Fig. 3 | Machine Learning

Fig. 3

From: On using nearly-independent feature families for high precision and confidence

Fig. 3

Each point corresponds to one problem. The x-coordinate for all points is the absolute difference in Max F1 performance of audio and visual-only base classifiers. For the first two plots, the y-coordinate is the gain, i.e., the difference in recall at 99 % (rec@99) when instances are ranked. The first plot shows the gains of NoisyOR (in rec@99) over the visual classifier, the 2nd is the gain of AVG over the Append+ classifier, and the 3rd is the gain of NoisyOR Adaptive (Sect. 3.7) over average. In general, the closer the performance of the two base classifiers, the higher the gain when using late fusion. For many of the problems, the difference in rec@99 is substantial

Back to article page