Profile cover photo
Profile photo
Dan Brickley
2,375 followers -
I work on Web standards, mostly. SemWeb, W3C, FOAF, Schema.org, NoTube.
I work on Web standards, mostly. SemWeb, W3C, FOAF, Schema.org, NoTube.

2,375 followers
About
Communities and Collections
View all
Posts

Post has attachment

Post has shared content

Post has shared content
Open source TensorFlow models. Josh Gordon at Google I/O demonstrates using Inception, a predefined model for trained from ImageNet, for image classification. It is very good at recognizing dogs. If you have a classifier, you can use it to generate art. E.g. Deep Dream. He shows how layers in the network learn edges and colors, textures, and higher-level patterns. Deep Dream is made by progressively exciting a particular filter somewhere in the network. He shows style transfer, with the edges, textures, and patterns layers coming from the style image, and the objects and predictions layers coming from the content image.

His next demo is Parsysaurus, the most accurate natural language parser. Given "The gostak distims the doshes", it can tell you "distims" is a verb, even though it's not a real word, present tense, that "gostak" is singular, and the subject of the sentence, "doshes" is plural and the object, etc. it supports 40 languages.

At the end he suggests a number of educational resources for neural networks.

Post has shared content
CycleGAN (https://junyanz.github.io/CycleGAN/) can translate between horses and zebras. It can also be used to make videos. Can't wait to see what movie makers will make of this technology.
Animated Photo

Post has shared content
Synthetic biology

I met some cool people this week, and here's one: Kate Adamala. She's a postdoc at the University of Minnesota. She creates artificial cells in the lab.

These aren't full-fledged cells that can reproduce and metabolize on their own. They're much simpler - but they're made from scratch, not from existing cells. She calls them protocells.

A typical protocell has some RNA inside a little membrane made of fatty acids. She can get the RNA to copy itself, and she can get different protocells to fuse, building more complicated systems from smaller parts.

My own career as a postdoc was much more boring! Kids these days are amazing. :-)

Here's a paper of hers:

• Kate Adamala and J.W. Szostak, Non-enzymatic template-directed RNA synthesis inside model protocells, Science 342 (2013) 1098-1100. Available at http://www.protobiology.org/reprints/Adamala_Szostak_Science_2013.pdf

Abstract. Efforts to recreate a prebiotically plausible protocell, in which RNA replication occurs within a fatty acid vesicle, have been stalled by the destabilizing effect of Mg2+ [magnesium ions] on fatty acid membranes. Here we report that the presence of citrate protects fatty acid membranes from the disruptive effects of high Mg2+ ion concentrations while allowing RNA copying to proceed, while also protecting single-stranded RNA from Mg2+-catalyzed degradation. This combination of properties has allowed us to demonstrate the chemical copying of RNA templates inside fatty acid vesicles,
which in turn allows for an increase in copying efficiency by bathing the vesicles in a continuously refreshed solution of activated nucleotides.

Though it's one of the most recent on her website, this paper is not so new; she's doing even cooler stuff these days. Check out her work here:

http://www.protobiology.org/index.php

and in her talk.

Post has attachment

Post has shared content

Post has attachment

Post has attachment

Post has attachment
Wait while more posts are being loaded