Matches in SemOpenAlex for { <https://semopenalex.org/work/W3033311119> ?p ?o ?g. }
- W3033311119 abstract "The high energy costs of neural network training and inference led to the use of acceleration hardware such as GPUs and TPUs. While this enabled us to train large-scale neural networks in datacenters and deploy them on edge devices, the focus so far is on average-case performance. In this work, we introduce a novel threat vector against neural networks whose energy consumption or decision latency are critical. We show how adversaries can exploit carefully crafted $boldsymbol{sponge}~boldsymbol{examples}$, which are inputs designed to maximise energy consumption and latency. We mount two variants of this attack on established vision and language models, increasing energy consumption by a factor of 10 to 200. Our attacks can also be used to delay decisions where a network has critical real-time performance, such as in perception for autonomous vehicles. We demonstrate the portability of our malicious inputs across CPUs and a variety of hardware accelerator chips including GPUs, and an ASIC simulator. We conclude by proposing a defense strategy which mitigates our attack by shifting the analysis of energy consumption in hardware from an average-case to a worst-case perspective." @default.
- W3033311119 created "2020-06-12" @default.
- W3033311119 creator A5018809423 @default.
- W3033311119 creator A5040762162 @default.
- W3033311119 creator A5046649739 @default.
- W3033311119 creator A5046983053 @default.
- W3033311119 creator A5048483915 @default.
- W3033311119 creator A5069844959 @default.
- W3033311119 date "2020-06-05" @default.
- W3033311119 modified "2023-09-24" @default.
- W3033311119 title "Sponge Examples: Energy-Latency Attacks on Neural Networks" @default.
- W3033311119 cites W1526747343 @default.
- W3033311119 cites W1867219652 @default.
- W3033311119 cites W1963651058 @default.
- W3033311119 cites W1999085092 @default.
- W3033311119 cites W2000359198 @default.
- W3033311119 cites W2012465543 @default.
- W3033311119 cites W2018108780 @default.
- W3033311119 cites W2018975076 @default.
- W3033311119 cites W2051152775 @default.
- W3033311119 cites W2090394384 @default.
- W3033311119 cites W2140141884 @default.
- W3033311119 cites W2153579005 @default.
- W3033311119 cites W2162552722 @default.
- W3033311119 cites W2194775991 @default.
- W3033311119 cites W2200264815 @default.
- W3033311119 cites W2285660444 @default.
- W3033311119 cites W2319799779 @default.
- W3033311119 cites W2511844051 @default.
- W3033311119 cites W2535690855 @default.
- W3033311119 cites W2553427643 @default.
- W3033311119 cites W2603766943 @default.
- W3033311119 cites W2606722458 @default.
- W3033311119 cites W2613119772 @default.
- W3033311119 cites W2625457103 @default.
- W3033311119 cites W2657095061 @default.
- W3033311119 cites W2750990141 @default.
- W3033311119 cites W2773446523 @default.
- W3033311119 cites W2793320545 @default.
- W3033311119 cites W2794670651 @default.
- W3033311119 cites W2798956872 @default.
- W3033311119 cites W2801748224 @default.
- W3033311119 cites W2887659847 @default.
- W3033311119 cites W2889326796 @default.
- W3033311119 cites W2896006880 @default.
- W3033311119 cites W2898963688 @default.
- W3033311119 cites W2933138175 @default.
- W3033311119 cites W2941208651 @default.
- W3033311119 cites W2945146780 @default.
- W3033311119 cites W2958749653 @default.
- W3033311119 cites W2962763344 @default.
- W3033311119 cites W2962784628 @default.
- W3033311119 cites W2963163009 @default.
- W3033311119 cites W2963207607 @default.
- W3033311119 cites W2963310665 @default.
- W3033311119 cites W2963311060 @default.
- W3033311119 cites W2963341956 @default.
- W3033311119 cites W2963367920 @default.
- W3033311119 cites W2963378725 @default.
- W3033311119 cites W2963403868 @default.
- W3033311119 cites W2963446712 @default.
- W3033311119 cites W2963807318 @default.
- W3033311119 cites W2963809228 @default.
- W3033311119 cites W2964153729 @default.
- W3033311119 cites W2964923388 @default.
- W3033311119 cites W2965373594 @default.
- W3033311119 cites W2969580338 @default.
- W3033311119 cites W2970295111 @default.
- W3033311119 cites W2970604398 @default.
- W3033311119 cites W2981540061 @default.
- W3033311119 cites W2990704537 @default.
- W3033311119 cites W3003453708 @default.
- W3033311119 cites W3005957694 @default.
- W3033311119 cites W3008905965 @default.
- W3033311119 cites W3046102592 @default.
- W3033311119 cites W3091857398 @default.
- W3033311119 cites W3096618103 @default.
- W3033311119 cites W3101543398 @default.
- W3033311119 cites W9657784 @default.
- W3033311119 cites W240297117 @default.
- W3033311119 hasPublicationYear "2020" @default.
- W3033311119 type Work @default.
- W3033311119 sameAs 3033311119 @default.
- W3033311119 citedByCount "13" @default.
- W3033311119 countsByYear W30333111192019 @default.
- W3033311119 countsByYear W30333111192020 @default.
- W3033311119 countsByYear W30333111192021 @default.
- W3033311119 crossrefType "posted-content" @default.
- W3033311119 hasAuthorship W3033311119A5018809423 @default.
- W3033311119 hasAuthorship W3033311119A5040762162 @default.
- W3033311119 hasAuthorship W3033311119A5046649739 @default.
- W3033311119 hasAuthorship W3033311119A5046983053 @default.
- W3033311119 hasAuthorship W3033311119A5048483915 @default.
- W3033311119 hasAuthorship W3033311119A5069844959 @default.
- W3033311119 hasConcept C105795698 @default.
- W3033311119 hasConcept C111919701 @default.
- W3033311119 hasConcept C119599485 @default.
- W3033311119 hasConcept C127413603 @default.
- W3033311119 hasConcept C149635348 @default.
- W3033311119 hasConcept C154945302 @default.