ddiruscio/logseq
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
master
logseq/assets/SANER2024_paper_194_1700084424094_0.edn
T
ddiruscio 97fba407ce Major2
2025-06-05 22:07:12 +02:00

1635 lines
89 KiB
Clojure
Raw Permalink Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
{:highlights [{:id #uuid "6575f969-93b9-4193-b1e5-fc2d99dd4927",
:page 1,
:position {:bounding {:x1 77.6796875,
:y1 251.5703125,
:x2 476.1384735107422,
:y2 283.40625,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 157.2452850341797,
:y1 251.5703125,
:x2 476.1384735107422,
:y2 267.5703125,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 267.40625,
:x2 472.19305419921875,
:y2 283.40625,
:width 971,
:height 1256.5882352941176}),
:page 1},
:content {:text "Most existing works reveal that the deep learning system is extremely susceptible to adversarial examples (AEs)"},
:properties {:color "green"}}
{:id #uuid "6575f97a-99a2-4368-8506-d1760dd988c5",
:page 1,
:position {:bounding {:x1 77.6796875,
:y1 314.7734375,
:x2 475.642333984375,
:y2 346.609375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 319.63629150390625,
:y1 314.7734375,
:x2 475.642333984375,
:y2 330.7734375,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 330.609375,
:x2 284.881591796875,
:y2 346.609375,
:width 971,
:height 1256.5882352941176}),
:page 1},
:content {:text "optimized gradient-based technologies in white-box testing"},
:properties {:color "purple"}}
{:id #uuid "6575f9a5-9503-43bc-8289-588ccbb928cd",
:page 1,
:position {:bounding {:x1 77.6796875,
:y1 393.8125,
:x2 476.1583251953125,
:y2 441.359375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 77.6796875,
:y1 393.8125,
:x2 476.1516418457031,
:y2 409.8125,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 409.6484375,
:x2 475.9927673339844,
:y2 425.6484375,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 425.359375,
:x2 476.1583251953125,
:y2 441.359375,
:width 971,
:height 1256.5882352941176}),
:page 1},
:content {:text "formal analysis between gradient-based attack and loss minima of the loss function to prove that powerful adversaries will share similar feature representations with a high probability."},
:properties {:color "purple"}}
{:id #uuid "6575f9cb-a047-4cd7-9696-78deac13e7f2",
:page 1,
:position {:bounding {:x1 77.6796875,
:y1 536.0625,
:x2 475.5866394042969,
:y2 583.7265625,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 281.1526184082031,
:y1 536.0625,
:x2 475.37109375,
:y2 552.0625,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 551.890625,
:x2 475.5866394042969,
:y2 567.890625,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 567.7265625,
:x2 384.31072998046875,
:y2 583.7265625,
:width 971,
:height 1256.5882352941176}),
:page 1},
:content {:text " our results prove that AEs can efficiently discover the vulnerability of DL model but are not suitable to explore more inner logic as test suites"},
:properties {:color "purple"}}
{:id #uuid "6575f9e7-9dfd-46bc-b262-ef059b0946ec",
:page 1,
:position {:bounding {:x1 77.6796875,
:y1 685.34375,
:x2 475.8890075683594,
:y2 741.296875,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 427.4505920410156,
:y1 685.34375,
:x2 475.8890075683594,
:y2 703.34375,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 704.3203125,
:x2 475.2975769042969,
:y2 722.3203125,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 723.296875,
:x2 471.8170471191406,
:y2 741.296875,
:width 971,
:height 1256.5882352941176}),
:page 1},
:content {:text "opacity of DL decisions renders it difficult to understand the trustworthiness of the model in response to open-world scenarios"},
:properties {:color "green"}}
{:id #uuid "6575f9fb-2f6b-414e-a68c-3f04f2841906",
:page 1,
:position {:bounding {:x1 77.6796875,
:y1 799.0703125,
:x2 475.88665771484375,
:y2 873.9921875,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 408.7023010253906,
:y1 799.0703125,
:x2 475.3397521972656,
:y2 817.0703125,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 818.0390625,
:x2 475.88336181640625,
:y2 836.0390625,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 837.015625,
:x2 475.88665771484375,
:y2 855.015625,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 855.9921875,
:x2 146.4271240234375,
:y2 873.9921875,
:width 971,
:height 1256.5882352941176}),
:page 1},
:content {:text " DL model should have a more rigorous test process and more predictable test outcomes, in order to prove the robustness and security of the model"},
:properties {:color "green"}}
{:id #uuid "6575fa27-6d2d-44a0-b41d-4557c3701eff",
:page 1,
:position {:bounding {:x1 77.6796875,
:y1 855.9921875,
:x2 475.60516357421875,
:y2 911.9375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 331.763427734375,
:y1 855.9921875,
:x2 475.60516357421875,
:y2 873.9921875,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 874.9609375,
:x2 474.7650451660156,
:y2 892.9609375,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 893.9375,
:x2 135.30136108398438,
:y2 911.9375,
