notesum.ai

Published at November 7

Verification of Neural Networks against Convolutional Perturbations via Parameterised Kernels

cs.LG

cs.AI

cs.CR

cs.CV

Released Date: November 7, 2024

Authors: Benedikt Brückner, Alessio Lomuscio¹

Aff.: ¹Safe Intelligence

Arxiv: http://arxiv.org/abs/2411.04594v1

$s$	strength	v/us/to	time	v/us/to	time	v/us/to	time	v/us/to	time	v/us/to	time	v/us/to	time
		Box Blur		Sharpen		Motion Blur $0^{\circ}$		Motion Blur $45^{\circ}$		Motion Blur $90^{\circ}$		Motion Blur $135^{\circ}$
3	0.2	46/4/0	4781	46/4/0	4235	47/3/0	3924	46/4/0	4985	46/4/0	3739	46/4/0	4950
3	0.4	45/5/0	7156	46/4/0	6241	47/3/0	6092	45/5/0	7286	46/4/0	5729	45/5/0	7252
3	0.6	43/7/0	8519	46/4/0	7677	45/5/0	7284	44/6/0	8709	45/5/0	6895	40/10/0	7943
3	0.8	37/13/0	8864	46/4/0	8621	44/6/0	8358	41/9/0	9904	43/7/0	7732	36/14/0	8398
3	1.0	32/18/0	9211	46/4/0	9325	42/8/0	8957	40/10/0	11442	42/8/0	8374	35/15/0	10120
5	0.2	46/4/0	6415	46/4/0	5454	47/3/0	5733	46/4/0	6291	46/4/0	5297	45/5/0	6190
5	0.4	42/8/0	8594	46/4/0	7614	42/8/0	7558	45/5/0	8976	45/5/0	7727	43/7/0	8582
5	0.6	35/15/0	9963	46/4/0	8942	40/10/0	8973	38/12/0	10372	43/7/0	8958	37/13/0	9935
5	0.8	24/26/0	9641	46/4/0	10255	33/17/0	9452	29/21/0	10706	36/14/0	9271	26/24/0	8919
5	1.0	22/28/0	11857	43/7/0	10658	30/20/0	10248	26/24/0	12164	34/16/0	10852	23/27/0	10269
7	0.2	46/4/0	7295	46/4/0	6155	44/6/0	6179	46/4/0	7022	46/4/0	6026	45/5/0	6846
7	0.4	42/8/0	9919	45/5/0	8310	41/9/0	8312	43/7/0	9798	45/5/0	8526	43/7/0	9722
7	0.6	38/12/0	13899	45/5/0	9647	35/15/0	9859	36/14/0	11507	40/10/0	10043	37/13/0	11704
7	0.8	22/28/0	11823	43/7/0	10715	28/22/0	10372	27/23/0	11890	32/18/0	10698	28/22/0	11971
7	1.0	14/36/0	10464	40/10/0	11178	24/26/0	11100	21/29/0	11747	21/29/0	8213	19/31/0	10387
9	0.2	46/4/0	7914	46/4/0	6454	45/5/0	6822	46/4/0	7596	46/4/0	6371	45/5/0	7507
9	0.4	43/7/0	11115	44/6/0	8482	41/9/0	9282	43/7/0	10660	45/5/0	9120	42/8/0	10301
9	0.6	37/13/0	14980	43/7/0	10028	34/16/0	11029	35/15/0	12874	40/10/0	11312	35/15/0	12814
9	0.8	22/28/0	14201	41/9/0	11066	23/27/0	10160	27/23/0	14035	25/25/0	9322	27/23/0	13803
9	1.0	10/40/0	7911	40/10/0	12457	19/31/0	10685	17/33/0	12090	17/33/0	8698	21/29/0	13471