Pratt, L.Y., "Using the discriminability based transfer algorithm to selectively bias neural network learning using the results of learning on related tasks", Technical Report MACS-93-29?, Dept. of Mathematical and Computer Sciences, Colorado School of Mines, Golden CO 80401. Home/Search Document Details and Download
Summary Related Articles Check |
This paper is cited in the following contexts:
Learning Background Effects in Object Recognition - Brent Taylor (Correct)
....learning are commonly used to learn domain theories for image based recognition tasks: artificial neural networks, pure induction methods, explanationbased learning (EBL) and theory revision. Although a powerful learning technique, artificial neural networks have not with the notable exception of [14] capitalized on prior knowledge, are difficult to understand, and require large training sets (on the order of 5000 examples) The reader is directed to [17] for an overview of neural networks for automatic target recognition. Pure induction methods like C4.5 [15] while flexible and produce ....
Pratt, L.Y., "Using the discriminability based transfer algorithm to selectively bias neural network learning using the results of learning on related tasks", Technical Report MACS-93-29?, Dept. of Mathematical and Computer Sciences, Colorado School of Mines, Golden CO 80401.
A Survey of Machine Learning Techniques for Automatic Target.. - Murphy, Taylor (Correct)
....domain theory from scratch. Given that neural networks require large databases to train from, this can be quite time consuming and cause the ATR to operate with a blind spot while the new domain theory is being constructed. The most notable exception to this lack of adaptability is work by Pratt [22] in transfer. Consider a network which performs adequately for recognizing targets in predominately forested regions but does not perform well for recognizing the same targets with the same inputs and outputs in extreme desert conditions. Rather than initialize building the new neural net for ....
....building the new neural net for recognition in the desert with random weights and possibly spend days training the network, a significant speed up in training may be possible by initializing weights extracted from the original network through a discriminability based transfer (DBT) algo8 rithm. In [22], networks for speech recognition, heart disease diagnosis, DNA promotion, and analysis of winning position in chess using the DBT approach trained faster and had better overall performance than networks either initialized with random weights or all the weights for the original network (literal ....
Pratt, L.Y., "Using the discriminability based transfer algorithm to selectively bias neural network learning using the results of learning on related tasks", Technical Report MACS-93-29?, Dept. of Mathematical and Computer Sciences, Colorado School of Mines, Golden CO 80401.
Online articles have much greater impact More about CiteSeer.IST Add search form to your site Submit documents Feedback
CiteSeer.IST - Copyright Penn State and NEC