Advances and Open Problems in Federated Learning
Open accessThe DOI matches its Crossref record; 509 of the 514 references it cites were found in a registry or an index; 188 of the 503 sources its claims cite were checked for support.
Abstract
Federated learning (FL) is a machine learning setting where many clients (e.g., mobile devices or whole organizations) collaboratively train a model under the orchestration of a central server (e.g., service provider), while keeping the training data decentralized. FL embodies the principles of focused data collection and minimization, and can mitigate many of the systemic privacy risks and costs resulting from traditional, centralized machine learning and data science approaches. Motivated by the explosive growth in FL research, this monograph discusses recent advances and presents an extensive collection of open problems and challenges.
Overview
CoChat Summary
TL;DR
This paper is a broad survey of federated learning, and the provided text focuses on privacy, security, robustness, and fairness challenges rather than reporting a single empirical study. It argues that FL changes the attack surface and makes many standard defenses harder to deploy because of decentralization, limited client visibility, and communication constraints.
Key findings
- Federated learning increases exposure to white-box evasion attacks because the model broadcast step means the model may be accessible to any malicious client; the text says this “increases the need for defenses against white-box evasion attacks.”
- For model update poisoning, the paper notes that targeted attacks can be strong even with few adversaries: it reports that 10% of participating devices compromised can be enough to introduce a backdoor, even with anomaly detectors at the server.
- Differential privacy is presented as a defense with two explicit strengths: it gives strong, worst-case protections and is closed under composition; but the paper also notes that in FL it can reduce accuracy and its hyperparameters are still hard to choose.
- Under secure aggregation, the server sees only aggregates, which helps privacy, but the paper says the method has limitations: it assumes a semi-honest server in key setup, can still leak through per-round aggregates, is not efficient for sparse vectors, and does not enforce well-formed client inputs.
Methods
- Not a single experimental study; this is a survey / open-problems paper organizing the FL literature across settings, efficiency, privacy, attacks, and fairness.
- The text synthesizes prior work and uses qualitative analysis of threat models, privacy mechanisms (e.g., DP, secure aggregation, MPC, TEEs), and optimization results; it also reproduces convergence-rate tables for IID and non-IID federated optimization methods.
- It discusses the FL setting using a cross-device / cross-silo / decentralized taxonomy and references typical FL workflows, datasets, and software frameworks rather than introducing new data collection.
Limitations
- The paper explicitly states that many open questions remain: e.g., how to defend against inference-time attacks in FL, how to reconcile privacy with robust aggregation, and how to make secure aggregation compatible with non-linear defenses.
- It says that formal analysis is often missing or incomplete for several areas, including split learning privacy guarantees, which are described as “still an open problem,” and the compatibility of some defenses with secure aggregation.
- Evident from the text: because this is a survey, it does not present new experiments or a unified benchmark; conclusions are drawn from existing work and summarized tables, not from a new empirical evaluation.
Model-written summary of a shortened version of the paper’s full text (it is longer than we read). Check it against the paper before relying on it.
Integrity signals
Registry
Cites
Statistics
Figures




Verification
Claims the paper makes
33 Valid · 129 Not valid · 141 Not checked · 285 Own results and interpretations
Some cited sources support this claim and some don’t — read each source’s verdict below.
In the paper
“This suggests, as shown by Reyzin et al. [392] for practical settings, that assuming honest (or at least semi-honest) behaviour by the server in a key distribution phase (as done in [80, 58]) is required in scenarios where private channels among clients are needed.”
What the sources say
- Disputedcites [392] Niu, 2019, Secure federated submodel learning sourceExperimental — accuracy not yet measured
In “B. Building Blocks” (p. 8), the paper says clients use the server as a relay but establish their secret key through client-to-client Diffie–Hellman exchange, with encryption to protect messages against the intermediary server and signatures to defend against active adversaries. It contrasts this setup with approaches that rely on a trusted third party for key generation and distribution, rather than stating that its setup requires an honest server during key distribution. (LLM, conf=0.78)
Checked by a reasoning model — may be wrong.
