Patrick Gruber




Nov 03: Deep Learning Nanodegree Program by Udacity
Deep Learning Specialization by


Dec 14: USMx - BIF003x: Statistical Analysis in Bioinformatics
Jul 31: ColumbiaX - CSMM.101x: Artificial Intelligence (AI)

Recent Work (I2A) is where caregivers and at risk people will go to see if they may have a treatable dementia or really have Alzheimer's Disease. We will have an augmented intelligence system (code has been in development for 3 years) powering the backend. Parts of the system will soon be available as open source projects. Tools include a polypharmacy engine, anticholinergenic engine, essential nutrient engine and others. There will also be a community forum.

The crypto challenge

For the final semester of my undergrad, I decided to take a computer security class. Three months into the class, the professor announced that he posted a crypto challenge, where students that were able to solve it would receive an automatic "A". We received an encrypted PDF file, and the C++ code that was used to encrypt it. This is the final report:

Complete package

The crypto challange had apparently been active for the last 12 years, and until now, nobody was able to solve it.

Exploring the Deep Learning literature

After finishing the "" certification, I began reading the paper "A simple neural network module for relational reasoning". The problem included questions such as: "What is the shape of the object that is furthest from the yellow object?". The neural network needs to learn three components; What yellow, furthest and shape means. This was the final result, after training the model for 20 hours on the AWS cloud:

Relation Network

Both the training and validation set achieved an accuracy of about 70%. Compared to the original paper, my implementation used mean to concatenate CNN output with the question. Also, the CNN configuration was different. Tweaking the overall structure could help improve the performance even further, but lack of resources prevented me from doing so.

The Google Foobar Challenge

I had heard of Google's recruiting methods before. This one however, took me a bit by surprise. One day while searching for something related to python, the Google website opened a black frame, with the text: "You speak our language". Clicking the provided link, I was brought to "". The entire website was structured, similarly to a Unix shell with restricted commands. By typing "request", a new challange was presented, and a countdown started. Challenges included algorithms for tree searches, path finding (A*), and Markov models.


There are some games hidden on this site. Type "help" and hit enter.
Open in new tab