prof. dr. Lambert Schomaker Shattering two binary dimensions over a number of classes Kunstmatige Intelligentie / RuG
2 Samples and classes In order to understand the principle of shattering sample points into classes we will look at the simple case of two dimensions of binary value
3 2-D feature space f1f1 f2f2
4 2-D feature space, 2 classes f1f1 f2f2
5 the other class… f1f1 f2f2
6 2 left vs 2 right f1f1 f2f2
7 top vs bottom f1f1 f2f2
8 right vs left f1f1 f2f2
9 bottom vs top f1f1 f2f2
10 lower-right outlier f1f1 f2f2
11 lower-left outlier f1f1 f2f2
12 upper-left outlier f1f1 f2f2
13 upper-right outlier f1f1 f2f2
14 etc f1f1 f2f2
15 2-D feature space f1f1 f2f2
16 2-D feature space f1f1 f2f2
17 2-D feature space f1f1 f2f2
18 XOR configuration A f1f1 f2f2
19 XOR configuration B f1f1 f2f2
20 2-D feature space, two classes: 16 hypotheses f 1 =0 f 1 =1 f 2 =0 f 2 = “hypothesis” = possible class partioning of all data samples
21 2-D feature space, two classes, 16 hypotheses f 1 =0 f 1 =1 f 2 =0 f 2 = two XOR class configurations: 2/16 of hypotheses requires a non-linear separatrix
22 XOR, a possible non-linear separation f1f1 f2f2
23 XOR, a possible non-linear separation f1f1 f2f2
24 2-D feature space, three classes, # hypotheses? f 1 =0 f 1 =1 f 2 =0 f 2 = ……………………
25 2-D feature space, three classes, # hypotheses? f 1 =0 f 1 =1 f 2 =0 f 2 = …………………… 3 4 = 81 possible hypotheses
26 Maximum, discrete space Four classes: 4 4 = 256 hypotheses Assume that there are no more classes than discrete cells Nhypmax = ncells nclasses
27 2-D feature space, three classes… f1f1 f2f2 In this example, is linearly separatable from the rest, as is . But is not linearly separatable from the rest of the classes.
28 2-D feature space, four classes… f1f1 f2f2 Minsky & Papert: simple table lookup or logic will do nicely.
29 2-D feature space, four classes… f1f1 f2f2 Spheres or radial-basis functions may offer a compact class encapsulation in case of limited noise and limited overlap (but in the end the data will tell: experimentation required!)