Is everything OK with GANs? — Wed links
The below tweet from Nando de Freitas of DeepMind captures a lot of my recent feelings toward AI & Deep Learning…
It is beyond any doubt that over the next few years we will perfect the technology for automatically generating a video of anyone saying anything we type, with the right voice too. What implications do you think this will have? What are the applications? How do we mitigate risks?
— Nando de Freitas (@NandoDF) March 2, 2018
The powers of ML and deep learning are simply awesome to behold:
- Witness how Apple used GANs to make images of eyes look more realistic.
- Or the ability to produce realistic synthetic time series of medical data.
- I’ve been personally tinkering with deep learning models for the last couple years, and it’s been truly wondrous to gain experience working with them.
But… the risk of using this technology for ill rather than good is upon us.
This post by Giorgio Patrini addresses the issue quite well, with the opening paragraph stating where we currently stand:
TLDR; It is becoming widely evident that technology will enable total manipulation of video and audio content, as well as its digital creation from scratch. As a consequence, the meaning of evidence and trust will be critically challenged and pillars of the modern society such as information, justice and democracy will be shaken up and go through a period of crisis. Once tools for fabrication becomes a commodity, the effects will be more dramatic than the current phenomenon of fake news.
Giorgio goes on to discuss a couple potential solutions (“digital signatures” vs “learning-based detection”), but I think the short term priority should be on just getting the word out on how immediate these threats are.
I actually heard some discussion on this topic on the Bill Simmons podcast recently, so maybe we’re already on our way to having serious public discussions about it.
The links…
1. Cool visualization of how the English alphabet has evolved…
Made this chart last year as a Kickstarter award but am now making the image available as a free download. Just right click and save and feel free to share. pic.twitter.com/4BLh0aEBsT
— Matt Baker (@usefulcharts) April 6, 2018
(key_awaytaking: The letter E is people!?! They’re people!!…)
2. Good read on the recent China / trade war news.
(via the excellent @modestproposal1)
3. Tensorflow tutorials are starting to link directly to Colab notebooks.
4. On choosing hyper-parameters…
A disciplined approach to neural network hyper-parameters: Part 1 -- learning rate, batch size, momentum, and weight decay https://t.co/wmY9RUBQpB
— Jeremy Howard (@jeremyphoward) March 28, 2018
The start of Leslie Smith's magnum opus. Lots of surprising & important results.
5. On the recent autonomous vehicle crashes…
“We should remain concerned about automated driving but terrified about conventional driving.”
(via @jdh)
Lastly, no post of links would be complete without something interesting from Colossal, like these handmade sketchbooks by José Naranja.
(colossal_understatement: follow @Colossal)