Unreal Engine 4: Basic Neural Network (4.12)

Marketplace Link

Video:https://www.youtube.com/watch?v=Doeh3qNTiNo

Neural Networks are a new math-based approach to simulate an actual brain. They are currently being used to generalize large datasets and categorize images. They have only recently been introduced to the field of video games. They have been proven to be an invaluable tool in bringing characters to life and giving them organic thought and movement. The network takes 1 or more float values as inputs, then returns 1 or more float values as outputs. The key is in the training; giving the network inputs and the desired correlating outputs. In this way, the network adjusts, improving how it processes the data.

The uses are only limited by your imagination:
Have NPCs wander around, reacting to walls and other stimuli
Have enemies adapt to how the player plays
Teach the player's computer allies to be more effective
Have critters learn and grow in organic animal ways

Features:
Easy to setup and use
The network can save/load itself or return a savable structure value
Self-contained; no outside libraries or resources required other than Unreal Engine
Advanced settings allow control over the learning rate and the activation function's steepness

Technical Details
Number of Blueprints: 7
5 Blueprint class
2 Structures

Features:
An easy-to-use neural network for AI, data mining, or machine learning
Takes any number of floats, processes them, and returns any number of outputs
Any size or shape: any number of inputs, outputs, and hidden layers
Can be trained to find almost any pattern, even ones that may not be obvious to humans
Supports 'supervised training'

Download Link

I love neural networks, but sometimes they can be point less

21-02-2020, 04:55 PM

unimech Wrote: I love neural networks, but sometimes they can be point less

Agree, a cool tool to play around with for proof of concept or just to learn about weighted values and training. Not always practical for game design.