Raise exception when using "--neuronwise_optimize" parameter

[zhangzhaodi233]

Hi Zhaoyang,

I'm really into your tool - FROWN, and I'm using it to get certifier bounds on my pretrained models with sigmoid. But when I used the parameter "--neuronwise_optimize". The code abrupt and raise exception in the following line.

FROWN/get_bound_for_general_activation_function.py

Lines 82 to 84 in 144065d

	def getConvenientGeneralActivationBound(l,u, activation, use_constant=False, remain_tangent_line_info=False):
	if (l>u).sum()>0:
	raise Exception('l must be less or equal to u')

Then, I read the paper "Fastened CROWN: Tightened Neural Network Robustness Certificates" again. In this paper, there is a Condition 1 - self-consistency. In the Chapter of "Fastened CROWN", it mentioned that "Condition 1 is too strong a condition for our proposed optimization framework to be practical. Actually we propose to improve CROWN by breaking Condition 1."

So, I don't know if the raised exception is related to the failure to meet Condition 1.

Thanks and regards,
Zoey