- Not matchedcites [80] Ceballos, 2018, SplitNN-driven vertical partitioning
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Partly supportedcites [58] Cyffers, 2020, Privacy amplification by decentralization sourceExperimental — accuracy not yet measured
Section 2 (p. 3) says the model’s restriction on who can see messages can be enforced with secure communication channels, supporting the private-channels part of the claim. It does not describe a server-run key distribution phase or require an honest server; Section 3.3 (p. 5) instead says the approach does not rely on a central server. (LLM, conf=0.85)
hedgedstates a cause- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: passes — The paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Disputed
The cited sources support only part of this claim — read each source’s verdict below.
In the paper
“Significant related work predates the introduction of the term federated learning.”
What the sources say
- Not matchedcites [396] Cummings, 2019, Bringing differential privacy into the experimental economics lab: Theory and an application to a public-good game
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites [492] Zafar
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Partly supportedcites [11] Hard, 2018, Federated learning for mobile keyboard prediction sourceExperimental — accuracy not yet measured
The Related Work section (p. 2) describes federated learning as an extension of server-based distributed training, supporting the existence of antecedent research. It does not say when the term “federated learning” was introduced or establish that significant related work predates that introduction. (LLM, conf=0.72)
- Not matchedcites [457] Ma
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Not matched
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“Further, [469] provides a convergence analysis for the non-IID data case, but for the specific scheme based on matching decomposition sampling described above.”
What the sources say
- Could not verifycites [469] Li, 2019, Fair resource allocation in federated learning sourceExperimental — accuracy not yet measured
Section 4.4 (p. 9) and Appendix E.2 (p. 27) report empirical convergence comparisons under heterogeneous data. Searches for “matching decomposition sampling” and “convergence analysis non-IID data scheme sampling” found no passage describing that scheme or an analysis of its convergence, so our copy does not verify the attributed claim. (LLM, conf=0.76)
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Could not verify
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“While large companies or organizations can play this role in some application scenarios, a reliable and powerful central server may not always be available or desirable in more collaborative learning scenarios [459].”
What the sources say
- Could not verifycites [459] Lécuyer sourceExperimental — accuracy not yet measured
The Introduction and Conclusion discuss certified defenses against adversarial examples, not the availability or desirability of a central server. Searches for the quoted claim and for central servers in collaborative or distributed learning found no passage addressing that point. (LLM, conf=0.75)
hedged- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Could not verify
The cited sources that could be checked couldn’t confirm or dispute this claim.
In the paper
“Currently proposed algorithms include trees [118], linear and logistic regression [490, 224, 316], and neural networks [317].”
What the sources say
- Not matchedcites [118] Xie, 2019, Local adaalter: Communication-efficient stochastic gradient descent with adaptive learning rates
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Could not verifycites [490] Awasthi, 2020, Beyond individual and group fairness, CoRR sourceExperimental — accuracy not yet measured
The Introduction (p. 1) and Related Work (p. 12) discuss algorithmic fairness, not feature-partitioned cross-silo federated learning. Searches for that setting and for logistic regression found no passage describing the claimed algorithms; the paper’s mention of linear programming in Experiments (p. 11) is not linear regression. (LLM, conf=0.76)
- Not matchedcites [224] Bogdanov
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not matchedcites [316] Gennaro
Couldn’t be matched to a paper in Crossref, so it can’t be checked.
- Not checked — no full text of the sourcecites [317] Goldwasser, 1989, The knowledge complexity of interactive proof systems, SIAM J. Comput. source
plain- Asserted: passes — The authors put it forward as true.
- Checkable: passes — It can be checked against evidence.
- Load-bearing: unclear — It is unclear whether the paper’s conclusions depend on it.
- Based: passes — The paper gives a reason why it holds.
- Not matched
190 more claims · 51 with their cited sources checked
Create a free account to see all 588 claims and to verify the 13 references still open.
- Probably not a substantial claim. 309
- It may state two separate claims as one. 161
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
The source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“TLS and TEEs can help mitigate some of the security and privacy concerns that remain for FL55.”