:width 971,
:height 1256.5882352941176}),
:page 1},
:content {:text " “ISO/IEC TR 2911911 Software Testing - Guidelines on the testing of AI-based systems"},
:properties {:color "purple"}}
{:id #uuid "6575fa6a-4509-497d-984e-40d247cc0835",
:page 1,
:position {:bounding {:x1 77.6796875,
:y1 893.9375,
:x2 475.8837890625,
:y2 968.859375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 144.88815307617188,
:y1 893.9375,
:x2 475.3057556152344,
:y2 911.9375,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 912.9140625,
:x2 475.8823547363281,
:y2 930.9140625,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 931.890625,
:x2 475.8837890625,
:y2 949.890625,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 950.859375,
:x2 407.12579345703125,
:y2 968.859375,
:width 971,
:height 1256.5882352941176}),
:page 1},
:content {:text " DL testing should basically satisfy the following requirements: (1) high defects revealing-ability, (2) test suites should be more diverse (i.e., triggering different logic), and(3) test adequacy assurance (i.e., coverage criteria)"},
:properties {:color "purple"}}
{:id #uuid "6575facc-59b7-43b4-8e16-96651ad23617",
:page 1,
:position {:bounding {:x1 77.6796875,
:y1 989.4375,
:x2 475.8832702636719,
:y2 1045.390625,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 112.25824737548828,
:y1 989.4375,
:x2 475.2897415161133,
:y2 1007.4375,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 1008.4140625,
:x2 475.8832702636719,
:y2 1026.4140625,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 1027.390625,
:x2 224.05213928222656,
:y2 1045.390625,
:width 971,
:height 1256.5882352941176}),
:page 1},
:content {:text "the decision logic of DL model follows a data-driven programming paradigm which is determined by training data rather than code logic"},
:properties {:color "purple"}}
{:id #uuid "6575fb8c-c5b4-430f-ade9-bc145a087250",
:page 1,
:position {:bounding {:x1 495.015625,
:y1 249.6875,
:x2 893.2655639648438,
:y2 343.5859375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 590.9108276367188,
:y1 249.6875,
:x2 893.2230834960938,
:y2 267.6875,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 268.6640625,
:x2 893.2655639648438,
:y2 286.6640625,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 287.6328125,
:x2 892.632080078125,
:y2 305.6328125,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 306.609375,
:x2 892.3499145507812,
:y2 324.609375,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 325.5859375,
:x2 781.7601318359375,
:y2 343.5859375,
:width 971,
:height 1256.5882352941176}),
:page 1},
:content {:text "Empirical studies [3] suggest that preventing adversaries from computing effective AEs is unattainable until now, which gives adequate confidence to DL testers that AEs can achieve a high fault-revealing rate whatever the robustness degree of given DL model"},
:properties {:color "purple"}}
{:id #uuid "6575fc1a-1aae-4a27-9e4f-9ffd4b0db619",
:page 1,
:position {:bounding {:x1 495.015625,
:y1 382.5078125,
:x2 893.2225952148438,
:y2 457.4296875,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 659.364990234375,
:y1 382.5078125,
:x2 893.2225952148438,
:y2 400.5078125,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 401.484375,
:x2 893.221923828125,
:y2 419.484375,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 420.4609375,
:x2 893.2220458984375,
:y2 438.4609375,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 439.4296875,
:x2 760.1068115234375,
:y2 457.4296875,
:width 971,
:height 1256.5882352941176}),
:page 1},
:content {:text "Following this line, the optimized adversary will perturb the original example along with several specific feature representations to minimize the loss between the correct class and the targeted class."},
:properties {:color "green"}}
{:id #uuid "6575fc29-2ab6-4ff1-9214-c7eeaff40e66",
:page 1,
:position {:bounding {:x1 495.015625,
:y1 439.4296875,
:x2 893.2247924804688,
:y2 533.328125,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 764.2669677734375,
:y1 439.4296875,
:x2 893.2234497070312,
:y2 457.4296875,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 458.40625,
:x2 893.2247924804688,
:y2 476.40625,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 477.3828125,
:x2 892.635009765625,
:y2 495.3828125,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 496.3515625,
:x2 893.2177734375,
:y2 514.3515625,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 515.328125,
:x2 813.482666015625,
:y2 533.328125,
:width 971,
:height 1256.5882352941176}),
:page 1},
:content {:text "Generally, although adversarial examples can efficiently detect the vulnerability of DL model, they are not suitable to be test suites from the perspective of test adequacy. It remains an open question that what is the best test suites generation technology"},
:properties {:color "purple"}}
{:id #uuid "6575fc58-fc9a-47ea-9322-eabb9416f7a5",
:page 1,
:position {:bounding {:x1 495.015625,
:y1 667,
:x2 893.22021484375,
