The short definition of a parameter sweep is that it's the process of trying different training parameter values in order to find a good set of neural network weight values.
- By James McCaffrey
- 11/10/2015
Recurrent neural networks can solve some types of problems that regular feed-forward networks cannot handle.
- By James McCaffrey
- 10/12/2015
Neural network models can be created, saved and reused. Here's how.
- By James McCaffrey
- 09/10/2015
The differences between neural network binary classification and multinomial classification are surprisingly tricky. McCaffrey looks at two approaches to implement neural network binary classification.
- By James McCaffrey
- 08/18/2015
Let's explore mini-batch training, the third among a variety of back-propagation algorithms you can use for training a neural network.
- By James McCaffrey
- 07/21/2015
Here's how to use non-standard activation functions to customize your neural network system.
- By James McCaffrey
- 06/11/2015
The train-validate-test process is hard to sum up in a few words, but trust me that you'll want to know how it's done to avoid the issue of model overfitting when making predictions on new data.
- By James McCaffrey
- 05/13/2015
Back-Propagation is the most common algorithm for training neural networks. Here's how to implement it in C#.
- By James McCaffrey
- 04/14/2015
It's more complex than back propagation, but Rprop has advantages in training speed and efficiency.
- By James McCaffrey
- 03/09/2015
Now that you know how to work with multi-swarm optimization, it's time to take it up a level and see how to train your network to use it.
- By James McCaffrey
- 02/04/2015
Multi-swarm optimization (MSO) is a powerful variation of particle swarm optimization. Understanding how MSO works and how to implement it can be a valuable addition to your developer toolkit.
- By James McCaffrey
- 01/20/2015
Python is James's preferred language for hybrid environments. Here's how to implement neural network back-propagation training using it.
- By James McCaffrey
- 12/18/2014
A neural network implementation can be a nice addition to a Python programmer's skill set. If you're new to Python, examining a neural network implementation is a great way to learn the language.
- By James McCaffrey
- 11/12/2014
Simplex optimization is one of the simplest algorithms available to train a neural network. Understanding how simplex optimization works, and how it compares to the more commonly used back-propagation algorithm, can be a valuable addition to your machine learning skill set.
- By James McCaffrey
- 10/16/2014
Dr. McCaffrey walks you through how to use the Microsoft Azure Machine Learning Studio, a new front-end for Microsoft Azure Machine Learning, to get a neural prediction system up and running.
- By James McCaffrey
- 09/18/2014
There are two different techniques for training a neural network: batch and online. Understanding their similarities and differences is important in order to be able to create accurate prediction systems.
- By James McCaffrey
- 08/18/2014
Weight decay and weight restriction are two closely related, optional techniques that can be used when training a neural network. This article explains exactly what weight decay and weight restriction are, and how to use them with an existing neural network application or implement them in a custom application.
- By James McCaffrey
- 07/28/2014
Deep Neural Networks are the more computationally powerful cousins to regular neural networks. Learn exactly what DNNs are and why they are the hottest topic in machine learning research.
- By James McCaffrey
- 06/13/2014
Dropout training is a relatively new algorithm which appears to be highly effective for improving the quality of neural network predictions. It's not yet widely implemented in neural network API libraries. Learn how to use dropout training if it's available in an existing system, or add dropout training to systems where it's not yet available.
- By James McCaffrey
- 05/13/2014
To train a neural network you need some measure of error between computed outputs and the desired target outputs of the training data. The most common measure of error is called mean squared error. However, there are some research results that suggest using a different measure, called cross entropy error, is sometimes preferable to using mean squared error.
- By James McCaffrey
- 04/22/2014
Evolutionary optimization can be used to train a neural network. A virtual chromosome holds the neural network's weights and bias values, and the error term is the average of all errors between the network's computed outputs and the training data target outputs. Learn how to code the solution.
- By James McCaffrey
- 03/26/2014
Evolutionary optimization (EO) is a type of genetic algorithm that can help minimize the error between computed output values and training data target output values. Use this demo program to learn to the method.
- By James McCaffrey
- 02/18/2014
Understanding data encoding and normalization is an absolutely essential skill when working with neural networks. James McCaffrey walks you through what you need to know to get started.
- By James McCaffrey
- 01/15/2014
Although mathematically elegant, back-propagation isn't perfect. Instead consider using particle swarm optimization (PSO) to train your neural network; here's how.
- By James McCaffrey
- 12/18/2013
Particle swarm optimization isn't usually seen as the first-choice technique for training a neural network but, as James McCaffrey demonstrates, it's a useful alternative.
- By James McCaffrey
- 11/25/2013