ddiruscio/logseq
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
master
logseq/assets/Yohannis e Kolovos - 2022 - Towards model-based bias mitigation in machine lea.edn
T
ddiruscio 97fba407ce Major2
2025-06-05 22:07:12 +02:00

1496 lines
80 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 "6443aff4-d368-41c5-8e3f-0fab050a9696",
:page 1,
:position {:bounding {:x1 0,
:y1 270,
:x2 512.2776947021484,
:y2 521.0885620117188,
:width 1060.8,
:height 1372.8},
:rects ({:x1 0,
:y1 270,
:x2 0,
:y2 290,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 286,
:x2 0,
:y2 306,
:width 1060.8,
:height 1372.8}
{:x1 108.37986755371094,
:y1 465.765625,
:x2 512.2776947021484,
:y2 483.0989685058594,
:width 1060.8,
:height 1372.8}
{:x1 93.24479675292969,
:y1 484.76043701171875,
:x2 509.6593475341797,
:y2 502.09375,
:width 1060.8,
:height 1372.8}
{:x1 93.24479675292969,
:y1 503.7552185058594,
:x2 487.44065856933594,
:y2 521.0885620117188,
:width 1060.8,
:height 1372.8}),
:page 1},
:content {:text "users have to implement their code in general/statistical programming languages, which can be demanding for users with little programming and fairness in machine learning experience. "},
:properties {:color "purple"}}
{:id #uuid "6443b029-7c3c-4976-bdb2-636d041bfb2f",
:page 1,
:position {:bounding {:x1 0,
:y1 574.9895935058594,
:x2 967.5400390625,
:y2 1106,
:width 1060.8,
:height 1372.8},
:rects ({:x1 550.65625,
:y1 574.9895935058594,
:x2 967.5400390625,
:y2 592.3229370117188,
:width 1060.8,
:height 1372.8}
{:x1 551.1197967529297,
:y1 593.984375,
:x2 704.0908203125,
:y2 611.3177185058594,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 1086,
:x2 0,
:y2 1106,
:width 1060.8,
:height 1372.8}),
:page 1},
:content {:text "These are some instances that show how biases in machine learning can promote unfairness."},
:properties {:color "purple"}}
{:id #uuid "6443b074-9c01-41ef-b382-1a57aabf1efc",
:page 1,
:position {:bounding {:x1 680.4436645507812,
:y1 631.9739685058594,
:x2 925.261474609375,
:y2 649.3073120117188,
:width 1060.8,
:height 1372.8},
:rects ({:x1 680.4436645507812,
:y1 631.9739685058594,
:x2 925.261474609375,
:y2 649.3073120117188,
:width 1060.8,
:height 1372.8}),
:page 1},
:content {:text "bias metrics and debiasing algorithms"},
:properties {:color "purple"}}
{:id #uuid "6443b0c6-c048-4241-a9e7-a72164d9abf4",
:page 1,
:position {:bounding {:x1 0,
:y1 745.9453125,
:x2 968.1181030273438,
:y2 1282,
:width 1060.8,
:height 1372.8},
:rects ({:x1 568.390625,
:y1 745.9453125,
:x2 968.1181030273438,
:y2 763.4453125,
:width 1060.8,
:height 1372.8}
{:x1 551.1171875,
:y1 764.9375,
:x2 967.5322875976562,
:y2 782.4375,
:width 1060.8,
:height 1372.8}
{:x1 551.1171875,
:y1 783.9375,
:x2 967.5257568359375,
:y2 801.4375,
:width 1060.8,
:height 1372.8}
{:x1 551.1171875,
:y1 802.9296875,
:x2 605.0174560546875,
:y2 820.4296875,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 1230,
:x2 0,
:y2 1250,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 1246,
:x2 0,
:y2 1266,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 1262,
:x2 0,
:y2 1282,
:width 1060.8,
:height 1372.8}),
:page 1},
:content {:text "Data scientists usually work using their intuitions to narrow down the number of combinations of algorithms, parameters, and other factors to find the best models for given goals, datasets, and domains"},
:properties {:color "purple"}}
{:id #uuid "644a3766-d009-4482-a078-0d804d5c6544",
:page 1,
:position {:bounding {:x1 0,
:y1 802.9296875,
:x2 967.5376586914062,
:y2 1362,
:width 1060.8,
:height 1372.8},
:rects ({:x1 716.4734497070312,
:y1 802.9296875,
:x2 966.7346801757812,
:y2 820.4296875,
:width 1060.8,
:height 1372.8}
{:x1 551.1171875,
:y1 821.9296875,
:x2 967.53271484375,
:y2 839.4296875,
:width 1060.8,
:height 1372.8}
{:x1 551.1171875,
:y1 840.921875,
:x2 967.5376586914062,
:y2 858.421875,
:width 1060.8,
:height 1372.8}
{:x1 551.1171875,
:y1 859.9140625,
:x2 967.5238647460938,
:y2 877.4140625,
:width 1060.8,
:height 1372.8}
{:x1 551.1171875,
:y1 878.9140625,
:x2 967.5307006835938,
:y2 896.4140625,
:width 1060.8,
:height 1372.8}
{:x1 551.1171875,
:y1 897.90625,
:x2 945.8202514648438,
:y2 915.40625,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 1278,
