Since a large property of a neural network is indeed its initial or current random state, maybe we can consider this entire state as just a vector of randomly generated values. In reality, that’s what it is on the computer. Each of these values is generated via a random number generator in a specific sequence.Continue reading “Intelligent sequences of numbers”
Author Archives: Nicholas Komsa
Configurations of random variables
In my neural network program, I refactored some code and produced an error which I did not notice for some time. When I would run the program, eventually, out of 100 networks, a single or few networks would learn the pathing problem. With no difference in how they are trained, they are all taught exactlyContinue reading “Configurations of random variables”
Writing custom Random Number Generators and performance Comparisions
For some reason, I became concerned about the amount of time being spent by random number generators, so I took a look into them. I wrote a program which compares the speed of std::mt19937 ( mersenne twister ), a custom “twister loop”, a custom “pcg”, and custom “xorshift” generators. The standard twister was the slowest,Continue reading “Writing custom Random Number Generators and performance Comparisions”
Dumb Neural Networks
Lottery Ticket Theory says that some networks do not train well if at all, based on random properties of the network. In my current project, I have 100 agents, each with their own neural network, same shape. Each of these networks trains exactly the same way, but some end up acting smart and others dumb.Continue reading “Dumb Neural Networks”
Neural Networks and Constants
In my research, it is possible to train a network and have it learn about a constant such as PI, and put it to use in its function. However, if we pass PI in as an input, rather than having to ‘teach’ PI, then training is worlds faster. The network merely learns to use PIContinue reading “Neural Networks and Constants”
Brute Force Collision Detection with Data Oriented Design and modern C++.
I have started rewriting my game engine, and I’ve just written a de-collision algorithm for objects. It runs purely through brute-force, using no space-partitioning. In my opinion, this is a good demonstration of what’s possible through DoD and modern C++. The code for the decollision algorithm will be presented at the end, but here isContinue reading “Brute Force Collision Detection with Data Oriented Design and modern C++.”
C++ Binary Serializer
Binary Serialization can be sort of scary, maybe, to some people. Fortunately, it is super simple, straight forward, and even elegant, in C++. Binary data is the raw representation of information on the computer, and there are easy ways to work with it in C++. The benefits of using Binary are that it can saveContinue reading “C++ Binary Serializer”
Lottery Ticket Hypothesis in action
I have written an application which creates pools of 1000 neural networks. One test performs backpropagation training on them. A second test performs backpropagation, and a genetic algorithm. The amount of times training is called for each test is the same. The genetic algorithm seems to actually be able to converge on a lottery ticketContinue reading “Lottery Ticket Hypothesis in action”
My AI study so far
I have been studying neural networks for some time, and recently during a YSU hackthon, I managed to make interesting progress. After about a year long break, I return to this code and make large amounts of progress and a number of topics have presented in C++ software. I’m going to describe some of myContinue reading “My AI study so far”
Introduction to Data Oriented Design
Object oriented programming is the typical go-to method of programming, and it is super useful, especially in a program’s “sculpting”/exploration or design phase. The problem with it, however, is that modern computer hardware has evolved and typical OOP practices have not. The main issues are that hardware is getting more parallelized and cache-centric. What thisContinue reading “Introduction to Data Oriented Design”
Spot C++ 