14 comments

[ 62.0 ms ] story [ 1048 ms ] thread
Just walked out of this talk @ Clojure West today; when the video is put online, I think it's worth your time.

("IDE for Quadcopters" would be the catchy, soundbitey name)

Video (Clojure/west): https://www.youtube.com/watch?v=53lcg7EGYM4

[Edit: Just got turned private. If/when it returns, it’ll be here: https://www.youtube.com/channel/UCaLlzGqiPE2QRj6sSOawJRg ]

This video is private. :\",
Interesting video but when Tom was demoing the system there is no display of the demo. Is that a glitch in getting the video on YouTube?
The AV system was messed up during the preso. Might be that.
The video is largely useless without the visualizations, given that the video is exclusively focused on visualization...
This is the repo of the visualization software he demo'd. Not sure how accessible it would be to run locally, but I saw the demo in person and it was pretty cool: https://github.com/dollabs/planviz
It turns out that one of the A/V guys unplugged the laptop feed 5 min in and that's why the video is just talking heads.

I'm going to redo the demo and screen capture the video and splice it in somehow.

I'll post again here with an update.

--Tom

Thanks much!
Nice. Currently working with another probabilistic modelling extension of Clojure, named Anglican.

How would this compare to Anglican? (not so much in actual state of the library now, but in goals)

http://www.robots.ox.ac.uk/~fwood/anglican/

Anglican dev here.

Systems like Anglican perform Bayesian inference in user-defined models, which are expressed as programs in a general-purpose language – in our case a subset of Clojure. An inference back end then implements a number of generic sampling-based algorithms like Metropolis-Hastings, sequential Monte Carlo and black box variational inference.

From the description, PAMELA provides a probabilistic version of a process modeling language. This is a concept that I'm personally not familiar with, but the intended use case appears to be the modeling of control systems, where inference is performed over some form of latent states in the control process:

http://www.ai.mit.edu/projects/ddamba/publications/RMPL.pdf

I'd be interested to hear from one of the devs!