MAVN is a software application intended to help facilitate education and research with feedforward neural networks. It is specifically designed to recognize patterns in images.
MAVN 5.0 runs simulations that generate points on a two-dimensional image. The image uses a Voronoi diagraming algorithm to define the patterns (patterns can be thought of as shapes) within the image. The simulation is backed with a feed-forward neural network that represents the Voronoi diagram patterns (again, patterns are just shapes) within the image. The network is capable of deciding if the point that was generated by the simulation hit (landed on) a shape in the image. Based on the ratio of hits to misses for each shape on the image, statistical methods can be used to determine if the image is similar, or the same, as another image. You can learn more in the manuals page.
- Explore simple machine vision
- Define different shapes for recognition
- Try different point distributions
- Create feed-forward neural networks
- Develop statistical representations of images
- Build models in a simple spreedsheet format
- Save and reuse models with a simple .CSV format
- Open and modify models in any spreadsheet applicaiton
- Quickly run simulations
- Visualize neural networks in real-time