-
2019 Jun 10
Earthquakes in the Landscape of Neural Network
In this article, I want to direct your attention to the less known properties of one, quite famous, technique in the deep learning. I want to show you how beautiful and interesting could be a concept that typically left behind because of all more exciting ideas in this area.
-
2018 Mar 26
Making Art with Growing Neural Gas
Article shows how to generate unique styles from any image using Growing Neural Gas (GNG). In addition, it explains how this type of neural network works and what problems user might encounter while training it on different images.
-
2017 Dec 17
Create unique text-style with SOFM
Neupy's logo has been generated with a help of the neural network. This article shows the process and how it could be extended for some other text.
-
2017 Dec 13
The Art of SOFM
It's quite rare that algorithm can not only extract knowledge from the data, but also produce something beautiful using exactly the same set of training rules without any modifications.
SOFM is a great example of the algorithm that can produce simple work of art when used in the right way.
-
2017 Dec 09
Self-Organizing Maps and Applications
Self-Organizing Feature Map (SOFM or SOM) is a simple algorithm for unsupervised learning. It can be applied to solve vide variety of problems. It quite good at learning topological structure of the data and it can be used for visualizing deep neural networks.
This article explains how SOFM works and shows different applications where it can be used.
-
2016 Dec 17
Hyperparameter optimization for Neural Networks
This article explains different hyperparameter algorithms that can be used for neural networks. It covers simple algorithms like Grid Search, Random Search and more complicated algorithms like Gaussian Process and Tree-structured Parzen Estimators (TPE).
-
2016 Nov 12
Image classification, MNIST digits
This short tutorial shows how to design and train simple network for digit classification in NeuPy.
-
2015 Sep 21
Password recovery
Discrete hopfiled networks can be used to solve wide variety of problems. In this article, we try to use this type of network in order to memorizes user's password and then we try reconstruct it from partially corrupted version of this password.
-
2015 Sep 20
Discrete Hopfield Network
In this article, we describe core ideas behind discrete hopfield networks and try to understand how it works. In addition, we explore main problems related to this algorithm. And finally, we take a look into simple example that aims to memorize digit patterns and reconstruct them from corrupted samples.
-
2015 Jul 04
Predict prices for houses in the area of Boston
Boston house prices is a classical dataset for regression. This article shows how to make a simple data processing and train neural network for house price prediction.
-
2015 Jul 04
Visualize Algorithms based on the backpropagation
Typical neural networks have mullions of parameters and it's quite difficult to visualize the process. In the article, we visualize training of the network that has only 2 parameters. It allows us to explore different training algorithms and see how it behaves during the training
These type of visualizations can provide us with useful insights about the training process.