Let C be any concept class and #4 be any hypothesis class. Let ho and h, be representations of the identically 0 and identically 1 functions, respectively. Show that if there is a randomized algorithm for efficiently PAC learning C using 76, then there is a deterministic algorithm for efficiently PAC learning C using HU {ho, hi}.The random nature of the examples can be removed using the von Neumann method to turn a biased coin into a fair coin. This method hinges on the fact that the probability of drawing the two orders of labels (+, -) and (-, +) are the same. This can then be used to draw an equal amount of each of the two labels for the purposes of learning. This is done by drawing two examples and keeping the first example if the labels differ on both. If the distribution is very heavily biased towards one of the labels the probability is of drawing two different labels in succession is very low. And the deterministic method would return either ho for negative bias, or hi for a positively biased distribution, as both of these would be a good hypothesis in this case. This new algorithm is deterministic for efficiently pac learning, as it uses either a fair distribution of positive and negative examples for efficiently learning. Or in the case of the heavily biased distribution can fall back on ho or hi