This directory (ftp://ftp.sas.com/pub/neural/) contains the Neural
Network FAQ, several papers on neural networks, and various other
more-or-less related things.

In research published this week in Neuron, Marcos Frank, PhD, Assistant Professor of Neuroscience, at the University of Pennsylvania School of Medicine, postdoctoral researcher Sara Aton, PhD, and colleagues describe for the first time how cellular changes in the sleeping brain promote the formation of memories.

"This is the first real direct insight into how the brain, on a cellular level, changes the strength of its connections during sleep," Frank says.

The findings, says Frank, reveal that the brain during sleep is fundamentally different from the brain during wakefulness.

"We find that the biochemical changes are simply not happening in the neurons of animals that are awake," Frank says. "And when the animal goes to sleep it's like you’ve thrown a switch, and all of a sudden, everything is turned on that's necessary for making synaptic changes that form the basis of memory formation. It's very striking."

SIMBRAIN is a free tool for building, running, and analyzing neural-networks (computer simulations of brain circuitry). Simbrain aims to be as visual and easy-to-use as possible. See our design goals. Unique features of Simbrain include its integrated "world components" and its ability to represent a network's state space. Simbrain is written in Java and runs on Windows, Mac OS X, and Linux. Click here for a video introduction to some of Simbrain's features, check out the screenshots, or just download the software and start experimenting. Simbrain is open source, and constantly evolving. We'd love for you to join our team. To discuss any aspect of Simbrain check out the forum.

One of the last remaining problems in science is the riddle of consciousness. The human brain—a mere lump of jelly inside your cranial vault—can contemplate the vastness of interstellar space and grapple with concepts such as zero and infinity. Even more remarkably it can ask disquieting questions about the meaning of its own existence. "Who am I" is arguably the most fundamental of all questions. It really breaks down into two problems—the problem of qualia and the problem of the self. My colleagues, the late Francis Crick and Christof Koch have done a valuable service in pointing out that consciousness might be an empirical rather than philosophical problem, and have offered some ingenious suggestions. But I would disagree with their position that the qualia problem is simpler and should be addressed first before we tackle the "Self." I think the very opposite is true. I have every confidence that the problem of self will be solved within the lifetimes of most people reading this c...

§ Grounding transfer in neural networks § Evolution of syntax § Co-evolution of language and brain § Language universals § Language in evolutionary robots

There are many different approaches to optical character recognition problem. One of the most common and popular approaches is based on neural networks, which can be applied to different tasks, such as pattern recognition, time series prediction, function approximation, clustering, etc. In this article, I'll try to review some approaches for optical character recognition using artificial neural networks. The attached project is aimed as a research project, so don't try to find here a ready solution for scanned document processing.

We have created a network that allocates a new computational unit whenever an unusual pattern is presented to the network. This network forms compact representations, yet learns easily and rapidly. The network can be used at any time in the learning process and the learning patterns do not have to be repeated. The units in this network respond to only a local region of the space of input values. The network learns by allocating new units and adjusting the parameters of existing units. If the network performs poorly on a presented pattern, then a new unit is allocated that corrects the response to the presented pattern. If the network performs well on a presented pattern, then the network parameters are updated using standard LMS gradient descent. We have obtained good results with our resource-allocating network (RAN). For predicting the Mackey-Glass chaotic time series, RAN learns much faster than do those using backpropagation networks and uses a comparable number of synapses.

Using a simple four-layer perceptron and standard backprop (You're going to say I cheated.) While the Two-Spirals Problem may be (or actually, may not be) a valuable benchmark for neural network researchers working on new architectures, I believe that it much more valuable, and to a much larger number of people (those working on applied NN solutions), as an example of THE WRONG WAY to present data to a neural network. This paper is intended as a primer on how to evaluate and manipulate your dataset prior to building a neural network to improve or even guarantee your results.

