Is Bitcoin in a Bubble?

It is closer to the moon than 2017, so are we headed for another crash? Today is different. There is a huge price rise but not the same level of retail interest.

The Straussian Moment - Peter Thiel (pdf)

The Straussian Moment - Peter Thiel (pdf) This was strangely hard to find and only available as pictures on Evernote, so I uploaded it as a searchable PDF. Peter Robinson also has a great interview with Thiel on his essay.

How Economic Development Happens - Synthesis of Competition and Command

I wondered about economic development since I was a kid, but the answers in economics were always unsatisfying. 18 year old me went to economics class to discover the mysteries of the world but was confronted with the Solow growth model.

Money Multiplier is Bullshit

Undergraduate economics commonly teaches that banks take in deposits and lend these out, increasing the money supply. This is wrong. The incorrect depiction of a fractional reserve banking system has caused widespread confusion, with some people thinking of it as a strange conspiracy, and is a staple of paranoid ramblings on Facebook.

Keeping up with Deep Learning Research

When I started my PhD I was overwhelmed trying to keep up with new research. Trying to read everything will take forever, but by not reading enough we risk rediscovering old ideas, or doing unimportant work.

Open Research 8: Optimizing Affine Transforms in VAEs

Optimizing Affine Transforms in VAEs Our goal is to create a VAE which will only encode a subset of affine transforms, but generalize to all. This is shown below: Previously we showed takes a higher capacity VAE to encode all possible orientations of an object, compared to encoding a single orientation.

Open Research 7: Differentiable Affine Transforms

Differentiable 3D Affine Transforms Affine transforms were made popular for deep learning in Spatial Transformer Networks, where they introduced an affine transform layer with to transform an image to some cannonical orientation before it was processed by the network.

Open Research 6: Why Deep Learning Research Doesn't Matter

Why Deep Learning Research Doesn’t Matter I started a PhD in deep learning because it was a fast moving field that was having growing impact on the world, and dreams of even greater developments.

Open Research 5: 3D Disentangled Representations and Iteration Speed

3D Disentangled Representations Last week I started my work on this by implementing a 3d VAE, and now it’s time to do something with it. The goal is to disentangle the orientation of an object from its shape, by optimizing over the orientation during training.

Open Research 4: Abandoning Weight Sharing, Bitter Lessons and Reasonable Ambition

Disentangled Representations, Bitter Lessons, and the Edge of Ability For a long time I’ve been obsessed with the idea of learning disentangled representations of the world. The general problem of learning disentangled representations is impossible without assumptions, just like how there is no classsifier to rule them all.