Search

13 Feb 2023: The compu-tational burden is small since the routine only needs to berun when configuring the test. ... We compareit to the CUSUM method, linear trend methods, and uni-formity tests.

13 Feb 2023: open,hard problem as it would require expensive integration over possible test datadistributions. ... Acc. ura. cy. (l) pref: cpu. Figure 4: Test set classification accuracy on classification and preference learningdatasets.

13 Feb 2023: The top row has three classes, the middle row hasfour classes and the bottom has five. ... In almost all of our tests, Archipelagohad lower test classification error than the NCNM.

13 Feb 2023: We propose estimating the likelihood of a test sequence given a learned model using particlefiltering. ... 6Different particle initialisations apply if we do not assume that the test sequence immediatelyfollows the training sequence.

19 Jun 2024: 3. Prior literature in social, economic, legal, and political sciences distinguishing between directdiscrimination and indirect discrimination makes similar observations as we do in this paper. ... For each of the classifiers, we also compute

13 Feb 2023: Table 2: Hold out test. Values are log10(p) where p is the p-value for the Spearman rank correlationhypothesis test. ... The results are shown in Figure 2.2, where we are also able to include binary variables unlikefor the Spearman test.

19 Jun 2024: In this paper, we propose to quantify unfairness using inequal-ity indices that have been extensively studied in economics andsocial welfare [3, 10, 19]. ... For all experiments, we repeatedly split the data into 70%-30%train-test sets 10 times and

13 Feb 2023: for πn in the middle of the range, versus at the ex-tremes. ... tower” cluster, and the middle row arethe images which have the most even membership in bothclusters.

13 Feb 2023: b Sampleoutputsgeneratedbythemodelafter learning. a b c. Figure4: Generative weightsof a three-layeredRGBN after beingtrainedon thestereodisparityproblem.a Weightsfrom thetop layerhidden unit to the 24 middle-layerhidden units. ... Eachof the 24

13 Feb 2023: Thiscorresponds to using the model in recall/test phase under uncer-tain input. ... which we draw 100 samples (dots under it).In the middle of the figure, the solid line represents the true under-lying function.