• Corpus ID: 235458306

Optimality and Stability in Federated Learning: A Game-theoretic Approach

  title={Optimality and Stability in Federated Learning: A Game-theoretic Approach},
  author={Kate Donahue and Jon M. Kleinberg},
Federated learning is a distributed learning paradigm where multiple agents, each only with access to local data, jointly learn a global model. There has recently been an explosion of research aiming not only to improve the accuracy rates of federated learning, but also provide certain guarantees around social good properties such as total error. One branch of this research has taken a game-theoretic approach, and in particular, prior work has viewed federated learning as a hedonic game, where… 

Tables from this paper

Models of fairness in federated learning
For egalitarian fairness, a tight multiplicative bound on how widely error rates can diverge between agents federating together is obtained and it is shown that sub-proportional error is guaranteed for any individually rational federating coalition.
Collaboration Equilibrium in Federated Learning
This paper proposes the concept of benefit graph which describes how each client can benefit from collaborating with other clients and advances a Pareto optimization approach to identify the optimal collaborators and theoretically proves that it can reach a CE from the benefit graph through an iterative graph operation.
Sample Optimality and All-for-all Strategies in Personalized Federated and Collaborative Learning
This work introduces theoretical lower bounds on the number of samples required from all agents to approximately minimize the generalization error of a fixed agent, and provides strategies matching these lower bounds in the all-for-one and all- for-all settings where respectively one or all agents desire to minimize their own local function.
Mechanisms that Incentivize Data Sharing in Federated Learning
This work shows how a naive scheme leads to catastrophic levels of free-riding where the benefits of data sharing are completely eroded, and introduces accuracy shaping based mechanisms to maximize the amount of data generated by each agent.
To Federate or Not To Federate: Incentivizing Client Participation in Federated Learning
An algorithm called IncFL is proposed that explicitly maximizes the fraction of clients who are incentivized to use the global model by dynamically adjusting the aggregation weights assigned to their updates, and can also improve the generalization performance of theglobal model on unseen clients.
FadMan: Federated Anomaly Detection across Multiple Attributed Networks
The proposed algorithm FadMan is a vertical federated learning framework for public node aligned with many private nodes of different features, and is validated on two tasks — correlated anomalies detection on multiple attributed networks and anomaly detection on an attributeless network — using real-world datasets.


One for One, or All for All: Equilibria and Optimality of Collaboration in Federated Learning
A framework for incentiveaware learning and data sharing in federated learning is introduced and stable and envy-free equilibria capture notions of collaboration in the presence of agents interested in meeting their learning objectives while keeping their own sample collection burden low are captured.
Incentive Mechanism Design for Federated Learning: Hedonic Game Approach
The Nash-stable set is proposed which determines the family of hedonic games possessing at least one Nash- stable partition, and the conditions of non-emptiness of the Nash- unstable set are analyzed.
Model-sharing Games: Analyzing Federated Learning Under Voluntary Participation
This work derives exact expected MSE values for problems in linear regression and mean estimation and uses these values to analyze the resulting game in the framework of hedonic game theory; it constructively shows that there always exists a stable partition of players into coalitions.
An Incentive Mechanism for Federated Learning in Wireless Cellular Networks: An Auction Approach
This paper considers a FL system that involves one base station (BS) and multiple mobile users, and proposes the primal-dual greedy auction mechanism, which can guarantee three economic properties, namely, truthfulness, individual rationality and efficiency.
A Game-Theoretic Approach to Coalition Formation in Green Cloud Federations
An algorithm is devised, based on cooperative game theory, that can be readily implemented in a distributed fashion, and that allows a set of CPs to cooperatively set up their federations in such a way that their individual profit is increased with respect to the case in which they work in isolation.
A game-theoretic approach to coalition formation in fog provider federations
An algorithm is proposed, based on cooperative game theory, that enables each FIP to decide with whom to cooperate in order to increase its profits, and the effectiveness of the proposed algorithm is demonstrated through an experimental evaluation considering various workload intensities.
Fair Resource Allocation in Federated Learning
This work proposes q-Fair Federated Learning (q-FFL), a novel optimization objective inspired by fair resource allocation in wireless networks that encourages a more fair accuracy distribution across devices in federated networks.
The price of stability for network design with fair cost allocation
It is established that the fair cost allocation protocol is in fact a useful mechanism for inducing strategic behavior to form near-optimal equilibria, and its results are extended to cases in which users are seeking to balance network design costs with latencies in the constructed network.
The Stability of Hedonic Coalition Structures
This work considers the partitioning of a society into coalitions in purely hedonic settings, and shows that if coalitions can be ordered according to some characteristic over which players have single-peaked preferences, then there exists an individually stable coalition partition.
Game of Gradients: Mitigating Irrelevant Clients in Federated Learning
A principled approach is followed to address the above FRCS problems and a new federated learning method using the Shapley value concept from cooperative game theory is developed, which is presented as a cooperative game involving the gradients shared by the clients.