Neural trees

A computer-implemented method of training a predictor to predict an outcome y given an example x, the method comprising:

• storing in a memory a plurality of training examples comprising examples x for which outcomes y are known;
• accessing from the memory at least one decision tree comprising a plurality of nodes connected by edges, the nodes comprising a root node, internal nodes and leaf nodes;
• wherein individual ones of the nodes and individual ones of the edges each have an assigned module comprising parameterized, differentiable operations, such that for each of the internal nodes the module computes a binary outcome for selecting a child node of the internal node;
• a processor configured to grow the decision tree by, for a current node in a layer of the tree furthest from the root node, deciding whether to:
• add another module to the incoming edge of the current node, add another node to the current node, or terminate growing for the current node;
• wherein the decision is made by using a validation set of the training examples.

A predictor, comprising:

• a memory which stores at least one example x for which an outcome y is not known;
• the memory storing at least one decision tree comprising a plurality of nodes connected by edges, the nodes comprising a root node, internal nodes and leaf nodes;
• wherein, individual ones of the nodes and individual ones of the edges each have an assigned module, comprising parameterized, differentiable operations, such that for each of the internal nodes the module computes a binary outcome for selecting a child node of the internal node;
• a processor configured to compute the prediction y by processing the example x using a plurality of the differentiable operations selected according to a path through the tree from the root node to a leaf node;
• and wherein the decision tree has been formed using a growing process which is dependent on a set of training data comprising examples x for which outcomes y are known.