:x2 0,
:y2 1298,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 1294,
:x2 0,
:y2 1314,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 1310,
:x2 0,
:y2 1330,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 1326,
:x2 0,
:y2 1346,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 1342,
:x2 0,
:y2 1362,
:width 1060.8,
:height 1372.8}),
:page 1},
:content {:text "ith lots of experimentation and trial and error [12 ], they have to go through all the narrowed combinations and test the produced models to identify which models are the best. Moreover, regardless of the availability of machine learning libraries, data scientists have to craft the search process from scratch in general/statistical programming languages (e.g., Python, R)."},
:properties {:color "green"}}
{:id #uuid "644a37aa-573a-4390-b59a-afcbcd227cc9",
:page 1,
:position {:bounding {:x1 0,
:y1 1068.8671875,
:x2 967.5353393554688,
:y2 1538,
:width 1060.8,
:height 1372.8},
:rects ({:x1 550.6484375,
:y1 1068.8671875,
:x2 967.5256958007812,
:y2 1086.3671875,
:width 1060.8,
:height 1372.8}
{:x1 551.1171875,
:y1 1087.859375,
:x2 967.5353393554688,
:y2 1105.359375,
:width 1060.8,
:height 1372.8}
{:x1 551.1171875,
:y1 1106.859375,
:x2 908.8331909179688,
:y2 1124.359375,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 1502,
:x2 0,
:y2 1522,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 1518,
:x2 0,
:y2 1538,
:width 1060.8,
:height 1372.8}),
:page 1},
:content {:text "They also do not have to code the implementation of the search process in general/statistical programming languages since it can be automatically generated and then fine-tuned later on."},
:properties {:color "green"}}
{:id #uuid "644a3855-1187-418f-bf42-23cb6705a978",
:page 1,
:position {:bounding {:x1 0,
:y1 1144.8515625,
:x2 967.5379638671875,
:y2 1586,
:width 1060.8,
:height 1372.8},
:rects ({:x1 679.5482788085938,
:y1 1144.8515625,
:x2 967.5379638671875,
:y2 1162.3515625,
:width 1060.8,
:height 1372.8}
{:x1 551.1171875,
:y1 1163.84375,
:x2 751.4285888671875,
:y2 1181.34375,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 1566,
:x2 0,
:y2 1586,
:width 1060.8,
:height 1372.8}),
:page 1},
:content {:text "model and automate bias measurement and mitigation in machine learning."},
:properties {:color "purple"}}
{:id #uuid "644a3871-b847-40e7-ab54-d14701ae1758",
:page 2,
:position {:bounding {:x1 0,
:y1 14,
:x2 511.7603759765625,
:y2 185.1015625,
:width 1060.8,
:height 1372.8},
:rects ({:x1 0,
:y1 14,
:x2 0,
:y2 34,
:width 1060.8,
:height 1372.8}
{:x1 337.1567077636719,
:y1 148.609375,
:x2 511.7603759765625,
:y2 166.109375,
:width 1060.8,
:height 1372.8}
{:x1 135.5546875,
:y1 167.6015625,
:x2 173.4437255859375,
:y2 185.1015625,
:width 1060.8,
:height 1372.8}),
:page 2},
:content {:text "bias measurement and mitigation"},
:properties {:color "yellow"}}
{:id #uuid "644a3877-8dbe-4b60-a0f4-57aa593c99c5",
:page 2,
:position {:bounding {:x1 379.03021240234375,
:y1 167.6015625,
:x2 509.6565246582031,
:y2 185.1015625,
:width 1060.8,
:height 1372.8},
:rects ({:x1 379.03021240234375,
:y1 167.6015625,
:x2 509.6565246582031,
:y2 185.1015625,
:width 1060.8,
:height 1372.8}),
:page 2},
:content {:text "bias mitigation model"},
:properties {:color "green"}}
{:id #uuid "644a39d3-3b4b-4023-b1e9-38cfeccf8d49",
:page 2,
:position {:bounding {:x1 194.8021697998047,
:y1 262.578125,
:x2 507.0864715576172,
:y2 280.078125,
:width 1060.8,
:height 1372.8},
:rects ({:x1 194.8021697998047,
:y1 262.578125,
:x2 507.0864715576172,
:y2 280.078125,
:width 1060.8,
:height 1372.8}),
:page 2},
:content {:text "experiment with different kinds of bias metrics"},
:properties {:color "green"}}
{:id #uuid "644a39da-3532-42a1-b98f-26f120dea15c",
:page 2,
:position {:bounding {:x1 0,
:y1 126,
:x2 509.9501647949219,
:y2 337.0625,
:width 1060.8,
:height 1372.8},
:rects ({:x1 0,
:y1 126,
:x2 0,
:y2 146,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 142,
:x2 0,
:y2 162,
:width 1060.8,
:height 1372.8}
{:x1 135.5546875,
:y1 281.5703125,
:x2 509.9501647949219,
:y2 299.0703125,
:width 1060.8,
:height 1372.8}
{:x1 135.5546875,
:y1 300.5703125,
:x2 509.6518859863281,
:y2 318.0703125,
:width 1060.8,
