-
Welcome to B4X forum!
B4X is a set of simple and powerful cross platform RAD tools:
- B4A (free) - Android development
- B4J (free) - Desktop and Server development
- B4i - iOS development
- B4R (free) - Arduino, ESP8266 and ESP32 development
Neural Networks
- Thread starter wonder
- Start date
- Similar Threads Similar Threads
Yes, I've been playing with it off and on over the past couple of months. Nothing too serious as I have some major projects to get done over the next year and a half, though. This is just a project that, if it works out, may turn into something serious in 2 or 3 years.
I'm using B4J and B4R to lay a neural network over the top of a mesh network I made last year using environmental sensors and some limited decision-making algorithms within and among those sensors (the B4R part) in the mesh. This neural overlay will incorporate the basic "Back Propagation" method of training as a start. The B4J part will analyze the outputs (checking the decision tree and verifying data with an outside source for the environmental stuff) and, somewhat, help coordinate multiple 'neurons' down the road.
It should be a fun little foray into neural networking programming with real-world application that can be seen and understood in real-time. I hope.
I'm using B4J and B4R to lay a neural network over the top of a mesh network I made last year using environmental sensors and some limited decision-making algorithms within and among those sensors (the B4R part) in the mesh. This neural overlay will incorporate the basic "Back Propagation" method of training as a start. The B4J part will analyze the outputs (checking the decision tree and verifying data with an outside source for the environmental stuff) and, somewhat, help coordinate multiple 'neurons' down the road.
It should be a fun little foray into neural networking programming with real-world application that can be seen and understood in real-time. I hope.
You could have a look at this, I posted a link to a library "DeepLearning4j":
https://www.b4x.com/android/forum/threads/deeplearning4j.80595/
It could be enough for what you need? Are you interested in building a neural network from scratch or interested in making an app / game with it?
Could you elaborate on that please? Did you write your own neural network algorithm?
It's in the beginning stages, but, yes, it's my own algorithm. I developed a mesh network with B4R and arduinos to monitor and react to environmental conditions over a large area. The network can adjust itself based on nodes (the arduinos with B4R code) being added / subtracted and entering / exiting the network. It also adjusts each node in the network to configure itself based on some environmental conditions either of the node itself or other nodes within the network in real-time. I added an aggregator to the network in the form of a rPI with B4J code earlier this year that does reporting, better storage and allows for easier and more long-distance transmission of the data. The network is just reacting ... like poking a paramecium with a stick.
The neural network overlay (as I call it) is being put as an extra "layer" (hence "overlay") over the mesh network code so that instead of just simply reacting ... it'll be able to learn and better adjust each node and the entire network based on the simple Back Propagation training method. I'm hoping to achieve better utilization of each node by reducing battery usage; avoiding some redundancy in measurements; and better sensor accuracy as a starting point with the neural network overlay. Second stage is having that paramecium, itself, know what kind of stick poked it and why it was being poked.