Fun fact: More data can hurt in linear regression. arxiv.org/abs/1912.07242 (aside: this is about the 3rd time that studying deep learning has taught me something about linear regression).

It took me a while to recap bc some papers are complex. Key points: - Destructing black box: convergence, generalization, neural tangent kernel - New approaches: Bayesian & uncertainty, graph, convex optimization - Neuroscience & bio-inspired algos - ... huyenchip.com/2019/12/18/key…

wired.com/story/facebook…

Here's the opening lecture I gave on recent developments in deep learning and AI, and hopes for 2020. It's humbling beyond words to have the opportunity to lecture at MIT and to be part of a community that seeks to understand and engineer intelligence: youtube.com/watch?v=0VH1Li…

Dismissing machine learning because it can't make sense of what it *hasn't* seen before it quite short-sighted. It is immensely valuable to be able to automatically recognize *what you are able to label* -- especially on hard pattern recognition problems, at super-human accuracy.

Humans would likely say image a) and c) of each row are from the same category, whereas b) is from a different one. On the contrary, a deep net robust to perturbations that turned a) into c) would assign the exact same category to all 3 images per row because: ||a-b|| = ||a-c||

1/ I interpret Hinton's question to be: If you had access to either a rigorously* validated machine learning surgeon w/a 90% cure rate that couldn't explain its decision or the opinion of a human surgeon w/ an 80% cure rate who could offer an explanation, which would you prefer?

Short and sweet, with some good insights on autonomous driving. kcbsradio.radio.com/media/audio-ch…

You now can train a really good image classification model on 25k images in less than 2 hours, entirely for free, on a Colab GPU. And it's dead easy. What a time to be alive I remember when you had to write your own CUDA and C++ and you needed a hardware budget

Sam Quinn did a great job explaining his process behind the OSINT and technical security analysis of the iParcelBox here: youtube.com/watch?v=t3Y37F…

I have a Deepfake joke but it’s hard to detect

"McAfee ATR did a fantastic job putting together this attack chain and finding a 0-day in the core Linux kernel itself to make it work." Team effort, 0day kudos to PL 🤘🦇🤘 with research team Sam Quinn Eoin Carroll Douglas McKee Mark Bereza 🇺🇦! techcommunity.microsoft.com/t5/internet-of… #mcafee #atr #assrc

Me, Douglas McKee, and Steve Povolny (he/him) will be streaming here tomorrow at 9amPT to discuss the 4 vulns be found in a teleconference robot it is 100% private so please no one join. it would be a real shame if some 𝑏𝑎𝑑 𝑎𝑐𝑡𝑜𝑟𝑠 listened in on our private meeting

This is a really cool project that needs to be seen! Submit your DeepFakes for analysis, and look for an upcoming demo on the detection framework! @purkinje16 Sherin Mathews Shivangee

We have a great opportunity for a researcher to join the ATR team here McAfee Labs careers.mcafee.com/job/netherland… #infosec #infosecjobs #Malware

I always get Normalization and Standardization mixed up. But they are different. Notes about them and why do we care. 🧵👇

Machine Learning concepts as animal GIFs A thread 🧵

Some takeaways from @openai's impressive recent progress, including GPT-3, CLIP, and DALL·E: [THREAD] 👇1/

I'm willing to fund hospitals in India that need funding to import bulk planeloads of oxygen or supplies into India to increase supply. Public hospitals/NGO's also pls reach out PMO India Ministry of Health The Times Of India Congress #IndiaFightsCOVID19 Hindustan Times The Indian Express Give

As I struggled to deal with the impact of COVID on my family members in India, I got delayed by a day for submitting my reviews for a conference & I got a message from a senior reviewer with the blurb below. My humble request to everyone - pls don't say this to anyone ever! [1/n]