:height 1372.8}
{:x1 135.5546875,
:y1 319.5625,
:x2 321.97027587890625,
:y2 337.0625,
:width 1060.8,
:height 1372.8}),
:page 2},
:content {:text "bias mitigation algorithms, datasets, classifiers, and their parameters to find the best combinations that reduce biases but with acceptable accuracy."},
:properties {:color "green"}}
{:id #uuid "644a3a41-c8f5-494d-ad66-3cdf2a45bb94",
:page 2,
:position {:bounding {:x1 0,
:y1 606,
:x2 509.6594543457031,
:y2 963.9140625,
:width 1060.8,
:height 1372.8},
:rects ({:x1 0,
:y1 606,
:x2 0,
:y2 626,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 622,
:x2 0,
:y2 642,
:width 1060.8,
:height 1372.8}
{:x1 234.9232635498047,
:y1 908.421875,
:x2 509.6564178466797,
:y2 925.921875,
:width 1060.8,
:height 1372.8}
{:x1 93.2421875,
:y1 927.421875,
:x2 509.6594543457031,
:y2 944.921875,
:width 1060.8,
:height 1372.8}
{:x1 93.2421875,
:y1 946.4140625,
:x2 189.07754516601562,
:y2 963.9140625,
:width 1060.8,
:height 1372.8}),
:page 2},
:content {:text "“the absence of any prejudice or favouritism toward an individual or group based on their inherent or acquired characteristics”"},
:properties {:color "green"}}
{:id #uuid "644a3ada-9495-4c00-8f51-927606cd2ff5",
:page 2,
:position {:bounding {:x1 0,
:y1 638,
:x2 511.3645324707031,
:y2 1020.8984375,
:width 1060.8,
:height 1372.8},
:rects ({:x1 0,
:y1 638,
:x2 0,
:y2 658,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 654,
:x2 0,
:y2 674,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 670,
:x2 0,
:y2 690,
:width 1060.8,
:height 1372.8}
{:x1 222.2281036376953,
:y1 946.4140625,
:x2 509.42115783691406,
:y2 963.9140625,
:width 1060.8,
:height 1372.8}
{:x1 92.671875,
:y1 965.40625,
:x2 509.6632995605469,
:y2 982.90625,
:width 1060.8,
:height 1372.8}
{:x1 93.2421875,
:y1 984.40625,
:x2 511.3645324707031,
:y2 1001.90625,
:width 1060.8,
:height 1372.8}
{:x1 93.2421875,
:y1 1003.3984375,
:x2 306.12713623046875,
:y2 1020.8984375,
:width 1060.8,
:height 1372.8}),
:page 2},
:content {:text "The absence of fairness can be caused by a bias which is a systematic error or distortion from the actual state of affairs due to flaws in data collection and processing, study design, analysis, and interpretation [ 38 ]."},
:properties {:color "green"}}
{:id #uuid "644a3ae6-cfaa-4b6b-aa8e-965b590cad14",
:page 11,
:position {:bounding {:x1 0,
:y1 1035.578125,
:x2 511.0310974121094,
:y2 1076.7109375,
:width 1060.8,
:height 1372.8},
:rects ({:x1 120,
:y1 1035.578125,
:x2 509.2393493652344,
:y2 1049.078125,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 1038,
:x2 0,
:y2 1058,
:width 1060.8,
:height 1372.8}
{:x1 120,
:y1 1049.3984375,
:x2 511.0310974121094,
:y2 1062.8984375,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 1054,
:x2 0,
:y2 1074,
:width 1060.8,
:height 1372.8}
{:x1 120,
:y1 1063.2109375,
:x2 491.269775390625,
:y2 1076.7109375,
:width 1060.8,
:height 1372.8}),
:page 11},
:content {:text "Ninareh Mehrabi, Fred Morstatter, Nripsuta Saxena, Kristina Lerman, and Aram Galstyan. 2021. A Survey on Bias and Fairness in Machine Learning. ACM Comput. Surv. 54, 6, Article 115 (jul 2021), 35 pages. https://doi.org/10.1145/3457607"},
:properties {:color "green"}}
{:id #uuid "644a3b17-7aea-4f0b-be03-dc0ff4d16772",
:page 2,
:position {:bounding {:x1 0,
:y1 782,
:x2 509.95281982421875,
:y2 1172.8671875,
:width 1060.8,
:height 1372.8},
:rects ({:x1 0,
:y1 782,
:x2 0,
:y2 802,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 798,
:x2 0,
:y2 818,
:width 1060.8,
:height 1372.8}
{:x1 110.515625,
:y1 1117.375,
:x2 509.95281982421875,
:y2 1134.875,
:width 1060.8,
:height 1372.8}
{:x1 93.2421875,
:y1 1136.3671875,
:x2 509.65777587890625,
:y2 1153.8671875,
:width 1060.8,
:height 1372.8}
{:x1 93.2421875,
:y1 1155.3671875,
:x2 319.35516357421875,
:y2 1172.8671875,
:width 1060.8,
:height 1372.8}),
:page 2},
:content {:text "For example, in 2006, it was found that an algorithm used for recidivism prediction produces a much higher false-positive rate for black people than white people"},
:properties {:color "green"}}
{:id #uuid "644a3b42-adb7-47a4-880d-e295bf7b7b85",
:page 2,
:position {:bounding {:x1 0,
:y1 224.5859375,
:x2 969.8950805664062,
:y2 994,
:width 1060.8,
