اشترك بالحزمة الذهبية واحصل على وصول غير محدود شمرا أكاديميا
تسجيل مستخدم جديدByzantine-Tolerant Machine Learning
84
0
0.0
(
0
)
اسأل ChatGPT حول البحث
ﻻ يوجد ملخص باللغة العربية
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 (SGD), the heart of most machine learning algorithms. Assuming a set of $n$ workers, up to $f$ of them being Byzantine, we ask how robust can SGD be, without limiting the dimension, nor the size of the parameter space. We first show that no gradient descent update rule based on a linear combination of the vectors proposed by the workers (i.e, current approaches) tolerates a single Byzantine failure. We then formulate a resilience property of the update rule capturing the basic requirements to guarantee convergence despite $f$ Byzantine workers. We finally propose Krum, an update rule that satisfies the resilience property aforementioned. For a $d$-dimensional learning problem, the time complexity of Krum is $O(n^2 cdot (d + log n))$.
قيم البحث
اقرأ أيضاً
This work presents a new distributed Byzantine tolerant federated learning algorithm, HoldOut SGD, for Stochastic Gradient Descent (SGD) optimization. HoldOut SGD uses the well known machine learning technique of holdout estimation, in a distributed
fashion, in order to select parameter updates that are likely to lead to models with low loss values. This makes it more effective at discarding Byzantine workers inputs than existing methods that eliminate outliers in the parameter-space of the learned model. HoldOut SGD first randomly selects a set of workers that use their private data in order to propose gradient updates. Next, a voting committee of workers is randomly selected, and each voter uses its private data as holdout data, in order to select the best proposals via a voting scheme. We propose two possible mechanisms for the coordination of workers in the distributed computation of HoldOut SGD. The first uses a truthful central server and corresponds to the typical setting of current federated learning. The second is fully distributed and requires no central server, paving the way to fully decentralized federated learning. The fully distributed version implements HoldOut SGD via ideas from the blockchain domain, and specifically the Algorand committee selection and consensus processes. We provide formal guarantees for the HoldOut SGD process in terms of its convergence to the optimal model, and its level of resilience to the fraction of Byzantine workers. Empirical evaluation shows that HoldOut SGD is Byzantine-resilient and efficiently converges to an effectual model for deep-learning tasks, as long as the total number of participating workers is large and the fraction of Byzantine workers is less than half (<1/3 for the fully distributed variant).
This paper proposes the first implementation of an atomic storage tolerant to mobile Byzantine agents. Our implementation is designed for the round-based synchronous model where the set of Byzantine nodes changes from round to round. In this model we
explore the feasibility of multi-writer multi-reader atomic register prone to various mobile Byzantine behaviors. We prove upper and lower bounds for solving the atomic storage in all the explored models. Our results, significantly different from the static case, advocate for a deeper study of the main building blocks of distributed computing while the system is prone to mobile Byzantine failures.
This paper proposes the first implementation of a self-stabilizing regular register emulated by $n$ servers that is tolerant to both mobile Byzantine agents, and emph{transient failures} in a round-free synchronous model. Differently from existing Mo
bile Byzantine tolerant register implementations, this paper considers a more powerful adversary where (i) the message delay (i.e., $delta$) and the period of mobile Byzantine agents movement (i.e., $Delta$) are completely decoupled and (ii) servers are not aware of their state i.e., they do not know if they have been corrupted or not by a mobile Byzantine agent.The proposed protocol tolerates emph{(i)} any number of transient failures, and emph{(ii)} up to $f$ Mobile Byzantine agents. In addition, our implementation uses bounded timestamps from the $mathcal{Z}_{13}$ domain and it is optimal with respect to the number of servers needed to tolerate $f$ mobile Byzantine agents in the given model.
The appeal of serverless (FaaS) has triggered a growing interest on how to use it in data-intensive applications such as ETL, query processing, or machine learning (ML). Several systems exist for training large-scale ML models on top of serverless in
frastructures (e.g., AWS Lambda) but with inconclusive results in terms of their performance and relative advantage over serverful infrastructures (IaaS). In this paper we present a systematic, comparative study of distributed ML training over FaaS and IaaS. We present a design space covering design choices such as optimization algorithms and synchronization protocols, and implement a platform, LambdaML, that enables a fair comparison between FaaS and IaaS. We present experimental results using LambdaML, and further develop an analytic model to capture cost/performance tradeoffs that must be considered when opting for a serverless infrastructure. Our results indicate that ML training pays off in serverless only for models with efficient (i.e., reduced) communication and that quickly converge. In general, FaaS can be much faster but it is never significantly cheaper than IaaS.
For mitigating Byzantine behaviors in federated learning (FL), most state-of-the-art approaches, such as Bulyan, tend to leverage the similarity of updates from the benign clients. However, in many practical FL scenarios, data is non-IID across clien
ts, thus the updates received from even the benign clients are quite dissimilar. Hence, using similarity based methods result in wasted opportunities to train a model from interesting non-IID data, and also slower model convergence. We propose DiverseFL to overcome this challenge in heterogeneous data distribution settings. Rather than comparing each clients update with other client updates to detect Byzantine clients, DiverseFL compares each clients update with a guiding update of that client. Any client whose update diverges from its associated guiding update is then tagged as a Byzantine node. The FL server in DiverseFL computes the guiding update in every round for each client over a small sample of the clients local data that is received only once before start of the training. However, sharing even a small sample of clients data with the FL server can compromise clients data privacy needs. To tackle this challenge, DiverseFL creates a Trusted Execution Environment (TEE)-based enclave to receive each clients sample and to compute its guiding updates. TEE provides a hardware assisted verification and attestation to each client that its data is not leaked outside of TEE. Through experiments involving neural networks, benchmark datasets and popular Byzantine attacks, we demonstrate that DiverseFL not only performs Byzantine mitigation quite effectively, it also almost matches the performance of OracleSGD, where the server only aggregates the updates from the benign clients.
الأسئلة المقترحة
سجل دخول لتتمكن من نشر تعليقات
التعليقات
جاري جلب التعليقات
سجل دخول لتتمكن من متابعة معايير البحث التي قمت باختيارها
