Search

12 Apr 2024: Figure 9: Results on the rotated MNIST benchmark, showing the mean $pm$ std of the test error (top) and log-likelihood (bottom) across three different seeds. ... methods. Figure 10: Results on the corrupted CIFAR-10 benchmark, showing the mean $pm$ std

12 Apr 2024: Reducing the prior precision $delta$ results in higher validation accuracy, but also a larger train-test gap, corresponding to more overfitting. ... Continual Learning: I personally think continual learning is a very good way to test approximate Bayesian

12 Apr 2024: Practitioners should also test the hell out of their inference schemes to gain confidence in them. ... The acid test is whether your inference scheme works on the real world data you care about, so test cases also need to replicate aspects of this

12 Apr 2024: Examples include inferring the mass of the Higgs boson ($X$) from collider data ($D$); estimating the prevalence of Covid 19 infections ($X$) from PCR test data ($D$); or reconstructing files ($X$) ... One way to view them is as unit tests that the

12 Apr 2024: Jekyll 2024-04-12T16:32:5900:00 https://mlg.eng.cam.ac.uk/blog/feed.xml MLG Blog Blog of the Machine Learning Group at the University of Cambridge An introduction to Flow Matching 2024-01-20T00:00:0000:00 2024-01-20T00:00:0000:00