:y2 741.921875,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 679.3114013671875,
:y1 667,
:x2 893.22021484375,
:y2 685,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 685.9765625,
:x2 893.2186279296875,
:y2 703.9765625,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 704.9453125,
:x2 892.6763916015625,
:y2 722.9453125,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 723.921875,
:x2 547.7325439453125,
:y2 741.921875,
:width 971,
:height 1256.5882352941176}),
:page 1},
:content {:text "theoretical understanding of adversarial attacks from the perspective of connections between gradients’ geometrical properties and local minima of the loss function"},
:properties {:color "green"}}
{:id #uuid "6575fc5e-aa50-4a91-aeb7-f7a4d124500e",
:page 1,
:position {:bounding {:x1 495.015625,
:y1 723.921875,
:x2 892.3944091796875,
:y2 760.8984375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 719.21142578125,
:y1 723.921875,
:x2 892.3944091796875,
:y2 741.921875,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 742.8984375,
:x2 709.8743286132812,
:y2 760.8984375,
:width 971,
:height 1256.5882352941176}),
:page 1},
:content {:text " optimized AEs tend to fall into several limited local minima."},
:properties {:color "green"}}
{:id #uuid "6575fc73-d773-4f24-b1d4-7cebde97d1db",
:page 1,
:position {:bounding {:x1 495.015625,
:y1 761.875,
:x2 893.22412109375,
:y2 817.8203125,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 562.5776977539062,
:y1 761.875,
:x2 893.22412109375,
:y2 779.875,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 780.84375,
:x2 893.2237548828125,
:y2 798.84375,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 799.8203125,
:x2 773.4390258789062,
:y2 817.8203125,
:width 971,
:height 1256.5882352941176}),
:page 1},
:content {:text "we then combine feature visualization technologies[8] and optimized gradient-based attack algorithms to reveal AEs share similar feature representations"},
:properties {:color "green"}}
{:id #uuid "6575fd79-3da1-4a44-a4b0-34c40b5bf882",
:page 1,
:position {:bounding {:x1 495.015625,
:y1 932.640625,
:x2 893.21875,
:y2 969.6171875,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 581.73291015625,
:y1 932.640625,
:x2 893.21875,
:y2 950.640625,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 951.6171875,
:x2 852.814697265625,
:y2 969.6171875,
:width 971,
:height 1256.5882352941176}),
:page 1},
:content {:text "multi-objective search techniques can produce more diverse test suites and cover more decision logic."},
:properties {:color "purple"}}
{:id #uuid "6575fdcf-a11b-424d-8d93-2affba291eb3",
:page 1,
:position {:bounding {:x1 495.015625,
:y1 1027.390625,
:x2 892.2930908203125,
:y2 1064.359375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 603.3133544921875,
:y1 1027.390625,
:x2 892.2930908203125,
:y2 1045.390625,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 1046.359375,
:x2 720.128662109375,
:y2 1064.359375,
:width 971,
:height 1256.5882352941176}),
:page 1},
:content {:text "theoretical analysis that the optimized AEs have similar feature representation"},
:properties {:color "green"}}
{:id #uuid "6575fdde-9130-4657-9b00-c2e1f9e5f0c6",
:page 1,
:position {:bounding {:x1 495.015625,
:y1 1103.2890625,
:x2 892.2769165039062,
:y2 1140.2578125,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 697.96728515625,
:y1 1103.2890625,
:x2 892.2769165039062,
:y2 1121.2890625,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 1122.2578125,
:x2 666.37158203125,
:y2 1140.2578125,
:width 971,
:height 1256.5882352941176}),
:page 1},
:content {:text "easons of AEs can increase common coverage criteria"},
:properties {:color "green"}}
{:id #uuid "6575fdfc-9b5b-46e3-8fb7-8d972a9b4ef5",
:page 2,
:position {:bounding {:x1 77.6796875,
:y1 499.875,
:x2 475.88275146484375,
:y2 536.7265625,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 195.52154541015625,
:y1 499.875,
:x2 475.88275146484375,
:y2 517.875,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 518.7265625,
:x2 290.7762451171875,
:y2 536.7265625,
:width 971,
:height 1256.5882352941176}),
:page 2},
:content {:text "earch-based technologies have promising performance compared with AEs"},
:properties {:color "purple"}}
{:id #uuid "6575fe32-510b-43d2-97b2-3d4c343b36f8",
:page 2,
:position {:bounding {:x1 137.61050415039062,
:y1 684.59375,
:x2 150.74913024902344,
:y2 702.59375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 137.61050415039062,
:y1 684.59375,
:x2 150.74913024902344,
:y2 702.59375,
:width 971,
:height 1256.5882352941176}),
:page 2},
:content {:text nil},
:properties {:color "red"}}
{:id #uuid "6575fe4a-23d3-4ef9-a0e0-b2fb8e820abb",
:page 2,
:position {:bounding {:x1 344.796875,
:y1 684.59375,
:x2 412.890625,
:y2 702.59375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 344.796875,
:y1 684.59375,
:x2 412.890625,
:y2 702.59375,
:width 971,
:height 1256.5882352941176}),
:page 2},
:content {:text "F(Θ, x, y)"},
:properties {:color "yellow"}}
{:id #uuid "6575fe88-160b-4af2-8180-8a09adc28b65",
