Hi all!
My name is Adam Howes, and I’m currently a final year PhD student in Bayesian statistics at Imperial College London.
At the moment, I’m working on approximate Bayesian inference methods for the Naomi HIV model. It’s a spatio-temporal evidence synthesis model, which we currently use empirical Bayes to fit via Template Model Builder. In collaboration with Alex Stringer at Waterloo I’m trying to use adaptive Gauss-Hermite quadrature instead, via the aghq package. The motivation is to eventually get something closer to exact Bayesian inference via MCMC which is not feasible in the application setting as it takes too long.
Before that, I worked on estimating HIV risk group proportions at a district level, using a spatio-temporal multinomial model. This work is recently out, and I’ve also written a more informal blog post about it.
Last year, I visited the MIT Media Lab to work on the Nucleic Acid Observatory project, a proposal to do comprehensive, pathogen agnostic, environmental surveillance. The hope is to be able to identify novel biological threats. One way you might do this is just looking for exponential growth, and this is mostly what I worked on.
I don’t have direct experience with infectious disease forecasting or nowcasting, but looking around the forum think there is quite a lot of overlap with the methods used, as well as a shared mindset.
Looking forward, I’m thinking about what to do after the PhD! I enjoy Bayesian statistics almost in its own light, but am especially keen to work in impactful settings. (Of course a lot rests on the definition of “impactful” here!)
Nice to meet you!
5 Likes
Hi Adam,
Great to see you at the recent seminar and great to have you on here!
The motivation is to eventually get something closer to exact Bayesian inference via MCMC which is not feasible in the application setting as it takes too long.
We have had a chat about this work before and it sounds very cool. Obviously, there are lots of other settings where faster inference w/ minimal approximation cost would be very useful. For real-time work I think this is particularly the case as have to fit so many models/they need to evolve on pretty rapid deadlines.
Last year, I visited the MIT Media Lab to work on the Nucleic Acid Observatory project, a proposal to do comprehensive, pathogen agnostic, environmental surveillance. The hope is to be able to identify novel biological threats. One way you might do this is just looking for exponential growth, and this is mostly what I worked on.
We should really have someone at the seminar series to talk about efforts in this area as I agree there is a lot of overlap especially as we are all really aiming at informing/improving decisions as fast as possible. Would you be interested or have thoughts on someone we could reach out to?
Looking forward, I’m thinking about what to do after the PhD!
Anyone looking at this, @athowes is very good so get grant writing/job offering 
1 Like
For real-time work I think this is particularly the case
Yes exactly! On this topic, do you know how long the nowcasting models you have, say in this package or elsewhere, tend to take to run with Stan? And have any other inference methods besides HMC been tried?
Part of the motivation for the work I’m doing is that HMC takes too long, and the R-INLA software is not compatible with the model. For this reason, I’ve been writting something similar to INLA but built using TMB instead. That way you can have whatever model structure you would like. Looking at epinowcast.stan I think you’re going to be in a similar situation. And we heard at the seminar the other day that HMC isn’t the most friendly for users when it stochastically works or doesn’t work, and the error messages can be hard to interpret. So yes, still work in progress, but would be interesting perhaps at some point to see about trying the Epinowcast model using other inference procedures.
Would you be interested or have thoughts on someone we could reach out to?
I think I’ve been out of the loop for a while, but would think that someone like Mike McLaren or Charlie Whittaker could be up for it. I could send them an email if you’d like?
Anyone looking at this
Haha, thank you. I think unwarranted praise for the time being, but I’m keen to keep working to get better!