Neural Network

May

A few weeks ago I completed the 16 week "Neural Networks for Machine Learning" on-line course from University of Montreal to deepen my understanding of Artificial Neural Networks and their fascinating properties. The course is taught by Professor, Geoffrey Hinton, which is one of the grand old men in ...

May

I was recently contacted by Gaurav Chaudhary, who is the CEO of Roots Analysis,  a firm that provide market research and consulting for the pharmaceutical industry. It turns out that Wildcard Pharmaceutical Consulting was mentioned in their report: Deep Learning in Drug Discovery and Diagnostics, 2017 ...

Mar

The process of expanding an otherwise limited dataset in order to more efficiently train a neural network is known as Data Augmentation For images there have been used a variety of techniques, such as flipping, rotation, sub-segmenting and cropping, zooming. The mirror image of a cat is ...

Mar

Deep Neural Networks and Spectroscopic Data - a chemometric study A Masters project is available in a collaboration between Wildcard Pharmaceutical Consulting and Department of Food, Spectroscopy and Chemometrics, KU. We are looking for a motivated student with interest in spectroscopy and computer programming to take on a project involving ...

Feb

In another blog post I demonstrated how to build a deep neural network with Keras in Python to model some toxicity dataset from the Tox21 challenge. The network was however not systematically optimized, but merely put together with a few manually selected hyper parameters. Hyper parameters are understood as the parameters which are not tuned by the data ...

Jan

I found some interesting toxicology datasets from the Tox21 challenge, and wanted to see if it was possible to build a toxicology predictor using a deep neural network. I don't know how many layers a neural network actually has to have to be called "deep", but its a buzz word, so ...

Jan

In the last blogpost the battle tested principal components analysis (PCA) was used as a dimensionality reduction tool. This time we'll take a deeper look into chemical space by using a deep learning neural autoencoder, by testing some of the newer tools based on neural networks which has shown promising results. ...

Dec

Neural networks are interesting models underlying much of the newest AI applications and algorithms. Recent advances in training algorithms and GPU enabled code together with publicly available highly efficient libraries such as Google's Tensorflow or Theano makes them highly interesting for modelling molecular data. Here I explore the high level Neural Network ...

Nov

Neural Networks are interesting algorithms, but sometimes also a bit spooky. In this blog post I explore the possibilities for teaching the neural networks to generate completely novel drug like molecules. I have experimented for some time with recurrent neural networks with the LSTM architecture. In short, recurrent neural networks differ from more traditional feed forward neural networks because they do ...