Robust Large Margin Deep Neural Networks — mentioned by derivt

Sat, 24 Feb 2018 11:45:45 GMT

This kind of attack relies on a low margin DNN, see (1), a low spectral norm of the input-output jacobian matrix guarantees good generalization error. So a one pixel attack exploits a weak eigenvalue (small absolute value) of the jacobian matrix. So to create a one pixel attack, compute: 1)the eigenvalues of the jacobian of input-ouput matrix, 2) takes the the smaller eigenvalue lambda_1 3) compute or approximate the function lambda_1 = f(input) 4) compute j = argmax_{i=1..n} d(lambda_1)/d(input_i) at the point in which the spectral norm is maximum. So to create the attack change the j-pixel…

Paper and discussion history. A mention is not an endorsement.

Robust Large Margin Deep Neural Networks · HackerPapers discussions

Robust Large Margin Deep Neural Networks — mentioned by derivt