John Bradshaw

Machine learning for molecules researcher.

[Sam Goldman @samgoldman19]

Today we release our new work on mass spectra prediction, SCARF: https://t.co/HH74QmnfT8

SCARF is a two-step generative model to predict spectra from molecules (inverse of our last paper, MIST) at the level of chemical formulae (panel C) using prefix tree data structures. 1/4

[Niki Kilbertus @k__niki]

I’m looking for motivated PhD students to join my group @helmholtz_ai (@HelmholtzMunich ) to work on causality and reliable ML systems. Please apply before November 30!

RT welcome :)

https://t.co/jM4mOGHzyx

[yingzhen @liyzhen2]

My initial tests for answering:
Are generative classifiers more robust (than discriminative ones) to recent adversarial attacks?

https://t.co/hPenJVU30d

Also proposed an idea to detect attacks with conditional generative models.

[James Hensman @jameshensman]

Probabilistic models underpin principled AI. https://t.co/ETWqNfAYwX (My first attempt at something like this, be gentle) @PROWLER_IO

[John Bradshaw @j__bradshaw]

@SebastianGP13 @alexggmatthews Hey! the pseudo points were initialised in location by k means on NN features. For small networks the NN was not pre-trained before.