:height 1372.8},
:rects ({:x1 627.9932250976562,
:y1 224.5859375,
:x2 969.8950805664062,
:y2 242.0859375,
:width 1060.8,
:height 1372.8}
{:x1 551.1171875,
:y1 243.578125,
:x2 969.2374877929688,
:y2 261.078125,
:width 1060.8,
:height 1372.8}
{:x1 551.1171875,
:y1 262.578125,
:x2 967.5278930664062,
:y2 280.078125,
:width 1060.8,
:height 1372.8}
{:x1 551.1171875,
:y1 281.5703125,
:x2 759.232421875,
:y2 299.0703125,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 942,
:x2 0,
:y2 962,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 958,
:x2 0,
:y2 978,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 974,
:x2 0,
:y2 994,
:width 1060.8,
:height 1372.8}),
:page 2},
:content {:text "There are other examples of biases in machine learning, but these three examples already demonstrate that, in practice, machine learning is not always fair, and the unfairness can bring disadvantages to specific groups."},
:properties {:color "green"}}
{:id #uuid "644a3b6a-35ac-4991-8cc8-ef1568904f34",
:page 2,
:position {:bounding {:x1 0,
:y1 553.4765625,
:x2 970.1663818359375,
:y2 1218,
:width 1060.8,
:height 1372.8},
:rects ({:x1 568.390625,
:y1 553.4765625,
:x2 970.1663818359375,
:y2 570.9765625,
:width 1060.8,
:height 1372.8}
{:x1 551.1171875,
:y1 572.4765625,
:x2 847.684814453125,
:y2 589.9765625,
:width 1060.8,
:height 1372.8}
{:x1 0,
:y1 1198,
:x2 0,
:y2 1218,
:width 1060.8,
:height 1372.8}),
:page 2},
:content {:text "Euclidean Distance, Manhattan Distance, and Mahalanobis Distance are metrics to measure individual fairness"},
:properties {:color "green"}}
{:id #uuid "644a3bca-a832-48ca-91ae-859bef40d6a1",
:page 2,
:position {:bounding {:x1 568.390625,
:y1 876.40625,
:x2 749.0089721679688,
:y2 893.90625,
:width 1060.8,
:height 1372.8},
:rects ({:x1 568.390625,
:y1 876.40625,
:x2 749.0089721679688,
:y2 893.90625,
:width 1060.8,
:height 1372.8}),
:page 2},
:content {:text "Statistical Parity Difference"},
:properties {:color "green"}}
{:id #uuid "644a3bf7-d6c6-496d-af65-4a854829aa7d",
:page 3,
:position {:bounding {:x1 215.2354736328125,
:y1 773.390625,
:x2 912.7332763671875,
:y2 790.890625,
:width 1060.8,
:height 1372.8},
:rects ({:x1 215.2354736328125,
:y1 773.390625,
:x2 912.7332763671875,
:y2 790.890625,
:width 1060.8,
:height 1372.8}),
:page 3},
:content {:text "The decision tree to automatically select the most appropriate bias metrics for particular situations."},
:properties {:color "green"}}
{:id #uuid "644a3c42-6144-4d05-9317-9f85ea65bca7",
:page 3,
:position {:bounding {:x1 79.5,
:y1 209,
:x2 895.5,
:y2 698,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects (),
:page 3},
:content {:text "[:span]", :image 1682586688904},
:properties {:color "green"}}
{:id #uuid "644a3c64-e752-4389-9cdd-a3d1948891ab",
:page 2,
:position {:bounding {:x1 0,
:y1 738.7734375,
:x2 893.0048217773438,
:y2 1426,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 675.3687744140625,
:y1 738.7734375,
:x2 893.0048217773438,
:y2 754.7734375,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 508.7265625,
:y1 756.3046875,
:x2 592.9102783203125,
:y2 772.3046875,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 1406,
:x2 0,
:y2 1426,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 2},
:content {:text "SAT is suspected to contain structural discrimination"},
:properties {:color "green"}}
{:id #uuid "644a3c76-900d-44cb-95a8-f9380b1fe2d6",
:page 2,
:position {:bounding {:x1 644.7821044921875,
:y1 756.3046875,
:x2 803.1978149414062,
:y2 772.3046875,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 644.7821044921875,
:y1 756.3046875,
:x2 803.1978149414062,
:y2 772.3046875,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 2},
:content {:text "Statistical Parity Difference"},
:properties {:color "green"}}
{:id #uuid "644a3c7d-17a7-4b38-b524-b3eb3f5aa6f4",
:page 2,
:position {:bounding {:x1 508.7265625,
:y1 773.84375,
:x2 611.2523803710938,
:y2 789.84375,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 508.7265625,
:y1 773.84375,
:x2 611.2523803710938,
:y2 789.84375,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 2},
:content {:text "Disparate Impact"},
:properties {:color "green"}}
{:id #uuid "644a3c81-dd2f-4e74-a637-d0cb20e328bb",
