I wrote this program when I was 16 years old. After conducting extensive research, I implemented the Sigmoid Activation function and utilized feedforward calculations to predict output. I employed backpropagation techniques to update weights in this program, which consists of 2 layers, including one hidden layer. For the backpropagation process, I selected the longest-lived balls. The intelligence of a ball is determined by the previous generation and mutation. Random number generation is employed for mutation purposes.
The program includes a mutation-dependent feature where the survival of a species is predominantly influenced by mutation and mitosis. Each individual undergoes microevolution through an infinite series of mutations, aiming to reach the perfect gene. However, this process lacks genetic diversity, resulting in individuals with generally lower intelligence. Mutations can have both negative and positive effects, making it uncertain whether the next generation will surpass the previous one. Nevertheless, over time, a generation best suited to the environment will emerge, thanks to the workings of natural selection. The program's purpose is to enable the next generation to learn from its predecessors. Mutations will occur in the best-performing trait found in the individual that has progressed the furthest
What happens when there are multiple individuals with unique traits? Crossover occurs, resulting in offspring that inherit traits from both parents. Mutation continues to happen but on a larger scale. As a result, individuals are generally smarter and more diverse, leading to a more stable population. They are able to learn faster, exhibit greater intelligence, and require fewer generations to adapt.
Observing the simulation, after several runs, the balls (referring to the simulated entities) demonstrate survival skills within approximately 3-4 generations, which is significantly faster than the first program (which could take anywhere between 1-200 runs due to the unpredictable nature of mutation). The balls exhibit a combination of their parents' appearances, with the color being averaged between the dad and mom. While I didn't fully grasp neural networks, I was determined to complete what I had started. Motivated by this desire, I began coding from scratch and completed the task within less than 24 hours. The lesson learned here is to not overlook the basics, as they form the foundation for more advanced concepts.
Genetic variation is incredibly crucial. It serves as the primary catalyst for a stable environment and growth. Let us strive to conserve species and protect the environment. We are currently facing the sixth mass extinction, and this time, humanity itself is the meteor.
This program is written in java (It is a _very_ simple program with a simple neural network. Brains except uno input and decide whether to jump or not. Brains will accept more inputs --up to 3-- soon). Keep in mind that not everyone is born stupid; hence, you may need to re-run multiple times (possibility of a smart individual is 3/10, and possibility of all-are-smart is 0.027).