One from MIT - machine learning / neural network + math
https://www.medrxiv.org/content/10.1101/2020.04.03.20052084v1.full.pdf

Quote:

---

In this paper, we attempt to interpret and extrapolate from publicly available data using a mixed first-principles epidemiological equations and data-driven neural network model. Leveraging our neural network augmented model, we focus our analysis on four locales: Wuhan, Italy, South Korea and the United States of America, and compare the role played by the quarantine and isolation measures in each of these countries in controlling the effective reproduction number Rt of the virus. Our results unequivocally indicate that the countries in which rapid government interventions and strict public health measures for quarantine and isolation were implemented were successful in halting the spread of infection and prevent it from exploding exponentially

---

I for one welcome our new robot overlords.

Second up - the tsips:
https://covid-19.tacc.utexas.edu/media/filer_public/d8/c1/d8c133e3-8814-4b30-9d3f-f0992ca66886/ut_covid-19_mortality_forecasting_model.pdf

Quote:

---

we have developed an alternative curve-fitting method for forecasting COVID-19 mortality throughout the U.S. Our model is similar in spirit to the IHME model, but different in two important details.

1. For each U.S. state, we use local data from mobile-phone GPS traces made available by SafeGraph to quantify the changing impact of socialdistancing measures on "flattening the curve."

2. We reformulated the approach in a generalized linear model framework to correct a statistical flaw that leads to the underestimation of uncertainty in the IHME forecasts.

The incorporation of real-time geolocation data and several key modifications yields projections that differ noticeably from the IHME model, especially regarding uncertainty when projecting COVID-19 deaths several weeks into the future.

---

Do read the "same same, but different" section comparing and contrasting it to the IHME.

And here's their competing with IHME website
https://covid-19.tacc.utexas.edu/projections/

Thanks for posting both of those. The MIT study in particular helps show how many parameters and assumptions there are in these models - both of which are determined by background experience, first-hand knowledge, and comparative studies.

I do have a number of issues with the MIT study:

• 63 tunable parameters to force models to match real-world data...
• Little discussion for up to 3-week time-delay between contracting virus and death (14 days asymptomatic, 7days from onset of symptoms to death). This should be factored into start of quarantine and seeing actual results.
• The data to drive US prediction is based on 3 other data-sets: Wuhan (junk), S. Korea (totally different policy and approach), and Italy. All are very different
• Failure to include Sweden
  

Opinion with unnecessary adjectives, and heavily based on Wuhan data:

"Our results unequivocally indicate that the countries in which rapid government interventions and strict public health measures for quarantine and isolation were implemented were successful in halting the spread of infection and prevent it from exploding exponentially. In the case of Wuhan especially"

Opinion that seems very contradictory, although it backs up the point that quarantine isn't sustainable and delays the inevitable:

"In the case of the US, our model captures well the current infected curve growth and predicts a halting of infection spread by 20 April 2020. We further demonstrate that relaxing or reversing quarantine measures right now will lead to an exponential explosion in the infected case count, thus nullifying the role played by all measures implemented in the US since mid March 2020."

Assumptions that fail to match reality (i.e. natural social circles and culture-based social distancing that naturally occurs):

"An important assumption of the SEIR and SIR models is homogeneous mixing among the subpopulations. Therefore, they cannot account for social distancing or mass quarantine effects. Additional assumptions are uniform susceptibility and disease progress for every individual"

More models based on junk Chinese data:

"The time resolved data for the infected, Idata and recovered, Rdata case count for Covid-19 was obtained from Chinese National Health Commission"

Graduate students with backgrounds in VR, neural networks, engineering materials science:

" We are grateful to Haluk Akay, Hyungseok Kim and Wujie Wang for helpful discussions and suggestions..."

Yep and when they use trendy words like "neural network" and "machine learning", it's easy for people to buy into it and think it's legit. Those two phrases played very little into that study or the results, especially considering it was barely discussed in the paper. It's an algorithm used to refine data and adjust parameters such that models match data from Wuhan, S. Korea, and Italy. Those parameter are then applied to the US. That's it.

Everyone wants to think they are helping and contributing in some big way and claim their hero title and 15 minutes of fame. Hell I did it when I protest lockdown on Sunday. And I'm biased just like everyone else, and find data and studies that support my "hunch".

I just wish more people (especially world leaders) would admit that. Public faith would grow greatly if all bubbles, and studies supporting those bubbles, of the Venn diagram were shown with equal fervor. It would also be nice if we had some candid public discussion about what people gain by emphasizing specific Venn bubbles.

Well to be fair I think the neural network is playing a key part in their study. It's basically interpreting the difference between the analytical random mixing SIR / SEIR models and observed results.

Predictive value who knows? All they're doing is training it to reproduce an accurate curve, but neural networks are tricky because you don't really know what they're actually doing. This one brings to mind past performance is no indicator of future outcomes.