:page 2,
:position {:bounding {:x1 77.6796875,
:y1 813.140625,
:x2 475.9012451171875,
:y2 850.1171875,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 430.0771484375,
:y1 813.140625,
:x2 475.9012451171875,
:y2 831.140625,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 832.1171875,
:x2 335.12127685546875,
:y2 850.1171875,
:width 971,
:height 1256.5882352941176}),
:page 2},
:content {:text " weight W and gradient G of the trained model."},
:properties {:color "green"}}
{:id #uuid "6575fe97-bcb4-4ec5-915f-24760ec740fb",
:page 2,
:position {:bounding {:x1 134.86378479003906,
:y1 908.015625,
:x2 142.43447875976562,
:y2 926.015625,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 134.86378479003906,
:y1 908.015625,
:x2 142.43447875976562,
:y2 926.015625,
:width 971,
:height 1256.5882352941176}),
:page 2},
:content {:text "s"},
:properties {:color "red"}}
{:id #uuid "6575feb6-1a83-4f34-a46f-40f6b918d35e",
:page 2,
:position {:bounding {:x1 269.60955810546875,
:y1 1103.2890625,
:x2 278.21484375,
:y2 1121.2890625,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 269.60955810546875,
:y1 1103.2890625,
:x2 278.21484375,
:y2 1121.2890625,
:width 971,
:height 1256.5882352941176}),
:page 2},
:content {:text "d"},
:properties {:color "red"}}
{:id #uuid "6575ff2d-8286-4737-b715-cc30c39a854c",
:page 2,
:position {:bounding {:x1 557.4609375,
:y1 210.859375,
:x2 611.8125,
:y2 228.859375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 557.4609375,
:y1 210.859375,
:x2 611.8125,
:y2 228.859375,
:width 971,
:height 1256.5882352941176}),
:page 2},
:content {:text "||δ|| ≤ ϵ"},
:properties {:color "green"}}
{:id #uuid "6575ff46-9ff0-48a3-9507-d3ceb182b021",
:page 2,
:position {:bounding {:x1 97.8634033203125,
:y1 870.0625,
:x2 324.9140625,
:y2 889.34375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 97.8634033203125,
:y1 870.0625,
:x2 245.8984375,
:y2 888.0625,
:width 971,
:height 1256.5882352941176}
{:x1 258.0859375,
:y1 870.0625,
:x2 324.9140625,
:y2 888.0625,
:width 971,
:height 1256.5882352941176}
{:x1 247.6015625,
:y1 876.84375,
:x2 256.203125,
:y2 889.34375,
:width 971,
:height 1256.5882352941176}),
:page 2},
:content {:text " classification loss L(FΘ(x + δ), y)"},
:properties {:color "yellow"}}
{:id #uuid "6575ff5d-8dfe-488f-ac51-3ed607dbb67b",
:page 2,
:position {:bounding {:x1 495.015625,
:y1 450.6171875,
:x2 893.2210693359375,
:y2 487.4609375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 807.223388671875,
:y1 450.6171875,
:x2 893.2210693359375,
:y2 468.6171875,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 469.4609375,
:x2 591.7008056640625,
:y2 487.4609375,
:width 971,
:height 1256.5882352941176}),
:page 2},
:content {:text "first-order adversarial attack"},
:properties {:color "green"}}
{:id #uuid "65761905-fe5a-483f-8e0a-e20d76730180",
:page 3,
:position {:bounding {:x1 77.6796875,
:y1 132.828125,
:x2 475.8846435546875,
:y2 169.796875,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 456.9210510253906,
:y1 132.828125,
:x2 475.8846435546875,
:y2 150.828125,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 151.796875,
:x2 355.171142578125,
:y2 169.796875,
:width 971,
:height 1256.5882352941176}),
:page 3},
:content {:text "requires rigorous testing before deployment"},
:properties {:color "green"}}
{:id #uuid "6576191a-90b8-48a8-9718-961af7dd697f",
:page 3,
:position {:bounding {:x1 77.6796875,
:y1 189.75,
:x2 475.88812255859375,
:y2 302.6171875,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 444.50238037109375,
:y1 189.75,
:x2 475.62646484375,
:y2 207.75,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 208.7265625,
:x2 475.88812255859375,
:y2 226.7265625,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 227.6953125,
:x2 475.8876647949219,
:y2 245.6953125,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 246.671875,
:x2 475.8843994140625,
:y2 264.671875,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 265.6484375,
:x2 475.88653564453125,
:y2 283.6484375,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 284.6171875,
:x2 471.8479919433594,
:y2 302.6171875,
:width 971,
:height 1256.5882352941176}),
:page 3},
:content {:text "Testing is checking whether the actual outputs match the expected ones and ensure that the system is defect free from the perspective of software engineering, which comprises following key test procedures (i) requirement analysis, (ii) test suites generation, (iii) test oracle generation, and (iv) test evaluation"},
:properties {:color "green"}}
{:id #uuid "6576193b-cd14-43d8-8c3a-7ba7ec20cf48",
:page 3,
:position {:bounding {:x1 77.6796875,
:y1 360.390625,
:x2 475.8887023925781,
:y2 454.2890625,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 439.69287109375,
:y1 360.390625,
:x2 475.8887023925781,
:y2 378.390625,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 379.3671875,
:x2 475.8853454589844,
:y2 397.3671875,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 398.34375,
:x2 475.88751220703125,
