We had a partial magic pony reunion tonight. Someone suggested starting another compression startup. Maybe call it mgc pny

CA

Speaking as someone who has worked on anti-spam against determined adversaries, I found this a nice, thoughtful thread on the reality of spam. Thank you @paraga.

@jack can be the middle man and @fhuszar can head the data science team 🤩 Come on all, we can do this as a team.

More scientific than whatever people are doing without access to non public information out here and 1000 times better than 💩 on Twitter. 🤣

@elonmusk @paraga Can't we just sit down sign a NDA or whatever legal agreement needed to be signed and then sample 1000 users from mDAU, then together bring a few team members from both side to go through them with all the features needed?

The right way of counting spam users out of mDAU is to sample and verify out of mDAU instead of out of followers of one or multiple account. And this is what @Twitter has been doing to the best of its abilities.

Two pictures from Miami

Postdoc on Probabilistic Machine Learning with @ZoubinGhahrama1 and me in Cambridge. If you know anyone interested, please encourage them (or yourself) to apply. https://t.co/7I1nkLd0GA

Only two photos I took in NYC 🤣

The #ECIR2022 industry poster session is very popular! Check out the two nice posters from my colleagues @Twitter 🙂 @jjh

This month I joined @TwitterResearch to work on their search and recommendation. I'm very excited to work with some amazing RecSys practitioners and apply my research knowledge in an actual real-world setting!😃 It's 20% part-time, mainly I'm still assist. prof. at Radboud Uni.

Good news at the end of the week. We are pleased to announce the reproducibility track for #RecSys2022: https://t.co/decSSWXAWO Let me repeat that... here is the call for reproducibility papers for #RecSys2022!

We describe a method for maximising long term engagements by using model-based reinforcement learning. The policy is then used to make decisions about whether to send push notifications or not. More details are in the paper if you are interested.

⚠️⛈️ #StormEunice - be prepared⛈️ ⚠️ With #Storm Eunice set to hit Surrey tomorrow, we are urging you to take extra care when out driving on the roads. ☔ Friday 0500 - 2100 Latest info 👉 https://t.co/C3ZUZHG7Ep Advice 👉 https://t.co/2BPrsOfNF3 Stay #WeatherAware⚠️

Preliminary evidence suggests new Covid variant B.1.1.529 carries higher risk of reinfection than other variants, World Health Organization says https://t.co/Nvv2xhkjmZ

Learning from Label Proportions: This line of research tries to learn from grouped training data where only aggregated group label is provided much like the recent change in Ads eco system. It also proofs that you can get good instance label prediction.

If you put it this way you see how ridiculous the labour shortage and #Brexit is.

Checkout our team's recent paper where we took a deep dive into why multi-task learning models work well for post-click conversion prediction:

Photo from a Facebook friend. Starbucks at Shenzhen Airport in China.