P. J. Werbos, "Beyond regression: New Tools for Prediction and Analysis in the Behavioral Sciences," Ph.D. dissertation, Committee on Applied Mathmatics, Harvard University, Cambridge, MA, 1974.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....this, and gives its name to a class of neural nets also known as multi layer perceptrons (MLP) A detailed description is given in the PDP volumes [17, chapter 8] which made this approach widely known. This method had in fact been developed independently by Werbos (1971) and Parker (1982) [77, 78] but the PDP group initiative in 1986 started the revival of interest in the whole field. 3.2 Perceptual versus cognitive problems Neural networks have been applied to very different types of problems and different modes of operation apply. Consider, for instance, the difference between ....

P Werbos. Beyond regression: new tools for prediction and analysis in the behavioural sciences. PhD thesis, Harvard University, 1974.

.... programming languages similar to the Forth like one, but instead based on a handful of primitive instructions for massively parallel cellular automata [65, 67, 74, 72] or on a few nonlinear operations on matrix like data structures such as those used in recurrent neural network research [69, 44, 4]. For example, we could use the principles of oops to create a non gradient based, near bias optimal variant of Hochreiter s successful recurrent network metalearner [19] It should also be of interest to study probabilistic Speed Prior based oops variants [55] and to devise applications of ....

P. J. Werbos. Beyond Regression: New Tools for Prediction and Analysis in the Behavioral Sciences. PhD thesis, Harvard University, 1974.

....function and the entropy. If the loss function is quadratic, the task is to minimize the mean (expected) square error (MSE) between the actual observed or target values and the corresponding values predicted by the network (Equation 1) The backpropagation algorithm, developed initially by Werbos [31] and then independently by Rumelhart group [25] is commonly used for training the network. BP deploys the gradient of the empirical risk function to alter the parameter set # until the risk is minimum. BP in its simple form uses a single parameter, # representing the learning rate. For a complete ....

P. Werbos. Beyond regression: new tools for prediction and analysis in the behavioral sciences. PhD thesis, Harvard University, 1974.

....in data by minimizing a risk function. The data are presented to the network and the risk function is approximated empirically Remp by summing over all data instances as follows: Remp (#) P p=1 (Y o ) 1) The Back propagation algorithm (BP) developed initially by Werbos [25] and then independently by Rumelhart group [21] is commonly used for training the network. BP uses the gradient of the empirical risk function to alter the parameter set # until the empirical risk is minimum. BP in its simple form uses a single parameter, # representing the learning rate. For a ....

P. Werbos. Beyond regression: new tools for prediction and analysis in the behavioral sciences. PhD thesis, Harvard University, 1974.

.... and Pit s model of a neuron (1943) MP43] we deal with Rosenblatt s perceptron (1958) Ros58] and Minsky and Papert s related critics (1969) MiPa69] Furthermore, we briefly consider Kohonen s theory of self organizing feature maps (1982) Koh88] standard error backpropagation (1974 1986) [Wer74, RHW86], Hopfield s fix point recurrent neural networks (1982) Hop92] the Boltzmann machine (1985) AHS85] Broomhead and Lowe s radial basis function networks (1988) BrLo88] and Elman s recurrent neural networks (1990) Elm90] The historical notes are completed by focusing on the dilemma of ....

....noticeable downturn in the research field of neural networks. According to Cowan (1990) Cow90, p. 840] there are two other major reasons For instance, the credit assignment problem was addressed by the standard error backpropagation algorithm, which was firstly introduced by Paul Werbos in 1974 [Wer74]. Unfortunately, the backpropagation algorithm received only insu#cient consideration and it took until 1986, when D. Rumelhart, G. Hinton and R. Williams reinvented the ideas of Werbos [RHW86, p. 322 8] for the downturn in the area of neural networks: First, the absence of a technological ....

[Article contains additional citation context not shown here]

Werbos P. J.: Beyond Regression: New Tools for Prediction and Analysis in the Behavioral Sciences, PhD. Thesis, Harvard University, 1974.

....be solved. The main attribute of fuzzy logic is the robustness of its interpolative reasoning mechanism. Neural networks were introduced by [5] and [6] They are computational structures that can be trained to learn by examples. Using a supervised learning algorithm, such as the back propagation [7], and a training set that samples the relation between input and output, we can perform fine local optimization. Genetic algorithms [8] give us a method to perform randomized global search in a solution space. Usually a population of candidate solutions, encoded internally as chromosomes, is ....

Werbos P., Beyond regression: new tools for predictions and analysis in the behavioral science, PhD Thesis, Harvard University, 1974

....Learning The phrase Inductive Learning is used for the class of methods that generalize from observed training examples by identifying features that empirically distinguish positive from negative training examples. Decision tree learning [36] neural network learning such as backpropagation [4, 51, 34, 7, 8], inductive logic programming [33] and genetic algorithms are all examples of inductive methods that operate in this fashion. Pure inductive learning methods formulate general hypothesis by nding empirical regularities over the training examples. The key practical limit on these inductive ....