:y2 416.34375,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 417.3125,
:x2 475.8839416503906,
:y2 435.3125,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 436.2890625,
:x2 184.604736328125,
:y2 454.2890625,
:width 971,
:height 1256.5882352941176}),
:page 3},
:content {:text "More specifically, there are several optimized gradient-based adversarial attack techniques have successfully proved that neural network is extremely susceptible to adversarial examples, such as PGD, C&W. "},
:properties {:color "red"}}
{:id #uuid "65761943-4b65-420d-aa7f-95498a835ca3",
:page 3,
:position {:bounding {:x1 77.6796875,
:y1 436.2890625,
:x2 475.88470458984375,
:y2 492.2421875,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 304.58599853515625,
:y1 436.2890625,
:x2 475.02099609375,
:y2 454.2890625,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 455.265625,
:x2 475.88470458984375,
:y2 473.265625,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 474.2421875,
:x2 312.9601135253906,
:y2 492.2421875,
:width 971,
:height 1256.5882352941176}),
:page 3},
:content {:text "verage-based test suites generation approaches which intend to explore more internal decision logic of the neural network"},
:properties {:color "green"}}
{:id #uuid "65761969-09aa-489f-9761-4aadfa2e449a",
:page 3,
:position {:bounding {:x1 697.3612060546875,
:y1 320.0546875,
:x2 811.2003173828125,
:y2 338.0546875,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 697.3612060546875,
:y1 320.0546875,
:x2 811.2003173828125,
:y2 338.0546875,
:width 971,
:height 1256.5882352941176}),
:page 3},
:content {:text " [?], [?], [?], [?], "},
:properties {:color "red"}}
{:id #uuid "6576198c-cbb9-439a-8fd4-7f00d612118f",
:page 3,
:position {:bounding {:x1 495.015625,
:y1 417.6953125,
:x2 893.218505859375,
:y2 492.6171875,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 814.8228759765625,
:y1 417.6953125,
:x2 893.2055053710938,
:y2 435.6953125,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 436.6640625,
:x2 892.338623046875,
:y2 454.6640625,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 455.640625,
:x2 893.218505859375,
:y2 473.640625,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 474.6171875,
:x2 888.946533203125,
:y2 492.6171875,
:width 971,
:height 1256.5882352941176}),
:page 3},
:content {:text "Goodfellow et al. [4] argue that the main reason for AEs comes from the linear part of the model, in which the perturbation can significantly magnify as the dimension of weight increases"},
:properties {:color "green"}}
{:id #uuid "657619a1-eaba-4ccd-a331-4ce22784424d",
:page 3,
:position {:bounding {:x1 495.015625,
:y1 512.5625,
:x2 892.341796875,
:y2 549.5390625,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 636.0252075195312,
:y1 512.5625,
:x2 892.341796875,
:y2 530.5625,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 531.5390625,
:x2 853.4991455078125,
:y2 549.5390625,
:width 971,
:height 1256.5882352941176}),
:page 3},
:content {:text " AEs arise when data lies on a lower dimensional manifold in a high dimensional space. "},
:properties {:color "green"}}
{:id #uuid "657619ba-7612-483c-9911-90b546a309c8",
:page 3,
:position {:bounding {:x1 495.015625,
:y1 626.4140625,
:x2 893.2203369140625,
:y2 682.234375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 542.9551391601562,
:y1 626.4140625,
:x2 893.2203369140625,
:y2 644.4140625,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 645.2578125,
:x2 893.2178955078125,
:y2 663.2578125,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 664.234375,
:x2 639.24169921875,
:y2 682.234375,
:width 971,
:height 1256.5882352941176}),
:page 3},
:content {:text "some papers explore to explanations in the data, such as insufficient training data, and the extremely exceptional samples in the data se"},
:properties {:color "green"}}
{:id #uuid "657619c7-42fe-4283-80ae-df08a1d051b3",
:page 3,
:position {:bounding {:x1 549.6400146484375,
:y1 683.2109375,
:x2 837.0006103515625,
:y2 701.2109375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 549.6400146484375,
:y1 683.2109375,
:x2 837.0006103515625,
:y2 701.2109375,
:width 971,
:height 1256.5882352941176}),
:page 3},
:content {:text "robust features (useful and semantic-related) "},
:properties {:color "green"}}
{:id #uuid "657619d9-8af7-4991-aba7-b3661bafc786",
:page 3,
:position {:bounding {:x1 495.015625,
:y1 664.234375,
:x2 893.2236328125,
:y2 758.1328125,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 647.24560546875,
:y1 664.234375,
:x2 893.2196044921875,
:y2 682.234375,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 683.2109375,
:x2 893.2235107421875,
:y2 701.2109375,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 702.1875,
:x2 893.2236328125,
:y2 720.1875,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 721.15625,
:x2 893.2198486328125,
:y2 739.15625,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 740.1328125,
:x2 551.1311645507812,
:y2 758.1328125,
:width 971,
:height 1256.5882352941176}),
