今天要跟大家分享的內(nèi)容是一篇聯(lián)邦學(xué)習(xí)領(lǐng)域方面比較全面的綜述文章:Advances and Open Problems in Federated Learning。該文章發(fā)表于2021年,由來(lái)自麻省理工、斯坦福、谷歌等25所國(guó)際知名高校(機(jī)構(gòu))的學(xué)者聯(lián)合所著,共調(diào)研了400余篇文獻(xiàn),內(nèi)容非常豐富。由于篇幅所限,這里聚焦于幾個(gè)基礎(chǔ)方面進(jìn)行分享,并進(jìn)行一定的補(bǔ)充。
Federated learning is a machine learning setting where multiple entities (clients) collaborate in solving a machine learning problem, under the coordination of a central server or service provider. Each client’s raw data is stored locally and not exchanged or transferred; instead, focused updates intended for immediate aggregation are used to achieve the learning objective.
(1)數(shù)據(jù)集1. EMNIST數(shù)據(jù)集: 原始數(shù)據(jù)由671,585個(gè)數(shù)字圖像和大小寫(xiě)英文字符(62個(gè)類(lèi))組成。聯(lián)邦學(xué)習(xí)的版本將數(shù)據(jù)集拆分到3,400個(gè)不平衡Clients,每個(gè)Clients上的數(shù)字/字符為同一人所寫(xiě),由于每個(gè)人都有獨(dú)特的寫(xiě)作風(fēng)格,因此數(shù)據(jù)是非同分布的。來(lái)源:Gregory Cohen, Saeed Afshar, Jonathan Tapson, and Andre ? van Schaik. EMNIST: an extension of MNIST to handwritten letters. arXiv preprint arXiv:1702.05373, 2017.2.Stackoverflow數(shù)據(jù)集: 該數(shù)據(jù)集由Stack Overflow的問(wèn)答組成,并帶有時(shí)間戳、分?jǐn)?shù)等元數(shù)據(jù)。訓(xùn)練數(shù)據(jù)集包含342,477多個(gè)用戶和135,818,730個(gè)例子。其中的時(shí)間戳信息有助于模擬傳入數(shù)據(jù)的模式。下載地址:https://www.kaggle.com/stackoverflow/stackoverflow3.Shakespeare數(shù)據(jù)集: 該數(shù)據(jù)是從The Complete Works of William Shakespeare獲得的語(yǔ)言建模數(shù)據(jù)集。由715個(gè)字符組成,其連續(xù)行是Client數(shù)據(jù)集中的示例。訓(xùn)練集樣本量為16,068,測(cè)試集為2,356。來(lái)源:Sebastian Caldas, Peter Wu, Tian Li, Jakub Konecˇny ?, H Brendan McMahan, Virginia Smith, and Ameet Talwalkar. LEAF: A benchmark for federated settings. arXiv preprint arXiv:1812.01097, 2018.(2)開(kāi)源軟件包1.TensorFlow Federated: TensorFlow框架,專(zhuān)門(mén)針對(duì)研究用例,提供大規(guī)模模擬功能來(lái)控制抽樣。支持在模擬環(huán)境中加載分散數(shù)據(jù)集,每個(gè)Client的ID對(duì)應(yīng)于TensorFlow數(shù)據(jù)集對(duì)象。來(lái)源:The TFF Authors. TensorFlow Federated, 2019. URL:https://www.tensorflow.org/federated.2.PySyft: PyTorch框架,使用PyTorch中的聯(lián)邦學(xué)習(xí)。適用于考慮隱私保護(hù)的機(jī)器學(xué)習(xí),采用差分隱私和多方計(jì)算(MPC)將私人數(shù)據(jù)與模型訓(xùn)練分離。來(lái)源:Theo Ryffel, Andrew Trask, Morten Dahl, Bobby Wagner, Jason Mancuso, Daniel Rueckert, and Jonathan Passerat-Palmbach. A generic framework for privacy preserving deep learning, 2018.