:page 2,
:position {:bounding {:x1 0,
:y1 738.7734375,
:x2 893.1856079101562,
:y2 1458,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 595.6687622070312,
:y1 738.7734375,
:x2 893.0048217773438,
:y2 754.7734375,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 508.7265625,
:y1 756.3046875,
:x2 893.1378173828125,
:y2 772.3046875,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 508.7265625,
:y1 773.84375,
:x2 893.1856079101562,
:y2 789.84375,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 508.7265625,
:y1 791.375,
:x2 589.2079467773438,
:y2 807.375,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 1406,
:x2 0,
:y2 1426,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 1422,
:x2 0,
:y2 1442,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 1438,
:x2 0,
:y2 1458,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 2},
:content {:text "For example, SAT is suspected to contain structural discrimination [ 26, 35 ]. Statistical Parity Difference [ 18 , 26 , 35 ] and Disparate Impact [ 20, 26, 35 ] are the common metrics to measure equal fairness"},
:properties {:color "yellow"}}
{:id #uuid "644a3df9-68ba-4e15-b2c8-c98f7422d25a",
:page 3,
:position {:bounding {:x1 0,
:y1 943.671875,
:x2 899.3744506835938,
:y2 1730,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 508.7265625,
:y1 943.671875,
:x2 895.6664428710938,
:y2 959.671875,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 508.7265625,
:y1 961.203125,
:x2 893.2443237304688,
:y2 977.203125,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 508.1953125,
:y1 978.7421875,
:x2 899.3744506835938,
:y2 994.7421875,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 508.7265625,
:y1 996.2734375,
:x2 716.59375,
:y2 1012.2734375,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 1678,
:x2 0,
:y2 1698,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 1694,
:x2 0,
:y2 1714,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 1710,
:x2 0,
:y2 1730,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 3},
:content {:text "Debiasing algorithms have been developed to reduce biases in machine learning, and they can be categorised based on the stages where they are applied in machine learning pipelines: pre-processing, in-processing, and post-processing"},
:properties {:color "green"}}
{:id #uuid "644a3e02-ae48-46a5-9755-0a242f3941eb",
:page 3,
:position {:bounding {:x1 0,
:y1 996.2734375,
:x2 893.2438354492188,
:y2 1762,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 723.791259765625,
:y1 996.2734375,
:x2 893.1499633789062,
:y2 1012.2734375,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 508.7265625,
:y1 1013.8125,
:x2 893.2438354492188,
:y2 1029.8125,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 508.7265625,
:y1 1031.34375,
:x2 634.95654296875,
:y2 1047.34375,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 1726,
:x2 0,
:y2 1746,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 1742,
:x2 0,
:y2 1762,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 3},
:content {:text "The work in [ 26 , 35 ] provides guidance for choosing the most appropriate debiasing algorithms for specific situations."},
:properties {:color "green"}}
{:id #uuid "644a3e1b-78e8-4449-a717-d97ba70e719f",
:page 4,
:position {:bounding {:x1 85,
:y1 130,
:x2 900,
:y2 460,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects (),
:page 4},
:content {:text "[:span]", :image 1682587161734},
:properties {:color "green"}}
{:id #uuid "644a3e40-5617-48f2-a0e7-309689551deb",
:page 3,
:position {:bounding {:x1 0,
:y1 1066.4140625,
:x2 893.2422485351562,
:y2 1810,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 563.9707641601562,
:y1 1066.4140625,
:x2 893.2422485351562,
:y2 1082.4140625,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 508.7265625,
:y1 1083.9453125,
:x2 803.2451171875,
:y2 1099.9453125,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 1790,
:x2 0,
:y2 1810,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 3},
:content {:text "particularly in the pre-processing stage, since biases that are reduced in that stage are intrinsic in datasets"},
:properties {:color "green"}}
{:id #uuid "644a3e55-8d49-4324-bcda-faebf39ba3e9",