:page 3},
:content {:text " Further studies point out that the existence of robust features (useful and semantic-related) and nonrobust features (highly useful yet perturbation-sensitive) in any figures, while non-robust features directly cause the presence of AEs. "},
:properties {:color "red"}}
{:id #uuid "65761a10-90c0-4c36-9d90-824318e5abe2",
:page 2,
:position {:bounding {:x1 268.06201171875,
:y1 741.515625,
:x2 348.9328918457031,
:y2 759.515625,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 268.06201171875,
:y1 741.515625,
:x2 348.9328918457031,
:y2 759.515625,
:width 971,
:height 1256.5882352941176}),
:page 2},
:content {:text "training loss"},
:properties {:color "green"}}
{:id #uuid "65761a80-5337-4826-a03f-d2d77606de10",
:page 4,
:position {:bounding {:x1 77.6796875,
:y1 403.4921875,
:x2 475.9081115722656,
:y2 440.46875,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 93.5,
:y1 403.4921875,
:x2 475.9081115722656,
:y2 421.4921875,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 422.46875,
:x2 210.4762725830078,
:y2 440.46875,
:width 971,
:height 1256.5882352941176}),
:page 4},
:content {:text "RQ1. What is the difference between the adversarial attack and DL test?"},
:properties {:color "green"}}
{:id #uuid "65761a8f-7cdc-4897-bafb-69f2b2c133d7",
:page 4,
:position {:bounding {:x1 77.6796875,
:y1 479.265625,
:x2 475.0063781738281,
:y2 516.2421875,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 393.40960693359375,
:y1 479.265625,
:x2 475.0063781738281,
:y2 497.265625,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 498.2421875,
:x2 218.57528686523438,
:y2 516.2421875,
:width 971,
:height 1256.5882352941176}),
:page 4},
:content {:text "AEs can be regarded as test suite"},
:properties {:color "green"}}
{:id #uuid "65761ab0-87d6-41a4-a44f-0dffbbc27076",
:page 4,
:position {:bounding {:x1 77.6796875,
:y1 593.234375,
:x2 475.60162353515625,
:y2 630.2109375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 118.2445068359375,
:y1 593.234375,
:x2 475.60162353515625,
:y2 611.234375,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 612.2109375,
:x2 195.3828125,
:y2 630.2109375,
:width 971,
:height 1256.5882352941176}),
:page 4},
:content {:text "The relationship between adversarial examples and coverage criteria "},
:properties {:color "yellow"}}
{:id #uuid "65761af8-a21e-4eba-a0e8-fd173f85cc15",
:page 4,
:position {:bounding {:x1 137.00527954101562,
:y1 933.8984375,
:x2 211.0941162109375,
:y2 951.8984375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 137.00527954101562,
:y1 933.8984375,
:x2 211.0941162109375,
:y2 951.8984375,
:width 971,
:height 1256.5882352941176}),
:page 4},
:content {:text "Theorem 1"},
:properties {:color "yellow"}}
{:id #uuid "65761b1b-66fd-44e7-95cc-8b4342c03b98",
:page 4,
:position {:bounding {:x1 77.6796875,
:y1 1046.359375,
:x2 475.8890686035156,
:y2 1083.3359375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 265.2218933105469,
:y1 1046.359375,
:x2 475.8890686035156,
:y2 1064.359375,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 1065.3359375,
:x2 262.5850524902344,
:y2 1083.3359375,
:width 971,
:height 1256.5882352941176}),
:page 4},
:content {:text " highly predictive but non-robust features in standard datasets"},
:properties {:color "blue"}}
{:id #uuid "65761b65-7d7f-4620-8bad-add40dcb7321",
:page 4,
:position {:bounding {:x1 495.015625,
:y1 250.9453125,
:x2 893.2239379882812,
:y2 306.890625,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 565.982666015625,
:y1 250.9453125,
:x2 893.2239379882812,
:y2 268.9453125,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 269.921875,
:x2 893.220947265625,
:y2 287.921875,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 288.890625,
:x2 836.11572265625,
:y2 306.890625,
:width 971,
:height 1256.5882352941176}),
:page 4},
:content {:text "while assumption 2 concentrates on utilizing geometric intuition to show how the decision boundary changed between classes evolves during the training process"},
:properties {:color "red"}}
{:id #uuid "65761b77-2610-4b4b-9a2e-5a8d3ea4a51e",
:page 4,
:position {:bounding {:x1 495.015625,
:y1 433.2734375,
:x2 893.3123779296875,
:y2 489.2265625,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 830.1015625,
:y1 433.2734375,
:x2 893.3123779296875,
:y2 451.2734375,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 452.25,
:x2 893.2224731445312,
:y2 470.25,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 471.2265625,
:x2 755.2138671875,
:y2 489.2265625,
:width 971,
:height 1256.5882352941176}),
:page 4},
:content {:text "optimized gradient-based adversarial examples are not inappropriate to be test suites in respect of test coverage"},
:properties {:color "green"}}
{:id #uuid "65761b8e-cdfa-484f-b46d-edffbffc7b43",
:page 4,
:position {:bounding {:x1 495.015625,
:y1 529.28125,
:x2 893.224853515625,
:y2 604.203125,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 704.4907836914062,
:y1 529.28125,
:x2 893.2216796875,
