects: “Foundational Theory,” which aims to establish a coherent and unifying theoretical framework for FedFMs. “Data,” addressing the difficulties in leveraging domain-specific knowledge from private data while maintaining privacy; “Heterogeneity,” examining variations in data, model, and computational resources across clients; “Security and Privacy,” focusing on defenses against malicious attacks and model theft; and “Efficiency,” highlighting the need for improvements in training, communication, and parameter efficiency. For each problem, we offer a clear mathematical definition on the objective function, analyze existing methods, and discuss the key challenges and potential solutions. This in-depth exploration aims to advance the theoretical foundations of FedFMs, guide practical implementations, and inspire future research to overcome these obstacles, thereby enabling the robust, efficient, and privacy-preserving FedFMs in various real-world applications.

Most federated learning-based recommender systems allow clients to access a well-trained high-quality model locally, which provides adversaries with the opportunity to infringe the legitimate copyright of the model. In response, we study an emerging and important problem, i.e., copyright protection of a federated recommendation model, which has not yet been addressed in the community of federated learning or recommender systems. We propose the first backdoor-based ownership verification scheme for federated recommendation (OVFR), which enables the server to claim its ownership for a given suspicious recommendation model. First, we propose to generate a trigger set tailored to recommendation scenarios. In particular, we generate some fake users and items, and then construct a set of fake users with fake interaction records as a trigger set. Moreover, we ensure that the distribution of the popularity of the fake items follows a long-tailed distribution for the effectiveness of the incorporated watermarking. To provide robustness assurance, we propose two different hybrid strategies to make the embeddings of the fake items similar to those of the real items. Second, we focus on effectively learning from a trigger set for recommendation scenarios. In particular, we design an MSE loss function and a contrastive loss function for incorporating the backdoor-based watermarking into the item embeddings, since the item embeddings are often more valuable and easier to be accessed than other parameters of a federated recommendation model. We then design a contrastive loss function to reduce the risk of the fake items being detected. Extensive experiments on three public datasets show the effectiveness of our OVFR in terms of ownership verification, model performance, and robustness.

Federated learning empowers privacy-preserving, multi-party secure model training without the necessity of sharing raw data. In recent years, knowledge distillation has emerged as a promising solution to address the significant challenge of model heterogeneity within federated learning. However, current research often overlooks the potential threats posed by Byzantine attacks, which can significantly compromise the security of federated distillation. Previous work on Byzantine attacks has been primarily focused on manipulating local gradients to compromise global model, lacking attacks on logits in knowledge distillation scenarios. In this paper, we introduce two innovative attacks, shedding light on the inherent risks in federated distillation. The proposed attacks include a top-k attack, which perturbs the top k values of logits in each column, and an impersonation attack, which emulates knowledge significantly deviating from the norm. To counter such attacks, we propose a robust aggregation strategy-FedTGD (Federated Top Guard Distillation), designed to ensure robust distillation with heterogeneous models. Specifically, FedTGD incorporates Density-Based Spatial Clustering of Applications with Noise (DBSCAN) and maximum cosine similarity on top-k values of logits to select benign knowledge. Experimental evaluations conducted on FEMNIST and CIFAR100 datasets, considering scenarios for both IID and Non-IID, reveal that top-k attack results in a substantial 27.16% accuracy reduction for FedMD. In contrast, our aggregation method shows a marginal 0.7% accuracy decrease under top-k attacks, outperforming state-of-the-art baselines.

The advent of Federated Learning (FL) highlights the practical necessity for the ’right to be forgotten’ for all clients, allowing them to request data deletion from the machine learning model’s service provider. This necessity has spurred a growing demand for Federated Unlearning (FU). Feature unlearning has gained considerable attention due to its applications in unlearning sensitive, backdoor, and biased features. Existing methods employ the influence function to achieve feature unlearning, which is impractical for FL as it necessitates the participation of other clients, if not all, in the unlearning process. Furthermore, current research lacks an evaluation of the effectiveness of feature unlearning. To address these limitations, we define feature sensitivity in evaluating feature unlearning according to Lipschitz continuity. This metric characterizes the model output’s rate of change or sensitivity to perturbations in the input feature. We then propose an effective federated feature unlearning framework called Ferrari, which minimizes feature sensitivity. Extensive experimental results and theoretical analysis demonstrate the effectiveness of Ferrari across various feature unlearning scenarios, including sensitive, backdoor, and biased features. The code is publicly available at https://github.com/OngWinKent/Federated-Feature-Unlearning

This paper proposes a general spectral analysis framework that thwarts a security risk in federated Learning caused by groups of malicious Byzantine attackers or colluders, who conspire to upload vicious model updates to severely debase global model performances. The proposed framework delineates the strong consistency and temporal coherence between Byzantine colluders' model updates from a spectral analysis lens, and, formulates the detection of Byzantine misbehaviours as a community detection problem in weighted graphs. The modified normalized graph cut is then utilized to discern attackers from benign participants. Moreover, the Spectral heuristics is adopted to make the detection robust against various attacks. The proposed Byzantine colluder resilient method, i.e., FedCut, is guaranteed to converge with bounded errors. Extensive experimental results under a variety of settings justify the superiority of FedCut, which demonstrates extremely robust model accuracy (MA) under various attacks. It was shown that FedCut's averaged MA is 2.1% to 16.5% better than that of the state of the art Byzantine-resilient methods. In terms of the worst-case model accuracy (MA), FedCut is 17.6% to 69.5% better than these methods.

