id	author	title	date	pages	extension	mime	words	sentences	flesch	summary	cache	txt
cord-029277-mjpwkm2u	Elboher, Yizhak Yisrael	An Abstraction-Based Framework for Neural Network Verification	2020-06-13		.txt	text/plain	8796	523	63	Different verification approaches may differ in (i) the kinds of neural networks they allow (specifically, the kinds of activation functions in use); (ii) the kinds of input properties; and (iii) the kinds of output properties. Because the complexity of verifying a neural network is strongly connected to its size [20] , our goal is to transform a verification query ϕ 1 = N, P, Q into query ϕ 2 = N , P, Q , such that the abstract networkN is significantly smaller than N (notice that properties P and Q remain unchanged). Together with a black-box verification procedure Verify that can dispatch queries of the form ϕ = N, P, Q , these components now allow us to design an abstraction-refinement algorithm for DNN verification, given as Algorithm 1 (we assume that all hidden neurons in the input network have already been marked pos/neg and inc/dec).	./cache/cord-029277-mjpwkm2u.txt	./txt/cord-029277-mjpwkm2u.txt