Brief Announcement: Byzantine-Tolerant Machine Learning
نویسندگان
چکیده
We report on Krum, the rst provably Byzantine-tolerant aggregation rule for distributed Stochastic Gradient Descent (SGD). Krum guarantees the convergence of SGD even in a distributed setting where (asymptotically) up to half of the workers can be malicious adversaries trying to attack the learning system.
منابع مشابه
Brief Announcement: Practical Synchronous Byzantine Consensus
This paper presents new protocols for Byzantine state machine replication and Byzantine agreement in the synchronous and authenticated setting. The PBFT state machine replication protocol tolerates f Byzantine faults in an asynchronous setting using n = 3f + 1 replicas. We improve the Byzantine fault tolerance to n = 2f + 1 by utilizing the synchrony assumption. Our protocol also solves synchro...
متن کاملBrief Announcement: When You Don't Trust Clients: Byzantine Proposer Fast Paxos
State machine replication is a general approach for constructing fault-tolerant services, and a key protocol underlying state machine replication is consensus. The set of Byzantine failures is so large that it has been applied for masking the effects of compromised systems, and so Byzantine-tolerant consensus has been used to construct systems that are meant to ameliorate the effect of compromi...
متن کاملByzantine-Tolerant Machine Learning
The growth of data, the need for scalability and the complexity of models used in modern machine learning calls for distributed implementations. Yet, as of today, distributed machine learning frameworks have largely ignored the possibility of arbitrary (i.e., Byzantine) failures. In this paper, we study the robustness to Byzantine failures at the fundamental level of stochastic gradient descent...
متن کاملFault-Tolerant Multi-Agent Optimization: Part III
We study fault-tolerant distributed optimization of a sum of convex (cost) functions with real-valued scalar input/output in the presence of crash faults or Byzantine faults. In particular, the goal is to optimize a global cost function 1 n ∑ i∈V hi(x), where V = {1, . . . , n} is the collection of agents, and hi(x) is agent i’s local cost function, which is initially known only to agent i. Thi...
متن کاملBrief Announcement: Decoupled Quorum-Based Byzantine-Resilient Coordination in Open Distributed Systems
The tuple space coordination model, originally introduced in the Linda programming language [2], uses a shared memory object called a tuple space to support coordination that is decoupled both in time – processes do not have to be active at the same time – and space – processes do not need to know each others’ addresses. The tuple space can be considered to be a kind of storage that stores tupl...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
عنوان ژورنال:
دوره شماره
صفحات -
تاریخ انتشار 2017