anishathalye

12 ms·

91.
by anishathalye 9y ago
It's a targeted adversarial example, we can do this for _any_ target class. We chose for our demo example to turn the turtle into a rifle, which was chosen uniformly at random from the 1000 ImageNet classes. In our paper, we have many
92.
by anishathalye 9y ago
Good idea! I'll add a description in a bit.
93.
by anishathalye 9y ago
Seashells isn't currently open-source (the goal was to have a hosted version to make it easy for people to pipe non-sensitive data). In the next version, I'm planning on adding end-to-end password-based encryption. Also, there are
94.
by anishathalye 9y ago
> Perhaps he means a no installation way to show htop on a web-page? Yup, exactly!
95.
by anishathalye 9y ago
A no-installation version: `htop | nc seashells.io 1337`
96.
by anishathalye 9y ago
Hi HN! I'm one of the researchers that produced this result: we figured out how to make 3D adversarial objects (currently fabricated using full-color 3D printing technology) that consistently fool neural networks in the physical world.
97.
Fooling Neural Networks in the Physical World with 3D Adversarial Objects(labsix.org)
243 points by anishathalye 9y ago|72 comments
98.
by anishathalye 9y ago
Sure! Sent you an email.
99.
by anishathalye 9y ago
They're also working on using a custom-built system written in Golang to test TiDB (though this is in the early stages): https://medium.com/@siddontang/use-chaos-to-test-the-distrib...
100.
by anishathalye 9y ago
It does take longer to synthesize robust adversarial examples - how much longer depends on the distribution of transformations. The rotation example in this blog post was pretty quick to create, maybe a minute or two on a Titan X. The print
101.
by anishathalye 9y ago
Oh, I didn't take it that way at all. In fact, I thought it was a great idea! :)
102.
by anishathalye 9y ago
Last week, I wrote a blog post ( https://blog.openai.com/robust-adversarial-inputs/ ) about how it's possible to synthesize really robust adversarial inputs for neural networks. The response was great, and I got sev
103.
A Guide to Synthesizing Adversarial Examples(anishathalye.com)
123 points by anishathalye 9y ago|22 comments
104.
by anishathalye 9y ago
How exactly would you imagine this being integrated with Slack? Would you want to share stuff from your terminal into Slack? Or you want to share content from Slack?
105.
by anishathalye 9y ago
Yeah, I totally understand you. I'm currently only using this for non-sensitive stuff (e.g. neural network training logs). I'm going to implement client-side encryption in the next version!
106.
by anishathalye 9y ago
An accounts system is coming in the next version. The public seashells URLs / HTML/CSS output ideas sounds neat, I'll add them to the potential features list! I'm not sure if the code is going to be open-sourced, but I&#
107.
by anishathalye 9y ago
In a sense, it's a no-signup-required hosted version of frontail where the client and server are separated and you don't need a server with a public IP to use it.
108.
by anishathalye 9y ago
Yeah, there are other similar services (many of which do a lot more than Seashells). The idea with Seashells was to have a zero-overhead service that was available to the public without signup (kind of in the spirit of a URL shortener: an
109.
by anishathalye 9y ago
Oh, I should update the text to clarify: sessions are garbage collected 24 hours after the program has terminated (so after there's no possibility of the program producing more output). There will probably be more options for persist
110.
by anishathalye 9y ago
Hi HN! I made a service to enable me to easily monitor long-running experiments (e.g. training neural nets). I thought it would be generally useful, so I'm opening it up to the public. I'm curious to hear what you all think! Read
111.
Show HN: Seashells – Pipe output from CLI apps to the web in real time(seashells.io)
356 points by anishathalye 9y ago|70 comments
112.
DeepSpec: The Science of Deep Specification(deepspec.org)
2 points by anishathalye 9y ago|0 comments
113.
by anishathalye 9y ago
Some people (e.g. Facebook) do this already. And it turns out that this doesn't really impact security all that much! Here's a recent research paper on the topic: pASSWORD tYPOS and How to Correct them Securely - http:/&#x2F
114.
Testing Distributed Systems for Linearizability(anishathalye.com)
1 points by anishathalye 9y ago|0 comments
115.
by anishathalye 9y ago
Thanks :) If anyone is interested in the benchmarking aspect, there's a discussion over on lobste.rs with aphyr, the creator of Knossos: https://lobste.rs/s/pdy68r/testing_distributed_systems_for Linearizabil
116.
Testing Distributed Systems for Linearizability(anishathalye.com)
7 points by anishathalye 9y ago|2 comments
117.
by anishathalye 10y ago
This is known behavior: according to the iOS 10 Security white paper [1], "iOS uses randomized Media Access Control (MAC) address when conducting Wi-Fi scans while it isn't associated with a Wi-Fi network... Note that Wi-Fi scans
118.
How do I create a topmost window? (2011)(blogs.msdn.microsoft.com)
2 points by anishathalye 10y ago|0 comments
119.
by anishathalye 10y ago
There are a ton of similar resources available here: http://dotfiles.github.io/ (reference dotfiles, frameworks, dotfile management utilities, and tutorials).
120.
by anishathalye 10y ago
Yeah, I'm lucky to have access to some really nice equipment at the Cypress Engineering Design Studio makerspace at MIT!

More ›