It is often difficult to predict the optimal neural network size for a particular application. Constructive or destructive methods that add or subtract neurons, layers, connections, etc. might offer a solution to this problem. We prove that one method, recurrent cascade correlation, due to its topology, has fundamental limitations in representation and thus in its learning capabilities. It cannot represent with monotone (i.e., sigmoid) and hard-threshold activation functions certain finite state automata. We give a "preliminary" approach on how to get around these limitations by devising a simple constructive training method that adds neurons during training while still preserving the powerful fully-recurrent structure. We illustrate this approach by simulations which learn many examples of regular grammars that the recurrent cascade correlation method is unable to learn.

We examine the representational capabilities of first-order and second-order single-layer recurrent neural networks (SLRNN's) with hard-limiting neurons. We show that a second-order SLRNN is strictly more powerful than a first-order SLRNN. However, if the first-order SLRNN is augmented with output layers of feedforward neurons, it can implement any finite-state recognizer, but only if state-splitting is employed. When a state is split, it is divided into two equivalent states. The judicious use of state-splitting allows for efficient implementation of finite-state recognizers using augmented first-order SLRNN's.

We show that a recurrent, second-order neural network using a real-time, forward training algorithm readily learns to infer small regular grammars from positive and negative string training samples. We present simulations that show the effect of initial conditions, training set size and order, and neural network architecture. All simulations were performed with random initial weight strengths and usually converge after approximately a hundred epochs of training. We discuss a quantization algorithm for dynamically extracting finite state automata during and after training. For a well-trained neural net, the extracted automata constitute an equivalence class of state machines that are reducible to the minimal machine of the inferred grammar. We then show through simulations that many of the neural net state machines are dynamically stable, that is, they correctly classify many long unseen strings. In addition, some of these extracted automata actually outperform the trained neural network for classification of unseen strings.