:page 4,
:position {:bounding {:x1 239.240234375,
:y1 538.4609375,
:x2 440.5684814453125,
:y2 554.4609375,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 239.240234375,
:y1 538.4609375,
:x2 440.5684814453125,
:y2 554.4609375,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 4},
:content {:text "Learning Fair Representation (LFR"},
:properties {:color "green"}}
{:id #uuid "644a3e77-292e-413c-9d38-524ff88516fc",
:page 4,
:position {:bounding {:x1 0,
:y1 748.875,
:x2 473.0163269042969,
:y2 1186,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 102.015625,
:y1 748.875,
:x2 473.0163269042969,
:y2 764.875,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 86.0703125,
:y1 766.40625,
:x2 263.6216735839844,
:y2 782.40625,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 1166,
:x2 0,
:y2 1186,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 4},
:content {:text "Bias mitigation can also be applied in in-processing by interfering with training algorithms. T"},
:properties {:color "green"}}
{:id #uuid "644a3ec7-5884-4911-b8a0-638a6265c4fb",
:page 5,
:position {:bounding {:x1 122.71415710449219,
:y1 581.078125,
:x2 233.7849884033203,
:y2 597.078125,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 122.71415710449219,
:y1 581.078125,
:x2 233.7849884033203,
:y2 597.078125,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 5},
:content {:text "predicted attribute,"},
:properties {:color "green"}}
{:id #uuid "644a3ecd-e055-4e5e-bbcf-4d8c17793358",
:page 5,
:position {:bounding {:x1 316.546875,
:y1 581.078125,
:x2 413.3851318359375,
:y2 597.078125,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 316.546875,
:y1 581.078125,
:x2 413.3851318359375,
:y2 597.078125,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 5},
:content {:text " favourable class"},
:properties {:color "green"}}
{:id #uuid "644a3ed1-a755-46e8-804a-52bc6e173ffb",
:page 5,
:position {:bounding {:x1 181.4375,
:y1 598.6171875,
:x2 289.6709899902344,
:y2 614.6171875,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 181.4375,
:y1 598.6171875,
:x2 289.6709899902344,
:y2 614.6171875,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 5},
:content {:text "sensitive attributes"},
:properties {:color "green"}}
{:id #uuid "644a3ed5-ae88-4c50-94ed-76a275fde574",
:page 5,
:position {:bounding {:x1 0,
:y1 414,
:x2 470.4781494140625,
:y2 632.1484375,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 0,
:y1 414,
:x2 0,
:y2 434,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 386.703125,
:y1 598.6171875,
:x2 470.4781494140625,
:y2 614.6171875,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 86.0703125,
:y1 616.1484375,
:x2 196.40066528320312,
:y2 632.1484375,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 5},
:content {:text "privileged and unprivileged classes"},
:properties {:color "green"}}
{:id #uuid "644a3ef2-5156-4b76-911c-3a22d4beadee",
:page 5,
:position {:bounding {:x1 0,
:y1 494,
:x2 470.48480224609375,
:y2 719.8203125,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 0,
:y1 494,
:x2 0,
:y2 514,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 85.640625,
:y1 686.2890625,
:x2 470.48480224609375,
:y2 702.2890625,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 86.0703125,
:y1 703.8203125,
:x2 351.94378662109375,
:y2 719.8203125,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 5},
:content {:text "They also select the bias metrics and mitigation algorithms that will be applied to datasets, models, and predictions"},
:properties {:color "green"}}
{:id #uuid "644a3f06-4ed0-4d0b-ac12-d922c6be39e9",
:page 5,
:position {:bounding {:x1 0,
:y1 526,
:x2 470.6231689453125,
:y2 807.4921875,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 0,
:y1 526,
:x2 0,
:y2 546,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 542,
:x2 0,
:y2 562,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 558,
:x2 0,
:y2 578,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 574,
:x2 0,
:y2 594,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 102.015625,
:y1 721.3515625,
:x2 470.45806884765625,
:y2 737.3515625,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 86.0703125,
:y1 738.890625,
