Query-based learning applied to partially trained multilayer perceptrons

Abstract

An approach is presented for query-based neural network learning. A layered perceptron partially trained for binary classification is considered. The single-output neuron is trained to be either a zero or a one. A test decision is made by thresholding the output at, for example, one-half. The set of inputs that produce an output of one-half forms the classification boundary. The authors adopted an inversion algorithm for the neural network that allows generation of this boundary. For each boundary point, the classification gradient can be generated. The gradient provides a useful measure of the steepness of the multidimensional decision surfaces. Conjugate input pairs are generated using the boundary point and gradient information and presented to an oracle for proper classification. These data are used to refine further the classification boundary, thereby increasing the classification accuracy. The result can be a significant reduction in the training set cardinality in comparison with, for example, randomly generated data points. An application example to power system security assessment is given.

Active learning for logistic regression: an evaluation

“…Uncertainty sampling is a term invented by Lewis and Gale (1994), though the ideas can be traced back to the query methods of Hwang et al (1991) and Baum (1991). We discuss the Lewis and Gale variant since it is widely implemented and general to probabilistic classifiers such as logistic regression.…”

Driven by Compression Progress: A Simple Principle Explains Essential Aspects of Subjective Beauty, Novelty, Surprise, Interestingness, Attention, Curiosity, Creativity, Art, Science, Music, Jokes

“…For example, although noise was unpredictable and led to wildly varying target signals for the predictor, in the long run these signals did not change the adaptive predictor parameters much, and the predictor of predictor changes was able to learn this. A standard RL algorithm [114,33,109] was fed with curiosity reward signals proportional to the expected long-term predictor changes, and thus tried to maximize information gain [16,31,38,51,14] within the given limitations. In fact, we may say that the system tried to maximize an approximation of the (discounted) sum of the expected first derivatives of the data's subjective predictability, thus also maximizing an approximation of the (discounted) sum of the expected changes of the data's subjective compressibility.…”

Efficient per query information extraction from a Hamming oracle

“…The probability of the same score can be derived by adding (12) and (10), since the score will remain the same if the mutation is either deleterious or neutral.…”

“…An oracle [6], [11], [12], [13], [21] is a common source of information in assisted search. The computational overhead of the oracle often dominates the time required for a search.…”