Bryan Liu

📢 Call for abstracts 📢 for the Conference on Digital Experimentation (CODE@MIT) Submit your best work *by Sept 9*! Online field experiments, statistical methods, practical & computational challenges, etc. ide.mit.edu/events/2022-co… #mitcode

Convenience sampling (n.) - a type of non-probability sampling that involves the sample being drawn from that part of the population that is close to hand. Example: "The Met Office is lying! It is not 40C at (my local area)!"

I was wondering why someone learning hypothesis testing told me "parametric tests are tests that assumes normal data", until I did a quick search on Google. Ah, that explains. Very damaging. (Join me in reporting this to Google with the Feedback button!)

Having that said, relatively simple interventions, like a video stream, an online Q&A board, facilitating local meet-ups to watch the talks and/or socialise, all makes a difference to researchers/practitioners of different background. We can all do better. 4/4

To be clear, I am not advocating for a purely virtual conference - in-person interaction is invaluable. I also appreciate any measures to be inclusive is hard work - anyone who had hosted a hybrid meeting would be able to tell how much overhead it would incur. 3/4

Absolutely gutted (AE read: devastated) that I will be unable to attend one of my favourite conferences @kdd_news this year as the US visa appointment queue in London is 200 days long and KDD 2022 is now a purely in-person event (seems to have no video stream either). 1/4

You know you may be too into London Transport if you start spelling "modern" as "Morden". You also know statistics may have taken over your life if you start spelling "Bayswater" as "Bayeswater".

The @NationalCareers and @OfficeforAI have launched a new Data Scientist Job Profile. I’m excited that this could go some of the way to support a broader range of people to enter AI-related jobs, including apprenticeships! #NationalAIStrategy nationalcareers.service.gov.uk/job-profiles/d…

For record, the headline of the article is modified on 13 Dec (2+ week later) as "an earlier headline incorrectly implied that the study found a causal link..." (see end of article). My previous tweet still stand and I am disappointed that it takes that long to curb the harm.

Why teaching "correlation does not imply causation" is paramount. (Enter article at your own risk, you have been warned.) theguardian.com/education/2021…

Ready to roll for the #NeurIPS2021 poster presentation! Dataset and Benchmark Poster Session 4, 10 Dec (Fri) 4:30-6:00PM GMT, Spot B2. Paper: arxiv.org/abs/2111.10198 Dataset: osf.io/64jsb

Looking forward to presenting in NeurIPS next week! twitter.com/StatMLIO/statu…

To cap the week off I am presenting _another_ poster in @CODEConference now - come join me in the poster room near the left entrance! Paper: openreview.net/pdf?id=79shW3z…

There's more! We realise there aren't any datasets that can support the design and running of experiments with adaptive stopping, which is increasingly popular. So we release one of our own: youtube.com/watch?v=qJXWgp…

Online controlled experiments - talked about by many, run by many, but their results are usually not available to the many. How can we address this? Check out our teaser video - We will be presenting in @CODEConference today! youtube.com/watch?v=yooVVk…

Is it just me, or the new tube section between Kennington and Battersea Power Station seems very windily-loud?

It is hard to imagine us, as people connected to the data science "flag" with various degree of strength, moving forward together and solving problems with this mindset. Of course, I might have misheard the message at 2:30am, and am more than happy to be corrected.

As someone trained in both Maths and CS, it's shocking to hear a recognised leader implying "stop over-representing your faction's work, my faction clearly has earlier & better contributions" in a conference keynote for an interdisciplinary field.