Matches in SemOpenAlex for { <https://semopenalex.org/work/W2808248759> ?p ?o ?g. }
- W2808248759 abstract "We study robust distributed learning that involves minimizing a non-convex loss function with saddle points. We consider the Byzantine setting where some worker machines have abnormal or even arbitrary and adversarial behavior. In this setting, the Byzantine machines may create fake local minima near a saddle point that is far away from any true local minimum, even when robust gradient estimators are used. We develop ByzantinePGD, a robust first-order algorithm that can provably escape saddle points and fake local minima, and converge to an approximate true local minimizer with low iteration complexity. As a by-product, we give a simpler algorithm and analysis for escaping saddle points in the usual non-Byzantine setting. We further discuss three robust gradient estimators that can be used in ByzantinePGD, including median, trimmed mean, and iterative filtering. We characterize their performance in concrete statistical settings, and argue for their near-optimality in low and high dimensional regimes." @default.
- W2808248759 created "2018-06-21" @default.
- W2808248759 creator A5001139821 @default.
- W2808248759 creator A5030261391 @default.
- W2808248759 creator A5030620564 @default.
- W2808248759 creator A5060774236 @default.
- W2808248759 date "2018-06-14" @default.
- W2808248759 modified "2023-09-27" @default.
- W2808248759 title "Defending Against Saddle Point Attack in Byzantine-Robust Distributed Learning" @default.
- W2808248759 cites W1505731132 @default.
- W2808248759 cites W1855339086 @default.
- W2808248759 cites W1884015253 @default.
- W2808248759 cites W1971942712 @default.
- W2808248759 cites W2009941369 @default.
- W2808248759 cites W2116612304 @default.
- W2808248759 cites W2116680824 @default.
- W2808248759 cites W2124768887 @default.
- W2808248759 cites W2131172946 @default.
- W2808248759 cites W2143238292 @default.
- W2808248759 cites W2189224591 @default.
- W2808248759 cites W2235185944 @default.
- W2808248759 cites W2283214199 @default.
- W2808248759 cites W2296832563 @default.
- W2808248759 cites W2398500126 @default.
- W2808248759 cites W2399994860 @default.
- W2808248759 cites W2480397354 @default.
- W2808248759 cites W2490498838 @default.
- W2808248759 cites W2511421375 @default.
- W2808248759 cites W2530417694 @default.
- W2808248759 cites W2549393568 @default.
- W2808248759 cites W2554864439 @default.
- W2808248759 cites W2561193401 @default.
- W2808248759 cites W2592318711 @default.
- W2808248759 cites W2592651140 @default.
- W2808248759 cites W2593047802 @default.
- W2808248759 cites W2593349530 @default.
- W2808248759 cites W2611515161 @default.
- W2808248759 cites W2751645773 @default.
- W2808248759 cites W2766900638 @default.
- W2808248759 cites W2767752488 @default.
- W2808248759 cites W2769394111 @default.
- W2808248759 cites W2770402827 @default.
- W2808248759 cites W2788308444 @default.
- W2808248759 cites W2788543015 @default.
- W2808248759 cites W2789903762 @default.
- W2808248759 cites W2791820624 @default.
- W2808248759 cites W2794080039 @default.
- W2808248759 cites W2794710858 @default.
- W2808248759 cites W2794854061 @default.
- W2808248759 cites W2798551148 @default.
- W2808248759 cites W2806196585 @default.
- W2808248759 cites W2949384269 @default.
- W2808248759 cites W2950383522 @default.
- W2808248759 cites W2950999680 @default.
- W2808248759 cites W2951773851 @default.
- W2808248759 cites W2952782294 @default.
- W2808248759 cites W2962737134 @default.
- W2808248759 cites W2963075511 @default.
- W2808248759 cites W2963306214 @default.
- W2808248759 cites W2963351358 @default.
- W2808248759 cites W2963416883 @default.
- W2808248759 cites W2963423870 @default.
- W2808248759 cites W2963446085 @default.
- W2808248759 cites W2963992805 @default.
- W2808248759 cites W2964008913 @default.
- W2808248759 cites W2964106499 @default.
- W2808248759 cites W2964135521 @default.
- W2808248759 cites W2964156132 @default.
- W2808248759 cites W2964279249 @default.
- W2808248759 cites W2995709067 @default.
- W2808248759 cites W3017335724 @default.
- W2808248759 cites W3111890340 @default.
- W2808248759 cites W3123272904 @default.
- W2808248759 cites W3137092842 @default.
- W2808248759 cites W616376197 @default.
- W2808248759 cites W643635367 @default.
- W2808248759 hasPublicationYear "2018" @default.
- W2808248759 type Work @default.
- W2808248759 sameAs 2808248759 @default.
- W2808248759 citedByCount "5" @default.
- W2808248759 countsByYear W28082487592018 @default.
- W2808248759 countsByYear W28082487592019 @default.
- W2808248759 countsByYear W28082487592021 @default.
- W2808248759 crossrefType "posted-content" @default.
- W2808248759 hasAuthorship W2808248759A5001139821 @default.
- W2808248759 hasAuthorship W2808248759A5030261391 @default.
- W2808248759 hasAuthorship W2808248759A5030620564 @default.
- W2808248759 hasAuthorship W2808248759A5060774236 @default.
- W2808248759 hasConcept C105795698 @default.
- W2808248759 hasConcept C11413529 @default.
- W2808248759 hasConcept C126255220 @default.
- W2808248759 hasConcept C134306372 @default.
- W2808248759 hasConcept C14036430 @default.
- W2808248759 hasConcept C185429906 @default.
- W2808248759 hasConcept C186633575 @default.
- W2808248759 hasConcept C2524010 @default.
- W2808248759 hasConcept C2681867 @default.
- W2808248759 hasConcept C2777127463 @default.
- W2808248759 hasConcept C33923547 @default.
- W2808248759 hasConcept C41008148 @default.