....an activation function) One commonly used transfer function is the logistic function: h(i) 1 net(i) 3) 3.1. 2 Learning with Neural Networks One of the most widely used learning algorithms is the backpropagation algorithm which has been the cornerstone of neural learning for quite sometime [4, 6, 34, 51]. Learning in a neural network involves modifying the weights and biases of the network in order to minimize a cost function. The cost function always includes an error term measure of how close the network s predictions are to the class labels for the examples in the training set. ....

P. Werbos. Beyond Regression: New Tools for Prediction and Analysis in the Behavioral Sciences. PhD thesis, Harvard University, 1974. 61

....recurrent networks one can think to the network state at time t as if it was obtained from the t th layer output of a multilayer network with T layers, where T is the length of the sequence. The other point of view is a mathematical one and it is based on the Werbos theory of ordered derivatives [5]. Werbos provided a mathematical tool to rigorously compute the derivatives of a certain variable with respect to another one in complex structures described by ordered mathematical relations (for example a neural network) Actually, with ordered derivatives it is possible to derive both BPTT and ....

P.J. Werbos. Beyond regression: New tools for prediction and analysis in the behavioural sciences. Ph.D. dissertation, Committee on Appl. Math., Harvard Univ., Cambridge, MA, Nov. 1974.

....of the forecast trends, and therefore it could be useful for trading. As the aim of financial forecasting is to maximize profit, we suggest that NMSE should not be used as the unique criterion of forecasting performance. 3 Backpropagation Networks with Profit Factor The backpropagation network [7, 8] is the most popular and most widely implemented neural network especially in the financial forecasting domain. It is based on a multi layered feedforward topology with supervised learning. The network is fully connected with every node in the lower layer linked to every node Table 2: Statistics ....

Werbos, P.J., Beyond Regression: New Tools for prediction and analysis in the behavioral Science, PhD Thesis, Harvard University, 1974.

No context found.

P. J. Werbos, "Beyond regression: New Tools for Prediction and Analysis in the Behavioral Sciences," Ph.D. dissertation, Committee on Applied Mathmatics, Harvard University, Cambridge, MA, 1974.

No context found.

P. J. Werbos. Beyond Regression: New Tools for Prediction and Analysis in the Behavioral Sciences. PhD thesis, Harvard University, 1974.

No context found.

P. J. Werbos. Beyond Regression: New Tools for Prediction and Analysis in the Behavioral Sciences. PhD thesis, Harvard University, 1974.

No context found.

Werbos PJ. Beyond regression: New tools for prediction and analysis in the behavioral sciences. PhD thesis, Harvard University, Cambridge, MA, 1974

No context found.

P. J. Werbos. Beyond Regression: New Tools for Prediction and Analysis in the Behavioral Sciences. PhD thesis, Harvard University, 1974.

No context found.

Werbos, P.J, Beyond regression : New tools for prediction and analysis in the behavioral sciences, Doctoral Dissertation, Apply. Math., Harvard University, November 1974.

No context found.

Werbos P. (1974), "Beyond regression: new tools for prediction and analysis in the behavioral sciences", PhD thesis, Harvard University.

No context found.

P.J. Werbos. Beyond Regression: New Tools for Prediction and Analysis in the Behavior Sciences. PhD thesis, Harvard University, Committee on Applied Mathematics, 1974.

No context found.

P. J. Werbos, Beyond Regression: New Tools for Prediction and Analysis in the Behavioral Sciences. PhD thesis, Harvard University, 1974.

No context found.

P. J. Werbos, Beyond Regression: New Tools for Prediction and Analysis in the Behavioral Sciences. PhD thesis, Harvard University, 1974.

No context found.

P.J. Werbos. Beyond regression: New tools for prediction and analysis in the behavioral sciences. PhD thesis, Harvard University, Cambridge, MA, 1974.

No context found.

Paul Werbos. Beyond Regression: New Tools for Prediction and Analysis in the Behavioral Sciences. PhD thesis, Harvard University, Cambridge, Mass., 1974. 3.

No context found.

P. Werbos. Beyond Regression: New tools for prediction and analysis in the behavioral science. PhD thesis, Department of Applied Mathematics, Harvard University, Cambridge, Mass., 1974.

No context found.

P. Werbos. Beyond Regression: New Tools for Prediction and Analysis in the Behavioural Sciences. PhD thesis, Harvard University, Cambridge, MA, 1974.

No context found.

Paul Werbos. Beyond Regression: New Tools for Prediction and Analysis in the Behavioral Sciences. PhD thesis, Harvard University, Cambridge, Mass., 1974. III.

No context found.

P. WERBOS, Beyond Regression: New Tools for Prediction and Analysis in the Behavioral Sciences, PhD thesis, Committee on Appl. Math., Harvard Univ., Cambridge, Mass., November 1974.

First 50 documents  Next 50

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