From Federated learning enables big data for rare cancer boundary detection (2022) · cites this paper as [55]
Experimental — accuracy not yet measuredThe source backs part of the claim, not all of it — or backs it for a different population, condition or comparison.
In the citing paper
“Patient data are bound to the institution that holds them by regulation, by sovereignty, and by ethics, so a commons of raw clinical data is foreclosed; what can be pooled is learning, through the federated-averaging mechanism of McMahan et al. (2017) and the broader apparatus surveyed by Kairouz et al. (2021), in which each hospital contributes computation local to its own records and only model”
From Commons-Governed Artificial Intelligence: A Taxonomy of Collective Governance (2026) · cites this paper as Kairouz et al. (2021)
Experimental — accuracy not yet measuredThe cited source says what the claim attributes to it.
In the citing paper
“Federated learning, introduced by McMahan et al. (2017) as communication-efficient training of a shared model over decentralized data through iterative averaging of locally computed updates, and surveyed comprehensively by Kairouz et al. (2021), lets a community train a common model without centralizing either the raw data or, in principle, the compute, since each participant contributes the”
From Commons-Governed Artificial Intelligence: A Taxonomy of Collective Governance (2026) · cites this paper as Kairouz et al., 2021
The check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“The application of the adaptive federated reinforcement learning approach thus can achieve high accuracy rate in attack detection (over 82%) with fast convergence rate and high learning rewards.”
From Federated Learning for Internet of Things: A Comprehensive Survey (2021) · cites this paper as [149]
Experimental — accuracy not yet measuredThe check wasn’t confident either way, so it gives no verdict — never disputed.
In the citing paper
“The application of the adaptive federated reinforcement learning approach thus can achieve high accuracy rate in attack detection (over 82%) with fast convergence rate and high learning rewards.”
From Federated Learning for Internet of Things: A Comprehensive Survey (2021) · cites this paper as [149]
Experimental — accuracy not yet measured
4 more claims about this paper
Create a free account to see every claim other papers make about this one.
Citations
Cites
Elements of Information Theory
Cover, T. M., & Thomas, J. A. (2001) · Wiley eBooks · doi:10.1002/0471200611 · Cited by 36,808
PyTorch: An Imperative Style, High-Performance Deep Learning Library
Paszke, A., Gross, S., Massa, F., Lerer, A., Bradbury, J., Chanan, G., Killeen, T., Lin, Z., Gimelshein, N., Antiga, L., et al. (2019) · arXiv · doi:10.48550/arxiv.1912.01703 · Cited by 31,473
Pan, S. J., & Yang, Q. (2010) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2009.191 · Cited by 23,387 · cited 1×
“Federated transfer learning [419] is another concept that considers challenging scenarios in which data parties share only a partial overlap in the user space or the feature space, and leverage existing transfer learning techniques [314] to build models collaboratively.”
BackgroundSelf-citationCommunication-Efficient Learning of Deep Networks from Decentralized Data
McMahan, H. B., Moore, E., Ramage, D., Hampson, S., & Arcas, B. A. Y. (2016) · arXiv · doi:10.48550/arxiv.1602.05629 · Cited by 14,215 · cited 3×
“[1]: “We term our approach Federated Learning, since the learning task is solved by a loose federation of participating devices (which we refer to as clients) which are coordinated by a central server.”
BackgroundInfluentialSelf-citationExplaining and Harnessing Adversarial Examples
Goodfellow, I. J., Shlens, J., & Szegedy, C. (2014) · arXiv · doi:10.48550/arxiv.1412.6572 · Cited by 10,381
Towards Evaluating the Robustness of Neural Networks
Carlini, N., & Wagner, D. (2017) · 2017 IEEE Symposium on Security and Privacy (SP) · doi:10.1109/sp.2017.49 · Cited by 8,859
Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks
Finn, C., Abbeel, P., & Levine, S. (2017) · arXiv · doi:10.48550/arxiv.1703.03400 · Cited by 8,512 · cited 1×
“Recently, a class of algorithms referred to as model-agnostic meta-learning (MAML) have been developed that meta-learn a global model, which can be used as a starting point for learning a good model adapted to a given task, using only a few local gradient steps [165].”