:x2 470.52490234375,
:y2 754.890625,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 86.0703125,
:y1 756.421875,
:x2 470.6231689453125,
:y2 772.421875,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 86.0703125,
:y1 773.953125,
:x2 470.58685302734375,
:y2 789.953125,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 86.0703125,
:y1 791.4921875,
:x2 355.4044189453125,
:y2 807.4921875,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 5},
:content {:text "Measure Original Biases Prior to Prediction (T02). In this task, users measure the original biases of the datasets in different metrics, such as accuracy, mean difference, etc. The results are later used as benchmarks when compared with the measurement results after prediction (T04) or bias mitigation (T06)."},
:properties {:color "yellow"}}
{:id #uuid "644a3f9a-2318-4e82-9f20-dca4539de90b",
:page 5,
:position {:bounding {:x1 0,
:y1 734,
:x2 472.0614318847656,
:y2 1035.4375,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 0,
:y1 734,
:x2 0,
:y2 754,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 750,
:x2 0,
:y2 770,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 766,
:x2 0,
:y2 786,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 782,
:x2 0,
:y2 802,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 102.015625,
:y1 949.296875,
:x2 472.0614318847656,
:y2 965.296875,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 86.0703125,
:y1 966.8359375,
:x2 470.58990478515625,
:y2 982.8359375,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 86.0703125,
:y1 984.3671875,
:x2 470.1664733886719,
:y2 1000.3671875,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 85.546875,
:y1 1001.8984375,
:x2 470.34130859375,
:y2 1017.8984375,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 86.0703125,
:y1 1019.4375,
:x2 436.4176330566406,
:y2 1035.4375,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 5},
:content {:text "Measure Original Biases Post Prediction (T04). In this task, users measure the original accuracies and biases after performing the prediction in Task T03. The results are used as benchmarks later when comparing them against biases after bias mitigation. This action is skipped if prediction or Task T03 is not performed."},
:properties {:color "yellow"}}
{:id #uuid "644a401e-db5d-41e8-9b21-73ef015968f6",
:page 9,
:position {:bounding {:x1 0,
:y1 727.28125,
:x2 893.0045166015625,
:y2 1730,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 593.3998413085938,
:y1 727.28125,
:x2 893.0045166015625,
:y2 743.28125,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 521.703125,
:y1 744.8203125,
:x2 728.9024047851562,
:y2 760.8203125,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 1710,
:x2 0,
:y2 1730,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 9},
:content {:text "Therefore, they can provide a solid foundation for a well-bounded DSL, such as FairML."},
:properties {:color "green"}}
{:id #uuid "644a4036-a9ae-4500-b2f7-673e6debd0e2",
:page 9,
:position {:bounding {:x1 0,
:y1 885.09375,
:x2 893.2422485351562,
:y2 1890,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 701.5010375976562,
:y1 885.09375,
:x2 893.2283935546875,
:y2 901.09375,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 522.234375,
:y1 902.625,
:x2 893.2422485351562,
:y2 918.625,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 522.234375,
:y1 920.1640625,
:x2 715.7637329101562,
:y2 936.1640625,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 1854,
:x2 0,
:y2 1874,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 1870,
:x2 0,
:y2 1890,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 9},
:content {:text "we can avoid it by leveraging 3rd party engines such as P2J, which can transform textual Python programs into Jupyter notebooks"},
:properties {:color "green"}}
{:id #uuid "644a4059-8cd7-4e2d-b769-30ff7ff33612",
:page 10,
:position {:bounding {:x1 0,
:y1 734,
:x2 470.3992614746094,
:y2 1014.921875,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 0,
:y1 734,
:x2 0,
:y2 754,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 104.3984375,
:y1 989.671875,
:x2 470.3992614746094,
