Brain Simulator 🧠:

Neural Network From Scratch using Java and processing environment. Brain is a neural network simulator tool implemented in Java using the Processing environment. The project aims to provide a visual representation of the decision-making process of a neural network controlling the movements of a snake. The tool incorporates a genetic algorithm to optimize the neural network's performance. Through real-time visualization, users can gain insights into the neural network's behavior, identify patterns, and witness the improvement achieved through the genetic algorithm.

Problem Statement:

The objective of Brain is to develop a snake brain simulation tool that showcases the decision-making process of a neural network (how the brain of the snake work) controlling the snake's movements. The neural network is optimized using a genetic algorithm, in addition to being designed from scratch and implemented in Java using the Processing environment. The tool provides a real-time or iterative visualization of the network's decision-making during brain activity.

Brain_Simulator.2023-12-11.10-31-20.online-video-cutter.com._3.mp4

Neural Network Architecture:

The neural network architecture for Brain is as follows:

Input Layer: The input layer consists of 24 nodes representing distances from the snake's body, walls, and food in eight directions. Each distance is encoded as a floating-point value
Hidden Layers: Brain includes two hidden layers, with each layer comprising 18 nodes. The Rectified Linear Unit (ReLU) activation function is used for the hidden layers
Output Layer: The output layer consists of four nodes representing the possible directions that the snake can move: up, down, left, and right. Each output node produces a value between 0 and 1, indicating the probability of selecting the corresponding direction.

Activation Function used:

ReLU function:

float reLU_Function(float y){
        return max(0,y) ;
}

Genetic Algorithm:

Brain incorporates a genetic algorithm to optimize the neural network's performance. The genetic algorithm involves the following steps:

Initialization: A population of neural networks is randomly generated, each with its own set of weights.
Evaluation: Each neural network in the population is evaluated by playing multiple snake games and measuring its fitness based on performance metrics such as survival time or score.
Selection: Neural networks with higher fitness are more likely to be selected for reproduction. Selection methods such as tournament selection or roulette wheel selection can be used.
Reproduction: Selected neural networks undergo crossover and mutation operations to create new offspring. Crossover combines the genetic material of two parent networks, while mutation introduces random changes to the offspring's weights.
Replacement: The offspring replace a portion of the previous population, ensuring the population size remains constant
Iteration: Steps 2 to 5 are repeated for multiple generations, allowing the population to evolve over time

Simulation Features:

Brain offers the following visualization features:

Overlay Simulation: The simulation overlay is presented on the snake's brain structure, providing a clear representation of the neural network's position, food cues, and other relevant information.
Activated Neurons: Activated neurons are highlighted during the decision-making process. This visual distinction can be achieved through color or size variations, making it easier to observe the active parts of the neural network.
Connection Visualization: Connections between neurons, representing the weights of the neural network, are visually displayed. This visualization may utilize lines, arrows, or other graphical elements to represent the strength and direction of the connections.
Real-time or Iterative Visualization: The visualization updates in real-time or at regular intervals to depict the decision-making process of the neural network as the snake moves and interacts with its environment.
Genetic Algorithm Metrics: The tool displays relevant metrics related to the genetic algorithm, such as the current generation, best fitness value, average fitness value, and other statistics to track the progress of the optimization process.

For more, check my article in Medium:

https://medium.com/@Esmail_A.Gumaan/brain-simulator-355dd3767e00

Esmail-ibraheem / Brain-Simulator