Background(2010) · Cited by 8,261
Intriguing properties of neural networks
Szegedy, C., Zaremba, W., Sutskever, I., Bruna, J., Erhan, D., Goodfellow, I., & Fergus, R. (2013) · arXiv · doi:10.48550/arxiv.1312.6199 · Cited by 8,143 · cited 4×
“attacks” to refer to attacks on model performance, not on data inference. Examples of adversarial attacks include data poisoning [63, 277], model update poisoning [42, 61], and model evasion attacks [377, 63, 186]. These attacks can be broadly classified into training-time attacks (poisoning attacks) and inference-time…”
BackgroundMethodInfluentialCalibrating Noise to Sensitivity in Private Data Analysis
Dwork, C., McSherry, F., Nissim, K., & Smith, A. (2006) · Lecture Notes in Computer Science · doi:10.1007/11681878_14 · Cited by 7,272
Research works printed but not linked: 37
- McMahan, 2017, Federated learning: Collaborative machine learning without centralized training data
- Differential Privacy Team, 2017, Learning with privacy at scale, Apple Machine Learning Journal, 1
- Pichai, 2019, Google’s Sundar Pichai: Privacy should not be a luxury good
- Apple , Private federated learning (NeurIPS 2019 Expo Talk Abstract), https://nips.cc/ExpoConferences/2019/schedule?talk_id=40, 2019.
- Abadi, 2015, Tensor Flow: Large-scale machine learning on heterogeneous systems
- Tang, D2: Decentralized training over decentralized data
- Koloskova, Decentralized deep learning with arbitrary communication compression
- Jaderberg, Decoupled neural interfaces using synthetic gradients, 1627
- Wang, Adaptive communication strategies for best error-runtime trade-offs in communication-efficient distributed SGD
- He, Milenas: Efficient neural architecture search via mixed-level reformulation
- Han, Geometric lower bounds for distributed parameter estimation under communication constraints, 1
- Bell, Secure single-server aggregation with (poly)logarithmic overhead, 1253
- Andrew, How to generate and exchange secrets (extended abstract), 162
- Nikolaenko, Privacy-preserving ridge regression on hundreds of millions of records, 334
- Barak, 2019, Secure evaluation of quantized neural networks, IACR Cryptology ePrint Archive, 2019, 131
- 2020, Lattigo 2.0.0
- 2020, Microsoft SEAL (release 3.6)
- 2020, SHELL
- Intel, 2012, Architecture instruction set extensions programming reference
- Shi, Privacy-preserving aggregation of time-series data
- Lie, Glimmers: Resolving the privacy/trust quagmire, 94
- Sabater, 2020, Distributed differentially private averaging with improved utility and robustness to malicious parties
- Amin, Bounding user contributions: A bias-variance trade-off in differential privacy, 263
- Feldman, Privacy amplification by iteration, 521
- Balle, Advances in Neural Information Processing Systems, 4623
- Cummings, Differentially private change-point detection, 10825
- Cummings, Differential privacy for growing databases, 8864
- Tramèr, Stealing machine learning models via prediction APIs, 601
- Bassily, 2020, Practical locally private heavy hitters, J. Mach. Learn. Res., 21, 1
- Cummings, 2019, Bringing differential privacy into the experimental economics lab: Theory and an application to a public-good game
- Goodfellow, Explaining and harnessing adversarial examples
- Wang, 2020, Attack of the tails: Yes, you really can backdoor federated learning
- Barocas, 2019, Fairness and Machine Learning
- Richardson, 2019, Dirty data, bad predictions: How civil rights violations impact police data, predictive policing systems, and justice
- Sambasivan, ‘privacy is not for me, it’s for those rich women’: Performative privacy practices on mobile phones by women in South Asia, 127
- Madras, Learning adversarially fair and transferable representations
- Ramage, 2020, Federated analytics: Collaborative data science without data collection