:y2 1002.171875,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 110.7734375,
:y1 1002.421875,
:x2 424.6435241699219,
:y2 1014.921875,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 10},
:content {:text " Julius A Adebayo et al. 2016. FairML: ToolBox for diagnosing bias in predictive modeling. Ph. D. Dissertation. Massachusetts Institute of Technology."},
:properties {:color "green"}}
{:id #uuid "644a4088-1bc7-4422-a45c-922a10764bfb",
:page 10,
:position {:bounding {:x1 0,
:y1 238,
:x2 471.9170837402344,
:y2 486.25,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 0,
:y1 238,
:x2 0,
:y2 258,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 254,
:x2 0,
:y2 274,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 270,
:x2 0,
:y2 290,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 286,
:x2 0,
:y2 306,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 302,
:x2 0,
:y2 322,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 192.58651733398438,
:y1 382.578125,
:x2 471.9170837402344,
:y2 398.578125,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 85.7109375,
:y1 400.109375,
:x2 470.5791320800781,
:y2 416.109375,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 86.0703125,
:y1 417.640625,
:x2 470.5916442871094,
:y2 433.640625,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 86.0703125,
:y1 435.1796875,
:x2 470.5902404785156,
:y2 451.1796875,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 86.0703125,
:y1 452.7109375,
:x2 470.5877380371094,
:y2 468.7109375,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 86.0703125,
:y1 470.25,
:x2 112.34947967529297,
:y2 486.25,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 10},
:content {:text ". They support data exploration, transformation, visualisation, and different algorithms for machine learning and data mining. Moreover, all of them are extensible, which means users can add new modules or scripts. However, as far as we are aware, there are no built-in modules for measuring and mitigating bias."},
:properties {:color "green"}}
{:id #uuid "644a4094-3643-4f83-bbbc-c0e0f2dbd69d",
:page 10,
:position {:bounding {:x1 0,
:y1 334,
:x2 470.4861145019531,
:y2 521.3125,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 0,
:y1 334,
:x2 0,
:y2 354,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 335.57037353515625,
:y1 487.78125,
:x2 470.4861145019531,
:y2 503.78125,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 86.0703125,
:y1 505.3125,
:x2 426.5237121582031,
:y2 521.3125,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 10},
:content {:text " Arbiter [ 52], a domain specific-language designed for ethical machine learning."},
:properties {:color "green"}}
{:id #uuid "644a40a0-218e-4db0-b416-464e9daf9813",
:page 10,
:position {:bounding {:x1 86.0703125,
:y1 557.921875,
:x2 427.32403564453125,
:y2 573.921875,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 86.0703125,
:y1 557.921875,
:x2 427.32403564453125,
:y2 573.921875,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 10},
:content {:text "transparency, fairness, accountability, and reproducibility"},
:properties {:color "green"}}
{:id #uuid "644a40bc-1b7d-4652-bb0c-e5ef852be51b",
:page 11,
:position {:bounding {:x1 0,
:y1 726.3046875,
:x2 893.6906127929688,
:y2 2034,
:width 979.1999999999999,
:height 1267.1999999999998},
:rects ({:x1 533.421875,
:y1 726.3046875,
:x2 892.126220703125,
:y2 738.8046875,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 533.421875,
:y1 739.0546875,
:x2 893.6906127929688,
:y2 751.5546875,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 533.421875,
:y1 751.8125,
:x2 807.6366577148438,
:y2 764.3125,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 1998,
:x2 0,
:y2 2018,
:width 979.1999999999999,
:height 1267.1999999999998}
{:x1 0,
:y1 2014,
:x2 0,
:y2 2034,
:width 979.1999999999999,
:height 1267.1999999999998}),
:page 11},
:content {:text "Julian Zucker and Myraeka d’Leeuwen. 2020. Arbiter: A Domain-Specific Language for Ethical Machine Learning. Association for Computing Machinery, New York, NY, USA, 421–425. https://doi.org/10.1145/3375627.3375858"},
:properties {:color "green"}}],
:extra {:page 1}}
Reference in New Issue View Git Blame Copy Permalink