The Bloom Filter Algorithm

A bloom filter is a probabilistic data structure that allows you to efficiently test an element for probable set membership. It supports the operations add(item) and contains(item), where add performs as expected and contains either yields a definite no or a probable yes, meaning false positives are possible but false negatives are not. Continue reading ⧗ 5'

RecipeNet - Training a Neural Network to Improve Recipes

This neural network can improve your cooking recipes by adding new and fitting ingredients. It's trained on the simplified-recipes-1M dataset which differentiates between 3500 different ingredients. Continue reading ⧗ 8'

Multi-label Image Classification

This is a gallery of some results in multi-label image classification I achieved last December. I used an Inception-v4 based convnet architecture which I trained for 2-3 hours on a dataset of more than 50,000 200x200px images scraped from unsplash.com. Continue reading ⧗ 7'

Spectral Clustering on Graphs

Spectral clustering is a clustering technique that can operate either on graphs or continuous data. It makes use of the eigenvectors of the laplacian- or similarity matrix of the data to find optimal cuts to separate the graph into multiple components. Continue reading ⧗ 0'

Understanding Nesterov Momentum (NAG)

Momentum and Nesterov Momentum (also called Nesterov Accelerated Gradient/NAG) are slight variations of normal gradient descent that can speed up training and improve convergence significantly. Continue reading ⧗ 4'

tqdm, imageio and Seaborn: Three essential python modules.

Learn how to use tqdm to display command line and Jupyter progress bars, imageio to easily load and save images and Seaborn to create beautiful graphs and visualizations. Continue reading ⧗ 4'

GAN-generated images from the Karolinska Face Dataset

This is a gallery of images generated by a CGAN trained on the KDEF dataset. The network consists of a standard deep-convolutional generator and discriminator with added gender and camera perspective conditional features. Continue reading ⧗ 3'

Why tf.data is much better than feed_dict and how to build a simple data pipeline in 5 minutes.

Most beginner tensorflow tutorials introduce the reader to the feed_dict method of loading data into your model where data is passed to tensorflow through the tf.Session.run() or tf.Tensor.eval() function calls. There is, however, a much better and almost easier way of doing this. Using the tf.data API you can create high-performance data pipelines in just a few lines of code. Continue reading ⧗ 8'