GYAN: A methodology for rule extraction from artificial neural networks

Artificial neural network (ANN) learning methods provide a robust and non-linear approach to approximating the target function for many classification, regression and clustering problems. ANNs have demonstrated good predictive performance in a wide variety of practical problems. However, there are strong arguments as to why ANNs are not sufficient for the general representation of knowledge. The arguments are the poor comprehensibility of the learned ANN, and the inability to represent explanation structures. The overall objective of this thesis is to address these issues by: (1) explanation of the decision process in ANNs in the form of symbolic rules (predicate rules with variables); and (2) provision of explanatory capability by mapping the general conceptual knowledge that is learned by the neural networks into a knowledge base to be used in a rule-based reasoning system. A multi-stage methodology GYAN is developed and evaluated for the task of extracting knowledge from the trained ANNs. The extracted knowledge is represented in the form of restricted first-order logic rules, and subsequently allows user interaction by interfacing with a knowledge based reasoner. The performance of GYAN is demonstrated using a number of real world and artificial data sets. The empirical results demonstrate that: (1) an equivalent symbolic interpretation is derived describing the overall behaviour of the ANN with high accuracy and fidelity, and (2) a concise explanation is given (in terms of rules, facts and predicates activated in a reasoning episode) as to why a particular instance is being classified into a certain category.

The sample complexity of pattern classification with neural networks: the size of the weights is more important than the size of the network

Sample complexity results from computational learning theory, when applied to neural network learning for pattern classification problems, suggest that for good generalization performance the number of training examples should grow at least linearly with the number of adjustable parameters in the network. Results in this paper show that if a large neural network is used for a pattern classification problem and the learning algorithm finds a network with small weights that has small squared error on the training patterns, then the generalization performance depends on the size of the weights rather than the number of weights. For example, consider a two-layer feedforward network of sigmoid units, in which the sum of the magnitudes of the weights associated with each unit is bounded by A and the input dimension is n. We show that the misclassification probability is no more than a certain error estimate (that is related to squared error on the training set) plus A3 √((log n)/m) (ignoring log A and log m factors), where m is the number of training patterns. This may explain the generalization performance of neural networks, particularly when the number of training examples is considerably smaller than the number of weights. It also supports heuristics (such as weight decay and early stopping) that attempt to keep the weights small during training. The proof techniques appear to be useful for the analysis of other pattern classifiers: when the input domain is a totally bounded metric space, we use the same approach to give upper bounds on misclassification probability for classifiers with decision boundaries that are far from the training examples.

Fine-grained plant classification using convolutional neural networks for feature extraction

We present an overview of the QUT plant classification system submitted to LifeCLEF 2014. This system uses generic features extracted from a convolutional neural network previously used to perform general object classification. We examine the effectiveness of these features to perform plant classification when used in combination with an extremely randomised forest. Using this system, with minimal tuning, we obtained relatively good results with a score of 0:249 on the test set of LifeCLEF 2014.

Improving deep convultional neural networks with unsupervised feature learning

The latest generation of Deep Convolutional Neural Networks (DCNN) have dramatically advanced challenging computer vision tasks, especially in object detection and object classification, achieving state-of-the-art performance in several computer vision tasks including text recognition, sign recognition, face recognition and scene understanding. The depth of these supervised networks has enabled learning deeper and hierarchical representation of features. In parallel, unsupervised deep learning such as Convolutional Deep Belief Network (CDBN) has also achieved state-of-the-art in many computer vision tasks. However, there is very limited research on jointly exploiting the strength of these two approaches. In this paper, we investigate the learning capability of both methods. We compare the output of individual layers and show that many learnt filters and outputs of the corresponding level layer are almost similar for both approaches. Stacking the DCNN on top of unsupervised layers or replacing layers in the DCNN with the corresponding learnt layers in the CDBN can improve the recognition/classification accuracy and training computational expense. We demonstrate the validity of the proposal on ImageNet dataset.