Cited by
The future of digital health with federated learning
Rieke, N., Hancox, J., Li, W., Milletarì, F., Roth, H. R., Albarqouni, S., Bakas, S., Galtier, M. N., Landman, B. A., Maier-Hein, K., et al. (2020) · npj Digital Medicine · doi:10.1038/s41746-020-00323-1 · Cited by 4,015
Text Data Augmentation for Deep Learning
Shorten, C., Khoshgoftaar, T. M., & Furht, B. (2021) · Journal of Big Data · doi:10.1186/s40537-021-00492-0 · Cited by 1,740
SCAFFOLD: Stochastic Controlled Averaging for Federated Learning
Karimireddy, S. P., Kale, S., Mohri, M., Reddi, S. J., Stich, S. U., & Suresh, A. T. (2019) · arXiv · doi:10.48550/arxiv.1910.06378 · Cited by 1,734
Self-citationA review of applications in federated learning
Li, L., Fan, Y., Tse, M., & Lin, K.-Y. (2020) · Computers & Industrial Engineering · doi:10.1016/j.cie.2020.106854 · Cited by 1,559
Federated Learning for Healthcare Informatics
Xu, J., Glicksberg, B. S., Su, C., Walker, P., Bian, J., & Wang, F. (2021) · Journal of Healthcare Informatics Research · doi:10.1007/s41666-020-00082-4 · Cited by 1,542
Metadata
- Authors
- Kairouz, P., & McMahan, H. B.
- Affiliations
- Google Research, USA
- Google (United States) (US)
- Venue
- Foundations and Trends in Machine Learning
- Publisher
- Emerald
- Published
- 2021-06-23
- Volume
- 14
- Issue
- 1-2
- Pages
- 1-210
- ISSN
- 1935-8237 (ISSN-L)
- 1935-8245
- OpenAlex
- W2995022099
- Cited by
- 7,772
- References
- 451
- Citations by year
- 2026: 1786 · 2025: 1736 · 2024: 1359 · 2023: 1149 · 2022: 781 · 2021: 665
- FWCI
- 282.57
- Type
- journal article
- Language
- en
- Open access
- green
- arxiv.org
- Locations
- Foundations and Trends® in Machine Learning (publishedVersion)
- Warwick Research Archive Portal (University of Warwick) (acceptedVersion)
- arXiv (Cornell University) (submittedVersion)
- HAL (submittedVersion, other-oa)
- HAL (submittedVersion, cc-by)
- Infoscience (Ecole Polytechnique Fédérale de Lausanne) (submittedVersion, cc-by-nc-nd)
- LillOA (Université de Lille (University Of Lille)) (submittedVersion, other-oa)
- Oskar-Bordeaux (Universite de Bordeaux) (submittedVersion)
- Rare & Special e-Zone (The Hong Kong University of Science and Technology) (submittedVersion)
- arXiv (Cornell University)
- Funders
- Agence Nationale de la Recherche (ANR-16-CE23-0016, ANR-20-CE23-0015)
- Topics
- Physical Sciences › Computer Science › Artificial Intelligence › Privacy-Preserving Technologies in Data
- Physical Sciences › Computer Science › Artificial Intelligence › Stochastic Gradient Optimization Techniques
- Physical Sciences › Computer Science › Artificial Intelligence › Domain Adaptation and Few-Shot Learning
- Keywords
- federated learning, decentralized data, privacy-preserving machine learning, data minimization, open problems, mobile devices, data collection
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
The claim verdicts this report counts are experimental: their accuracy is not yet measured.
A language model drew the claims from the paper on 2026-10-02, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).
A second model labelled each claim on 2026-10-02 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).
Cited sources are checked with a language model’s help (gpt-6-sol).
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
How to cite
Cite CoChat for the checks on this page, and the paper for its own findings.
- This page
- CoChat. (2026). Citation check: Advances and Open Problems in Federated Learning. https://app.cochat.ai/papers/10.1561/2200000083
- The paper
- Kairouz, P., & McMahan, H. B. (2021). Advances and Open Problems in Federated Learning. Foundations and Trends in Machine Learning. https://doi.org/10.1561/2200000083
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.