Deciding how many nodes and layers to use in a neutral network: many functions can be expressed in 1 or 2 hidden layers, more is often better and faster, and you usually have to experiment.
https://machinelearningmastery.com/how-to-configure-the-number-of-layers-and-nodes-in-a-neural-network/
miniblog.
Generating photos of fictional people using generative adversarial networks: https://thispersondoesnotexist.com/
Really interesting paper exploring adversarial inputs to ML models: https://arxiv.org/abs/1905.02175
They conclude:
* It's a property of the input data, not the training
* You can even train a model on non-robust features and obtain a model that works well on the original input data!
Exploring garbage collection accelerators in CPUs: https://spectrum.ieee.org/tech-talk/computing/hardware/this-little-device-relieves-a-cpu-from-its-garbage-collection-duties
A story of exploiting Android through releasing packages with the same name and ID on public repositories: https://blog.autsoft.hu/a-confusing-dependency/
The amount of TV people watch is determined more by genetics than environment: https://www.rationaloptimist.com/blog/the-genes-of-human-behaviour/
AV1 files are roughly 1/30th of the size of equivalent GIFs! https://www.singhkays.com/blog/its-time-replace-gifs-with-av1-video/
IFTTT is very limited compared to programming, but it's super convenient for adding functionality to a smartphone.
For example, I have a task that unmutes my phone if I miss a call. This is really handy functionality for me, and doesn't exist in the phone app itself.
For long term storage, the US Library of Congress recommends CSV, JSON, XML and... sqlite! https://www.sqlite.org/locrsf.html
I'm surprised since it's a more complex format tied to a single implementation. Extremely widespread though.
Woah, rather than selling security exploits, there's now a black market selling access to compromised corporate networks! https://arstechnica.com/information-technology/2019/05/hackers-breached-3-us-antivirus-companies-researchers-reveal/
I'm a big believer in 'Perlis languages', where you learn a language for new perspectives and approaches rather than needing it for a specific domain.
The tricky bit is: when can you say you've acquired those new perspectives?
Extraterm is a terminal emulator with a really interesting model of showing outputs in frames and allowing you to manipulate them: https://github.com/sedwards2009/extraterm
Cute (video) introduction to exceptions in Smalltalk. You can handle, resume, return alternative values or even re-execute in the same context!
https://youtu.be/x8q_04lrsDM
I'm really excited to learn about Luakit, a browser that's extensible in Lua! https://luakit.github.io/
I've followed uzbl and Conkeror with interest but both are inactive these days.
fping is a handy ping alternative, that allows specifying multiple hosts and other handy flags like timeout! https://fping.sourceforge.net/
The Stack Overflow/Discourse model is that you get more privileges as you're more active in a project.
Could we do the same with git repos? Newbies get more guidance and guard rails, seasoned contributors get automatic merge powers if CI passes?
Pharo 7 includes a tiling window manager! https://pharoweekly.wordpress.com/2018/07/09/tiling-shortcuts/
On the early metaphors that drove the original development of Unix: https://zge.us.to/txt/unix-harmful.html
(Lots of typewriter influence, and amazingly the first versions did not have a notion of pipes!)
gcc 9.1: better diagnostics, better optimisations (faster to optimise and faster at runtime), a new D frontend, and more! https://gcc.gnu.org/ml/gcc/2019-05/msg00024.html
A neat demo of the importance of good tracebacks in asynchronous code: https://vorpus.org/blog/beautiful-tracebacks-in-trio-v070/
Showing 461-480 of 736 posts