Conventionally, federated learning aims to optimize a single objective, typically the utility. However, for a federated learning system to be trustworthy, it needs to simultaneously satisfy multiple objectives, such as maximizing model performance, minimizing privacy leakage and training costs, and being robust to malicious attacks. Multi-Objective Optimization (MOO) aiming to optimize multiple conflicting objectives simultaneously is quite suitable for solving the optimization problem of Trustworthy Federated Learning (TFL). In this article, we unify MOO and TFL by formulating the problem of constrained multi-objective federated learning (CMOFL). Under this formulation, existing MOO algorithms can be adapted to TFL straightforwardly. Different from existing CMOFL algorithms focusing on utility, efficiency, fairness, and robustness, we consider optimizing privacy leakage along with utility loss and training cost, the three primary objectives of a TFL system. We develop two improved CMOFL algorithms based on NSGA-II and PSL, respectively, to effectively and efficiently find Pareto optimal solutions and provide theoretical analysis on their convergence. We design quantitative measurements of privacy leakage, utility loss, and training cost for three privacy protection mechanisms: Randomization, BatchCrypt (an efficient homomorphic encryption), and Sparsification. Empirical experiments conducted under the three protection mechanisms demonstrate the effectiveness of our proposed algorithms.

Vertical federated learning (VFL) allows an active party with labeled data to leverage auxiliary features from the passive parties to improve model performance. Concerns about the private feature and label leakage in both the training and inference phases of VFL have drawn wide research attention. In this paper, we propose a general privacy-preserving vertical federated deep learning framework called FedPass, which leverages adaptive obfuscation to protect the feature and label simultaneously. Strong privacy-preserving capabilities about private features and labels are theoretically proved (in Theorems 1 and 2). Extensive experimental results with different datasets and network architectures also justify the superiority of FedPass against existing methods in light of its near-optimal trade-off between privacy and model performance.

Cross-platform recommendation aims to improve recommendation accuracy by gathering heterogeneous features from different platforms. However, such cross-silo collaborations between platforms are restricted by increasingly stringent privacy protection regulations, thus data cannot be aggregated for training. Federated learning (FL) is a practical solution to deal with the data silo problem in recommendation scenarios. Existing cross-silo FL methods transmit model information to collaboratively build a global model by leveraging the data of overlapped users. However, in reality, the number of overlapped users is often very small, thus largely limiting the performance of such approaches. Moreover, transmitting model information during training requires high communication costs and may cause serious privacy leakage. In this paper, we propose a novel privacy-preserving double distillation framework named FedPDD for cross-silo federated recommendation, which efficiently transfers knowledge when overlapped users are limited. Specifically, our double distillation strategy enables local models to learn not only explicit knowledge from the other party but also implicit knowledge from its past predictions. Moreover, to ensure privacy and high efficiency, we employ an offline training scheme to reduce communication needs and privacy leakage risk. In addition, we adopt differential privacy to further protect the transmitted information. The experiments on two real-world recommendation datasets, HetRec-MovieLens and Criteo, demonstrate the effectiveness of FedPDD compared to the state-of-the-art approaches.

Federated learning models are collaboratively developed upon valuable training data owned by multiple parties. During the development and deployment of federated models, they are exposed to risks including illegal copying, re-distribution, misuse and/or free-riding. To address these risks, the ownership verification of federated learning models is a prerequisite that protects federated learning model intellectual property rights (IPR) i.e., FedIPR. We propose a novel federated deep neural network (FedDNN) ownership verification scheme that allows private watermarks to be embedded and verified to claim legitimate IPR of FedDNN models. In the proposed scheme, each client independently verifies the existence of the model watermarks and claims respective ownership of the federated model without disclosing neither private training data nor private watermark information. The effectiveness of embedded watermarks is theoretically justified by the rigorous analysis of conditions under which watermarks can be privately embedded and detected by multiple clients. Moreover, extensive experimental results on computer vision and natural language processing tasks demonstrate that varying bit-length watermarks can be embedded and reliably detected without compromising original model performances. Our watermarking scheme is also resilient to various federated training settings and robust against removal attacks.

In a federated learning scenario where multiple parties jointly learn a model from their respective data, there exist two conflicting goals for the choice of appropriate algorithms. On one hand, private and sensitive training data must be kept secure as much as possible in the presence of semi-honest partners; on the other hand, a certain amount of information has to be exchanged among different parties for the sake of learning utility. Such a challenge calls for the privacy-preserving federated learning solution, which maximizes the utility of the learned model and maintains a provable privacy guarantee of participating parties’ private data. This article illustrates a general framework that (1) formulates the trade-off between privacy loss and utility loss from a unified information-theoretic point of view, and (2) delineates quantitative bounds of the privacy-utility trade-off when different protection mechanisms including randomization, sparsity, and homomorphic encryption are used. It was shown that in general there is no free lunch for the privacy-utility trade-off , and one has to trade the preserving of privacy with a certain degree of degraded utility. The quantitative analysis illustrated in this article may serve as the guidance for the design of practical federated learning algorithms.