Machine Learning Methods. Similarly, an XXX learns by being fed data (e.g. images, or real-world visuals of tanks) from which it derives conclusions (e.g. recognising a ‘tank’). With a xxxxxxxx xxxx of data, the XXX develops detailed algorithmic rules, which enable it to draw correct conclusions in the field.179 This can be done through supervised learning, whereby human agents feed the machine labelled input data and the desired result,180 after which learning is effectuated from both correct predictions and backpropagation of errors.181 Namely, when an incorrect output is provided (e.g. misidentifying a bus for a tank) an error value is generated, which reflects the divergence between the correct 176 Xxxxxx Xxxxxxxx, The Master Algorithm: How the Quest for the Ultimate Learning Machine Will Remake Our World (Xxxxx Xxxx, 2015), 6. 177 Bharat Bhosal, ‘Curvelet Interaction with Artificial Neural Networks’ in Subana Xxxxxxxxxxxxxx and Xxxxxxx Xxxxxxxxxxxx (eds.), Artificial Neural Network Modelling (Springer International, 2016), 117. 178 Ibid. (noting that ANNs mimic this biological process with an artificial structure consisting of processing elements (nodes); interconnections between those elements/nodes; an activation function (rule), which transforms input into output (inside a node); and a learning function, which manages the weights of input-output pairs. Accordingly, ANNs ‘learn from example’ or, more correctly, from input- output data samples. This entails a strengthening or weakening of connections between the many nodes in the system via statistical correlation, until the desired behaviour is realised. Knowledge is therefore encoded in the strength of the neural connections). 179 Ibid. 180 The desired result is some variable X (in this case, a tank) that is designated as the target for prediction, explanation or inference. Moreover, the values of X in the dataset (the tanks in the image samples) constitute the “ground truth” values for learning. See Xxxxx Xxxxx, ‘Learning’, in Frankish and Xxxxxx (n 164) 151, 154. 181 Xxxxxxx Xxxxxxxxxxx, ‘Deep Learning in Neural Networks: An Overview’ (8 October 2014) <xxxxx://xxxxx.xxx/pdf/1404.7828.pdf> accessed 13 May 2018. and incorrect answer.182 The error value is then backpropagated through the XXX, which updates the input-output weights/connections between the nodes, to correctly output ‘tank’ when an image of a tank is next shown.183 Thus, the algorithm learns to generalise its training data to new instances of ‘tank’. Supervised...
