Spherical Defense

Security using Deep Learning

[Hatice Gunes @HatijeGyunesh]

Congratulations to @Cambridge_CL @AFAR_Cambridge MPhil alumnus @akbirkhan for persevering and getting his paper *Multi-dimensional Affect in Poetry (POCA) Dataset: Acquisition, Annotation and Baseline Results* accepted ACII'21 @acii_conf! Preprint: https://t.co/ntcqPxPgm7

[Joe Brown @JoeJB94]

Super excited to be awarded runner-up best paper at #BuildSys20 for our paper SMITE. It's a real honour!

https://t.co/VuIDbaQ26h

[akbir. @akbirkhan]

Now we're back to #Kanye2020, here's why we should all be more like Ye (and some other lessons):

https://t.co/k1KvJE5u7F

[Spherical Defense @SphericalAI]

The API market is set to grow at 20% for the next decade, and with this we'll need new **comprehensive** security.

Check out our latest blog on API security at:

https://t.co/1SbbiYulsy

[akbir. @akbirkhan]

@PyTorch optimisers with memory (Adam) use a _buffer which on init is assigned to a device.

Unlike every other tensor, this doesn't have public methods for device re-assignment.

So even with the new zip save (1.4.0), you can never save/resume training completely.

😭😭😭😭

[Martin Mundt @mundt_martin]

I've written a new #machinelearning blog post on the choice of feature amounts in convolutional neural networks.

It's the start of a series aimed to challenge common DL assumptions, so I hope that it's both sufficiently insightful and still provocative:

https://t.co/sR0glSzzCu

[〈 Berger | Dillon 〉 @InertialObservr]

Prof. Claudia de Rahm's theory of 'Massive Gravity' to explain Dark Energy is getting a lot of press as of late.

So let's dig into it. (thread: 1/n)

https://t.co/EPZnTsif8G

[Quantum Computing Memes for QMA-Complete Teens @QuantumMemeing]

what if the people at IBM are right and the real quantum supremacy is the friends we made along the way?

[Spherical Defense @SphericalAI]

Testing is the important and difficult in machine learning! Here are our tips:

[akbir. @akbirkhan]

Nobody teachers you how to test Deep Learning in Academia.

I wrote some tips on how to, in this blog:

https://t.co/kItK4tJmFc

[RSAC @OneRSAC]

At #RSAC Launch Pad at RSAC 2019, a panel of leading venture capitalists will listen to pitches from startups competing for a chance to secure funding and expedite growth. Learn about the 3 finalists, @_NuID, @SphericalAI and @styrainc, who made the cut: https://t.co/RStGUl8J5V

[Spherical Defense @SphericalAI]

"Watching the world champion get beaten (4–1), woke up an entire nation to the idea that AI is finally intelligent enough to compete at human level task"... check out Research Engineers @akbirkhan's latest blog

[akbir. @akbirkhan]

I just published a new blog on Kai-Fu Lees' book AI Superpowers. I recommend it for anyone in AI Safety or Policy! https://t.co/jaAIHlWH1Z

[Spherical Defense @SphericalAI]

Check out our new weekly blog post on #ML papers our research team found interesting.
https://t.co/haXQbtWW0Q

[Spherical Defense @SphericalAI]

Check out our first blog about how #AI learns to understand language. Written by one of our research engineers, @akbirkhan :https://t.co/p7DPZ9Mnid

#MachineLeaning #nlp

[Spherical Defense @SphericalAI]

AI's learning to work together just like human players
https://t.co/SPalhbGjIr

[akbir. @akbirkhan]

The Fourier Recurrent Unit - new RNN using Fourier coefficients:
1) Trig functions = great global support,
2) Fourier Basis = easy exponential calc

FRU also has no exploding/vanishing grad. Long time dependices encoded in diff frequencies!

#icml #icml2018

[akbir. @akbirkhan]

'Visualising and Understanding Atari Agent' - using perturbation based saliency to help visualise agents intentions/value functions.

E.g Space invaders: agent firstly learns to survive then maximise points.

#icml2018 #icml #rl

[Spherical Defense @SphericalAI]

"We're gonna have a completely pressure tape."
AI Benjamin made a movie...
https://t.co/Y2fS1H5Vlw

[Spherical Defense @SphericalAI]

Interesting day at Infosec @Infosecurity today at the Olympia. Very nice and enthusiastic presentation by @cylance.

[Ian Goodfellow @goodfellow_ian]

GANs can be used to automatically design dental crowns, that are then actually manufactured and used in the physical world. Crowns need to be made specifically for each patient and need to fit correctly with the other teeth and support biting and chewing. https://t.co/ov2sQvQTjL

[Spherical Defense @SphericalAI]

It is getting harder to tell if you are on the phone with an actual human... https://t.co/GO4YayNPaZ