I’ve been interested in helping develop a decentralized universal basic income for some time (sometimes called “CryptoUBI”). At first glance it might seem like a blockchain challenge, but the scalability demands of such a technology promptly rules out centralized ledgers.

This winter break I wanted to explore a blockchain alternative and I found Radix Engine, a pretty active project that just released a smart contract language (Scrypto) for developers to play around with.

This is a tad long so each section can be read separately and has a TLDR.

For my first deep learning project, I tried pumping raw audio into an LSTM and looked for neurons that matched orca whistles. For my next project I wanted to try using a GAN architecture and spectrograms to find stronger abstractions.

After completing a course on deep learning I was anxious to find a data set to try out for my first project. Recently Open AI demonstrated that sentiment could be detected by training a recurrent neural network (RNN) to make letter by letter predictions in a set of Amazon reviews. The amazing thing about the research was that a single “neuron” was able to predict sentiment even though the model was trained on a data set that wasn’t labeled. There’s also a great blog post from Andrej Karpathy that looks at the internals of RNNs and shows how different neurons fire character by character.