Posts & notes
Original projects, plus 15 posts marked Course notes that summarize lab and tutorial work from courses at TU Braunschweig. Each of those credits its course at the end.
Building an Interactive Family Tree
How I built a bilingual, offline-capable interactive family tree application.
Custom Object Detection with YOLOv5
Training YOLOv5 on the DIOR remote sensing dataset for aerial image object detection.
Numerical Analysis Approaches in Engineering
Data-driven numerical methods for solving complex engineering problems with practical examples.
Predicting Battery Life with LSTMs
NASA battery dataset, capacity degradation analysis, and stateful LSTM architectures for State of Health prediction.
The Complete Guide to Handling Missing Data
Detection, visualization with missingno, deletion strategies, imputation methods, KNN imputation, and the EM algorithm.
Kaggle House Prices: A Complete EDA
79-feature dataset analysis with ANOVA testing, Spearman correlation, feature engineering, and Ridge regression.
The Art of Brewing Excellent Coffee
From Turkish coffee to V60 pour-over: brewing methods and what makes a truly great cup.
From Feature Functions to Neural Networks
OOP feature function classes, CO2 prediction, and composing single and two-hidden-layer networks from scratch.
Introduction to Bayesian Regression
Gaussian priors, likelihood functions, closed-form posterior computation, and predictive distributions with uncertainty bands.
Explainable AI: Grad-CAM Visualizations
Visualizing what a CNN sees when classifying aerial imagery: opening the black box of deep learning.
Semantic Segmentation with U-Net
Pixel-wise segmentation on ISPRS Vaihingen aerial imagery with custom DataGenerator and IoU evaluation.
DeepLabV3+ vs U-Net: Advanced Segmentation
Comparing architectures with data augmentation and focal loss on the ISPRS Vaihingen benchmark.
Single-Class Object Localization with VGG16
Bounding box regression using a frozen VGG16 backbone for airplane localization on Caltech101.
Multi-Class Object Detection
Dual-head CNN for simultaneous classification and bounding box regression on Caltech101.
CNN Training: Optimizers, Initializers & Augmentation
Systematic comparison of optimizers, weight initializers, and data augmentation on UC Merced land use.
Transfer Learning with VGG16
Frozen feature extraction vs fine-tuning for aerial land use classification on UC Merced.
Polynomial Regression, Overfitting, and Regularization
Polynomial degree sweep, Ridge and Lasso regularization, ReLU/sine/Gaussian feature functions, and Mauna Loa CO2 data.
MLP for Image Classification
Multi-Layer Perceptron for Fashion MNIST: overfitting, data generators, and spatial invariance limits.
CNNs for Image Classification
Vanilla CNN architectures on Fashion MNIST: effects of padding, depth, and kernel size on generalization.
Introduction to Data Exploration with Python
Loading the sklearn diabetes dataset, DataFrame basics, histograms, scatter plots with colormaps, and pairwise correlation grids.
Linear Regression and Curve Fitting from Scratch
The 4-step ML framework: features, predict, loss, optimize. From sklearn to Vandermonde matrices and scipy optimization.
Working with Remote Sensing Data
Spectral band visualization, pansharpening, and NDVI vegetation index computation from satellite imagery.
No posts in this category yet.