Chris (@canagnos) will be joined by Luciano Floridi (@Floridi), University of Oxford, Carly Kind (@carlykind_), Ada Lovelace Institute, John Edwards, ICO and Sue Daley (@ChannelSwimSue), techUK.

Clarification: the marginal correlation between estimates of CFR and R0 induced by undetected cases is negative, not positive as my poor grammar suggested above. Put simply, it is either milder and more contagious, or deadlier and less contagious.

Below is a quick thread on why a short, sharp, sheer shutdown of 4-5 weeks will be effective and desirable, based on my conversations with @yaneerbaryam and others over the past weeks. First, why are ~4 weeks enough to do the trick? 1/n

As a result, Ioannidis' points are absolutely critical to think about and address in the mid-term, but I don't see how they change the picture in the immediate term. end/

But it is clear that deaths are still on an exponential track in Italy, which means they are not yet approaching immunity. As a result, what happened in Italy will likely happen elsewhere assuming no early lockdown and similar national health system capacities. 6/

Of course this correlation breaks down for longer-term forecasts, because a bigger R0/lower CFR means the exponential phase of the logistic curve runs out of steam faster (as we approach herd immunity faster). 5/

This correlation means we are much less uncertain about the number of severe cases coming our way than we are about either R0 or CFR: our uncertainty about either quantity "cancels out" in the case of forecasting number of severe cases and deaths. 4/

Moreover, the testing sampling bias renders our estimates of R0 and CFR *positively* correlated: if we have tons of undetected cases, these must be mild which brings CFR down, but it also means the virus is more contagious, so it increases R0. 3/

And yet in places like Italy we have clear evidence of exponential growth not only in cases, but also deaths, which is a more reliable statistic in that it is much more independent of testing (all severe cases are tested). 2/

Professor Ioannidis, an eminent scientist and role model for younger statisticians like myself, expresses concern about estimates of R0 and CFR, and claims that the virus might be less contagious and less fatal than has been claimed. Source here: https://t.co/k3YS0s37us 1/

A new challenge for self piloting deep AI.

You know what would be a great idea - a global climate change mitigation force. Countries would contribute according to their GDP. There are a lot of issues out there in places that don’t fall neatly in any one country’s territory.

It's time. Join Andrew in his support of the full federal legalization of marijuana. https://t.co/kvc553RELT

Proud to release my climate change plan today - we must fight like mad for a future we will be glad to pass to our children. #YangsClimatePlan 👍🇺🇸 https://t.co/tliM27XEqd

Only 17% of Americans approve of the way our Congress is handling its job. We need to change this.

I anticipate 2019 to bring more attention to alternative forms of supervision in machine learning, in an attempt to overcome label scarcity and label noise. Cybersecurity is among the fields likely to benefit most as I argue in this book chapter:

Roll three 6-sided dice. What is the chance that at least two of them will sum exactly 10? I have data that suggest that even professional statisticians get this wrong more than half the time. What is your take?

Our open-source toolbox for anomaly detection, DSIO, reached 300 stars! We are working hard to create synergies - read here about our tight integration with scikit-learn “ https://t.co/Fka9wCsQso scikit-learn integration”

Petition: Debate in Parliament the lack of an effective policy for the treatment of M.E.

More and more incredibly brave Iranian women are risking imprisonment by defying the law to wear mandatory hejab. They deserve international solidarity.