除此以外,文章也提供了其他軟件實(shí)現(xiàn)以及數(shù)據(jù)源供研究使用。
參考文獻(xiàn)
Bellet, A., Guerraoui, R., Taziki, M., and Tommasi, M. (2018), “Personalized and private peer-to-peer machine learning,” in International Conference on Artificial Intelligence and Statistics, PMLR, pp. 473–481.Colin, Igor, Aurélien Bellet, Joseph Salmon, and Stéphan Clémen?on. "Gossip dual averaging for decentralized optimization of pairwise functions." In International Conference on Machine Learning, pp. 1388-1396. PMLR, 2016.Lan, Guanghui. "An optimal method for stochastic composite optimization." Mathematical Programming 133, no. 1 (2012): 365-397.Kairouz, P., McMahan, H. B., Avent, B., Bellet, A., Bennis, M., Bhagoji, A. N., Bonawitz, K., Charles, Z., Cormode, G., Cummings, R., DOliveira, R. G. L., Eichner, H., Rouayheb, S. E., Evans, D., Gardner, J., Garrett, Z., Gascn, A., Ghazi, B., Gibbons, P. B., Gruteser, M., Harchaoui, Z., He, C., He, L., Huo, Z., Hutchinson, B., Hsu, J., Jaggi, M., Javidi, T., Joshi, G., Khodak, M., Konecn, J., Korolova, A., Koushanfar, F., Koyejo, S., Lepoint, T., Liu, Y., Mittal, P., Mohri, M., Nock, R., zgr, A., Pagh, R., Qi, H., Ramage, D., Raskar, R., Raykova, M., Song, D., Song, W., Stich, S. U., Sun, Z., Suresh, A. T., Tramr, F., Vepakomma, P., Wang, J., Xiong, L., Xu, Z., Yang, Q., Yu, F. X., Yu, H., and Zhao, S. (2021), “Advances and Open Problems in Federated Learning,” Foundations and Trends in Machine Learning, 14, 1–210.Li, Yuzheng, Chuan Chen, Nan Liu, Huawei Huang, Zibin Zheng, and Qiang Yan. "A blockchain-based decentralized federated learning framework with committee consensus," IEEE Network 35, no. 1 (2020): 234-241.McMahan, B., Moore, E., Ramage, D., Hampson, S., and y Arcas, B. A. (2017), “Communication-efficient learning of deep networks from decentralized data,” in Artificial Intelligence and Statistics, PMLR, pp. 1273–1282.Stich, Sebastian U. "Local SGD converges fast and communicates little. (2018)" arXiv preprint arXiv:1805.09767.Karimireddy, Sai Praneeth, Satyen Kale, Mehryar Mohri, Sashank Reddi, Sebastian Stich, and Ananda Theertha Suresh. (2020), "Scaffold: Stochastic controlled averaging for federated learning," In International Conference on Machine Learning, pp. 5132-5143. PMLR.Tang, H., Lian, X., Yan, M., Zhang, C., and Liu, J. (2018), “Decentralized training over decentralized data,” in International Conference on Machine Learning, PMLR, pp. 4848–4856.Vanhaesebrouck, P., Bellet, A., and Tommasi, M. (2017), “Decentralized collaborative learning of personalized models over networks,” in Artificial Intelligence and Statistics, PMLR, pp. 509–517.Xiangru Lian, Ce Zhang, Huan Zhang, Cho-Jui Hsieh, Wei Zhang, and Ji Liu. “Can Decentralized Algorithms Outperform Centralized Algorithms? A Case Study for Decentralized Parallel Stochastic Gradient Descent,” In NIPS, 2017.Xiang Li, Kaixuan Huang, Wenhao Yang, Shusen Wang, and Zhihua Zhang. “On the convergence of FedAvg on non-IID data,” arXiv preprint arXiv:1907.02189, 2019.