:y2 547.28125,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 548.25,
:x2 893.221435546875,
:y2 566.25,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 567.2265625,
:x2 893.224853515625,
:y2 585.2265625,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 586.203125,
:x2 630.1485595703125,
:y2 604.203125,
:width 971,
:height 1256.5882352941176}),
:page 4},
:content {:text "Given its mathematical proof and preliminary results showing that an optimized attacker will fall into an artificial large local minimum of the loss function with high probability"},
:properties {:color "green"}}
{:id #uuid "65761bac-8ab8-49f5-901d-096c782898e9",
:page 4,
:position {:bounding {:x1 495.015625,
:y1 662.1015625,
:x2 893.2225952148438,
:y2 737.0234375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 722.5388793945312,
:y1 662.1015625,
:x2 893.2197875976562,
:y2 680.1015625,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 681.0703125,
:x2 893.221923828125,
:y2 699.0703125,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 700.046875,
:x2 893.2225952148438,
:y2 718.046875,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 719.0234375,
:x2 650.366943359375,
:y2 737.0234375,
:width 971,
:height 1256.5882352941176}),
:page 4},
:content {:text "given a normal model, the optimized attacker will probabilistic search the perturbation at multiple local minima of the loss function and the conceptual figure is shown in Fig.2"},
:properties {:color "green"}}
{:id #uuid "65761c1a-1862-4896-89c2-f935c1505df6",
:page 4,
:position {:bounding {:x1 495.015625,
:y1 805.6015625,
:x2 893.3446044921875,
:y2 843.859375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 718.1559448242188,
:y1 805.6015625,
:x2 893.3446044921875,
:y2 823.6015625,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 824.578125,
:x2 551.9439697265625,
:y2 842.578125,
:width 971,
:height 1256.5882352941176}
{:x1 542.1015625,
:y1 831.359375,
:x2 545.1796875,
:y2 843.859375,
:width 971,
:height 1256.5882352941176}),
:page 4},
:content {:text " the smallest loss at x is to label yt."},
:properties {:color "green"}}
{:id #uuid "65761c2b-8802-4ebb-8aa4-0ad48046add2",
:page 4,
:position {:bounding {:x1 495.015625,
:y1 843.546875,
:x2 893.2520751953125,
:y2 881.6796875,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 840.25244140625,
:y1 843.546875,
:x2 893.2520751953125,
:y2 861.546875,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 862.5234375,
:x2 673.234375,
:y2 880.5234375,
:width 971,
:height 1256.5882352941176}
{:x1 685.421875,
:y1 862.5234375,
:x2 771.7890625,
:y2 880.5234375,
:width 971,
:height 1256.5882352941176}
{:x1 783.984375,
:y1 862.5234375,
:x2 839.46875,
:y2 880.5234375,
:width 971,
:height 1256.5882352941176}
{:x1 674.9375,
:y1 869.1796875,
:x2 683.5390625,
:y2 881.6796875,
:width 971,
:height 1256.5882352941176}
{:x1 773.4921875,
:y1 869.1796875,
:x2 782.09375,
:y2 881.6796875,
:width 971,
:height 1256.5882352941176}
{:x1 839.328125,
:y1 869.1796875,
:x2 842.40625,
:y2 881.6796875,
:width 971,
:height 1256.5882352941176}),
:page 4},
:content {:text "minimal perturbation δ such that FΘ(x + δ)̸ = FΘ(x) = yt"},
:properties {:color "green"}}
{:id #uuid "65761c58-25f6-4184-8b0b-0a75e6937c71",
:page 4,
:position {:bounding {:x1 703.875,
:y1 1073.6328125,
:x2 889.3423461914062,
:y2 1092.9140625,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 703.875,
:y1 1073.6328125,
:x2 793.25,
:y2 1091.6328125,
:width 971,
:height 1256.5882352941176}
{:x1 805.4375,
:y1 1073.6328125,
:x2 889.3423461914062,
:y2 1091.6328125,
:width 971,
:height 1256.5882352941176}
{:x1 794.953125,
:y1 1080.4140625,
:x2 803.5546875,
:y2 1092.9140625,
:width 971,
:height 1256.5882352941176}
{:x1 820.203125,
:y1 1080.4140625,
:x2 823.28125,
:y2 1092.9140625,
:width 971,
:height 1256.5882352941176}),
:page 4},
:content {:text " A(x) is (u, FΘ, yt)-effective"},
:properties {:color "green"}}
{:id #uuid "65761c8a-4653-4bc6-aa2a-3ea82ab3b522",
:page 5,
:position {:bounding {:x1 495.015625,
:y1 307.2421875,
:x2 893.2806396484375,
:y2 364.46875,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 510.7421875,
:y1 307.2421875,
:x2 893.2806396484375,
:y2 325.2421875,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 326.2109375,
:x2 893.2182006835938,
:y2 344.2109375,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 345.1875,
:x2 776.864501953125,
:y2 363.1875,
:width 971,
:height 1256.5882352941176}
{:x1 499.671875,
:y1 351.96875,
:x2 505.8203125,
:y2 364.46875,
:width 971,
:height 1256.5882352941176}),
:page 5},
:content {:text "Adversarial attacks aim at creating perturbed inputs to achieve two primary objectives-maximizing loss while keeping lp-norm distance from the original inputs."},
:properties {:color "green"}}
{:id #uuid "65761cec-9035-43ba-bdfc-4bdc431b5974",
:page 5,
:position {:bounding {:x1 722.9436645507812,
:y1 815.1484375,
:x2 888.736083984375,
:y2 833.1484375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 722.9436645507812,
:y1 815.1484375,