The code on this page is placed in the public domain with the hope that others will find it a useful starting place for developing their own software. Most is not well-documented nor thoroughly tested. Reinforcement Learning Read about a MATLAB implementation of Q-learning and the mountain car problem here. That page also includes a link to the MATLAB code that implements a GUI for controlling the simulation. Want to try your hand at balancing a pole? Try one of the following. The most recent version is first. * download Pole.hs, a Haskell implementation that uses Gtk2hs, or the executable for Intel machines, * download Pole.java. or run the Java applet, * download pole.tcl , a Tcl/Tk implementation Here is code for learning to balance a pole, used for experiments described in Strategy Learning with Multilayer Connectionist Representations, by C. Anderson, in the Proceedings of the Fourth International Workshop on Machine Learning, Irvine, CA, 1987. It includes C code and a README ex...

A convolutional neural network achieves 99.26% accuracy on a modified NIST database of hand-written digits.

A C++ class implementing a back-propagation algorithm neural net, that supports any number of layers/neurons.

lists of articles about reinforcement learning

Platt's resource-allocation network RAN Platt, 1991a, 1991b is modi ed for a reinforcement-learning paradigm and to restart" existing hidden units rather than adding new units. After restart- ing, units continue to learn via back-propagation. The resulting restart algorithm is tested in a Q-learning network that learns to solve an inverted pendulum problem. Solutions are found faster on average with the restart algorithm than without it.

The MNIST database of handwritten digits, available from this page, has a training set of 60,000 examples, and a test set of 10,000 examples. It is a subset of a larger set available from NIST. The digits have been size-normalized and centered in a fixed-size image. It is a good database for people who want to try learning techniques and pattern recognition methods on real-world data while spending minimal efforts on preprocessing and formatting.

In this dissertation, a novel training algorithm for neural networks, named Parameter Incremental Learning (PIL), is proposed, developed, analyzed and numerically validated. The main idea of the PIL algorithm is based on the essence of incremental supervised learning: that the learning algorithm, i.e., the update law of the network parameters, should not only adapt to the newly presented input-output training pattern, but also preserve the prior results. A general PIL algorithm for feedforward neural networks is accordingly derived, using the first-order approximation technique, with appropriate measures of the performance of preservation and adaptation. The PIL algorithms for the Multi-Layer Perceptron (MLP) are subsequently derived by applying the general PIL algorithm, augmented with the introduction of an extra fictitious input to the neuron. The critical point in obtaining an analytical solution of the PIL algorithm for the MLP is to apply the general PIL algorithm at the neuron...

mathematical description of the backpropagation algorithm

Written in Python. See http://www.python.org/ Placed in the public domain. Neil Schemenauer <nas@arctrix.com>

In this book we explore a computational approach to learning from interaction. Rather than directly theorizing about how people or animals learn, we explore idealized learning situations and evaluate the effectiveness of various learning methods. That is, we adopt the perspective of an artificial intelligence researcher or engineer. We explore designs for machines that are effective in solving learning problems of scientific or economic interest, evaluating the designs through mathematical analysis or computational experiments. The approach we explore, called reinforcement learning, is much more focused on goal-directed learning from interaction than are other approaches to machine learning.

Document details from CiteSeerX (Isaac Councill, Lee Giles): The purpose of this tutorial is to provide an introduction to reinforcement learning (RL) at a level easily understood by students and researchers in a wide range of disciplines. The intent is not to present a rigorous mathematical discussion that requires a great deal of effort on the part of the reader, but rather to present a conceptual framework that might serve as an introduction to a more rigorous study of RL. The fundamental principles and techniques used to solve RL problems are presented. The most popular RL algorithms are presented. Section 1 presents an overview of RL and provides a simple example to develop intuition of the underlying dynamic programming mechanism. In Section 2 the parts of a reinforcement learning problem are discussed. These include the environment, reinforcement function, and value function. Section 3 gives a description of the most widely used reinforcement learning algorithms. These include...