Machine Learning
Explore a brief overview of my projects in machine learning.
Background
Neural networks are interconnections of neurons through which information propagates to convert a set of inputs into an output.
Biological neural networks have been incredibly successful in the natural world, as shown by how they are responsible for virtually all information processing in animals (including humans). Inspired by the versatility and robustness of biological neural networks, artificial neural networks (ANNs) have become a forefront in machine learning research by using mathematical models to approximate the functioning of their biological counterparts. Despite being heavily simplified versions of biological neural networks, ANNs have brought on significant technological innovation, from beating world record GO players to powering advanced image recognition software.
Given their notable impact on research and innovation, I created several projects to apply neural networks and unpack their inner workings.
Project #1 – Single Layer Perceptron (2022)
My first ever neural-network project was a single-layer perceptron based on Rosenblatt’s original algorithm. I trained it on the classic 28×28 MNIST dataset, on which it reached 87.3% accuracy. However, as can be seen on the figure to the right, the weights were quite noisy, which signals sub-optimal learning. Additionally, the choice of algorithm constrained the neural network to only one layer. Nevertheless, it was an important learning experience regarding ANNs.
Weights created by the single-layer perceptron on the MNIST dataset
Project #2 – Hard Coded Networks (2023-2024)
Because I first started working on ANNs in 9th grade, I initially was missing much of the calculus-based knowledge needed to program effective networks. As such, in 2023 when I learnt the basics of differentiation, I worked to apply it to manually programming the backpropagation for ANNs. This first took the form of a 1×1 ANN, meaning that it had one input node, one output node, and no hidden layers. This was then expanded to hard-coded 2×2, 1×2×2, 2×2×1 etc. networks. These networks could converge, had working backpropagation, activation functions, biases, and some used Adam Optimization.
Schematics of 2 of the the hard coded networks I created
Project #2.1 – Research Paper on Network Convergence (2024)
Because I first started working on ANNs in 9th grade, I initially was missing much of the calculus-based knowledge needed to program effective networks. As such, in 2023 when I learnt the basics of differentiation, I worked to apply it to a neural network. This first took the form of a 1×1 ANN, meaning that it had one input node, one output node, and no hidden layers. This was then expanded to hard-coded 2×2, 1×2×2, 2×2×1 etc. networks. These networks effectively converged, although given their small size were not directly applicable to e.g. machine vision.
Schematics of 2 of the the hard coded networks I created
Results
So far, 5 people with CP have tested the add-on. The first tester was Levente, a 16-year-old student at a CP rehabilitation institution who is only able to use one hand at a time with limited coordination. With the help of the accessory, Levente achieved first place for the first time in a Forza race. After several hours of testing, he provided feedback for minor improvements that have since been implemented. Using the add-on, the further testers were also able to play Forza without limitations.
Figures 1, 2, 3: Pictures of testing the add-on
Conclusion
In sum, the project consists of an affordable and easily customizable 3D-printed adaptive gaming controller accessory, which allows children and adults with CP to participate in gaming. This accessory has the potential to reach a wide range of individuals and enable their engagement in gaming activities.
References
In sum, the project consists of an affordable and easily customizable 3D-printed adaptive gaming controller accessory, which allows children and adults with CP to participate in gaming. This accessory has the potential to reach a wide range of individuals and enable their engagement in gaming activities.