:x2 888.736083984375,
:y2 833.1484375,
:width 971,
:height 1256.5882352941176}),
:page 5},
:content {:text "Reference [17] and [18]"},
:properties {:color "red"}}
{:id #uuid "65761d6a-3ff9-44c8-91a8-8c0072becbf0",
:page 6,
:position {:bounding {:x1 245.24546813964844,
:y1 837.7734375,
:x2 470.9477233886719,
:y2 855.7734375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 245.24546813964844,
:y1 837.7734375,
:x2 470.9477233886719,
:y2 855.7734375,
:width 971,
:height 1256.5882352941176}),
:page 6},
:content {:text "ithout declaring the usage scope"},
:properties {:color "yellow"}}
{:id #uuid "65761db2-5459-4ea3-8298-8704fddaef3a",
:page 6,
:position {:bounding {:x1 77.6796875,
:y1 970.59375,
:x2 475.9845886230469,
:y2 1064.359375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 426.171875,
:y1 970.59375,
:x2 475.9845886230469,
:y2 988.59375,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 989.4375,
:x2 475.88702392578125,
:y2 1007.4375,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 1008.4140625,
:x2 475.8860168457031,
:y2 1026.4140625,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 1027.390625,
:x2 475.88653564453125,
:y2 1045.390625,
:width 971,
:height 1256.5882352941176}
{:x1 77.6796875,
:y1 1046.359375,
:x2 341.75067138671875,
:y2 1064.359375,
:width 971,
:height 1256.5882352941176}),
:page 6},
:content {:text "existing powerful optimized attack searches for small perturbations to minimal loss along the gradient, which means the adversarial inputs generated by the same label have a large possibility to share similar feature representations."},
:properties {:color "green"}}
{:id #uuid "65761dc1-a97a-465a-8069-cd0f7bd4494b",
:page 6,
:position {:bounding {:x1 77.6796875,
:y1 1084.3125,
:x2 290.12890625,
:y2 1102.3125,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 77.6796875,
:y1 1084.3125,
:x2 290.12890625,
:y2 1102.3125,
:width 971,
:height 1256.5882352941176}),
:page 6},
:content {:text "similar neuron activation pattern"},
:properties {:color "green"}}
{:id #uuid "65761dcf-0a82-48e4-b519-860b04155856",
:page 6,
:position {:bounding {:x1 495.015625,
:y1 602.15625,
:x2 893.218017578125,
:y2 639.1328125,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 571.3409423828125,
:y1 602.15625,
:x2 893.218017578125,
:y2 620.15625,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 621.1328125,
:x2 799.2186889648438,
:y2 639.1328125,
:width 971,
:height 1256.5882352941176}),
:page 6},
:content {:text ", the adversarial examples can only cover limited several decision logics in the theoretical proof."},
:properties {:color "yellow"}}
{:id #uuid "65761e03-5260-47b0-a2fd-bb5759a94536",
:page 6,
:position {:bounding {:x1 495.015625,
:y1 742.8984375,
:x2 893.224609375,
:y2 779.875,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 660.100830078125,
:y1 742.8984375,
:x2 893.224609375,
:y2 760.8984375,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 761.875,
:x2 845.5339965820312,
:y2 779.875,
:width 971,
:height 1256.5882352941176}),
:page 6},
:content {:text "(maximally amplifying its entries to the full range of [0, 1] to make it visually clearer)"},
:properties {:color "green"}}
{:id #uuid "65761ed4-a2c4-482a-abbe-d5a91c056cf3",
:page 9,
:position {:bounding {:x1 94.0859375,
:y1 566.4765625,
:x2 459.51678466796875,
:y2 641.3984375,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 207.3984375,
:y1 566.4765625,
:x2 459.51678466796875,
:y2 584.4765625,
:width 971,
:height 1256.5882352941176}
{:x1 94.0859375,
:y1 585.4453125,
:x2 459.4241027832031,
:y2 603.4453125,
:width 971,
:height 1256.5882352941176}
{:x1 94.0859375,
:y1 604.421875,
:x2 459.42279052734375,
:y2 622.421875,
:width 971,
:height 1256.5882352941176}
{:x1 94.0859375,
:y1 623.3984375,
:x2 156.50460815429688,
:y2 641.3984375,
:width 971,
:height 1256.5882352941176}),
:page 9},
:content {:text "Our theoretically analysis and empirical experiment demonstrate that optimized gradient-based attack methods only cover limited several neuron activate patterns. "},
:properties {:color "green"}}
{:id #uuid "65761ee7-72a5-4f71-96b7-4f0dfcec8e5d",
:page 10,
:position {:bounding {:x1 495.015625,
:y1 648.40625,
:x2 893.2220458984375,
:y2 723.328125,
:width 971,
:height 1256.5882352941176},
:rects ({:x1 510.7421875,
:y1 648.40625,
:x2 893.1497192382812,
:y2 666.40625,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 667.375,
:x2 893.2220458984375,
:y2 685.375,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 686.3515625,
:x2 892.3280029296875,
:y2 704.3515625,
:width 971,
:height 1256.5882352941176}
{:x1 495.015625,
:y1 705.328125,
:x2 547.87353515625,
:y2 723.328125,
:width 971,
:height 1256.5882352941176}),
:page 10},
:content {:text "To the best of our knowledge, our work is the first to formally claim the contradiction between the inherent properties of existing AE algorithms and the requirements of DL test."},
:properties {:color "green"}}],
:extra {:page 11}}
Reference in New Issue View Git Blame Copy Permalink