Tom Backert

Contact

I love to build things. Mostly Computer Science and Artificial Intelligence related.

All my projects are available on my GitHub.

MNIST: I trained a neural network on the classic MNIST Digit Recognition Problem. I also used GradCAM to visualize what features were most important for the models decision.
Brain Tumor Classifier: I built a Brain Tumor Classifier using Deep Convolutional Neural Networks (CNNs) to classify brain tumors on MRI Scans. Keras Xception was used as a base model with pretrained ImageNet weights. The base model was then finetuned for and applied on brain MRI scans via transfer learning. In the end, GradCAM was applied to visualize the models decision and therefore the tumor on the MRI scans.
- Resources: Tensorflow, Keras Xception, Xception paper, ImageNet, GradCAM
NanoGPT: I rebuilt Andrej Karpathy's NanoGPT to understand more about Large Language Models (LLMs) and the idea of transformers. It was trained on the Tiny Shakesspear dataset and was able to produce reasonable shakesspear language.
Back Propagation from scratch: I implemented the Backpropagation Algorithm from scratch to understand how neural networks actually learn. This was doing the Andrej Karpathy lecture.
Handson-ml: I have some notebooks on the Handson ML book.
Lunar Lander: I build an Reinforment Learning (RL) Agent who can land the Lunar Lander. I used the Gymnasium api and Stable Baselines3 for implementation of the DQN-Algorithm.
Ninja Game: I built a Ninja 2D Platformer game in Pygame.

The following are a collection of technical links that I've found interesting.

The following are a collection of learning resources that I've found helpful.