講座名稱：Privacy-preserving Federated Clustering and Classification by CVX Optimization (CVXopt) or AI-aided CVXopt

講座人：祁忠勇 教授

講座時(shí)間：4月22日15:30-17：30

地點(diǎn)：新科技樓1012報(bào)告廳

講座人介紹：

祁忠勇，1983年于美國(guó)南加州大學(xué)University of Southern California取得電氣工程專業(yè)博士學(xué)位，1983~1988年任職于美國(guó)噴氣推進(jìn)實(shí)驗(yàn)室（Jet Propulsion Laboratory, JPL），1989年至今在臺(tái)灣清華大學(xué)電機(jī)工程系擔(dān)任正教授。已發(fā)表學(xué)術(shù)論文170余篇（其中期刊論文60余篇，大部分為IEEE Transactions Signal Processing長(zhǎng)文）、專著2本、會(huì)議論文100余篇。研究方向廣泛，主要包括無(wú)線通訊信號(hào)處理、凸函數(shù)分析及優(yōu)化、盲信號(hào)分離、醫(yī)學(xué)及高光譜影像分析等。祁教授是信號(hào)處理領(lǐng)域國(guó)際知名學(xué)者，為IEEE Senior Member，曾擔(dān)任IEEE SPAWC、SPC等多個(gè)學(xué)術(shù)會(huì)議主席，為IEEE Transactions on Signal Processing、IEEE Transactions on Circuits and Systems I、IEEE Transactions on Circuits and Systems II、IEEE Signal Processing Letters等SCI學(xué)術(shù)期刊副主編、擔(dān)任EURASIP Signal Processing編委會(huì)成員等。

講座內(nèi)容：

Abstract: Federated learning (FL) has been a rapidly growing research area together with artificial intelligence (AI), where the model is trained over massively distributed clients under the orchestration of a parameter server (PS) without sharing clients’ data. In this presentation, by means of the widely known differential privacy (DP) theory for privacy preservation, we present a supervised classification algorithm by AI-aided convex optimization (CVXopt) and an unsupervised clustering algorithm by CVXopt, each developed by solving a non-convex and non-smooth (NCNM) FL problem. Their unique insightful properties and some privacy and convergence analyses are also presented, that can be used for the FL algorithm design guidelines. Extensive experiments on real-world data are presented to demonstrate the effectiveness of the presented algorithms and much superior performance over state-of-the-art FL algorithms, together with the validation of all the analytical results and properties. Finally, we draw some conclusions as well as some future research explorations.

主辦單位：通信工程學(xué)院