Talking to the incomparable @LaurelFogarty recently: LF: Do you know this Pólya book, "How to Solve It"? me: Yas queen! This book is about how to heuristically solve mathematical problems. Been a huge help to me over the years, both in research and teaching. Bits attached.

A novel method for combining multiple expert assessments with an integrated calibration of the experts (bias correction) based on training data, using Bayesian inference and hierarchical Gaussian processes. #BAjournal #AdvancePublication Jarno Vanhatalo projecteuclid.org/download/pdfvi…

Most model-based RL algos are complicated, with many moving parts and hidden design decisions. If you're new to the field, usually you discover this the hard way the first time you dig into details. (1/🧵)

KL divergence asymmetry

For some reason (because I kept hoping to write more and better someday, and someday kept not coming) I have never shared it, but here is a hands-on tutorial on #GaussianProcess in The Julia Language using Turing (https://bayes.club/@TuringLang): github.com/elizavetasemen… Big shoutout to fantastic Turing devs.

New preprint with @toxdom on "Quantifying sources of uncertainty in drug discovery predictions with probabilistic models". It discusses 7 sources of uncertainty in #MachineLearning models and code is all in The Julia Language and Turing (https://bayes.club/@TuringLang) arxiv.org/abs/2105.09474

What does a Gaussian process "look like"?🧐 My take on interactive visualisation of #GaussianProcesses aims to help you intuitively understand prior and posterior kernel functions: infinitecuriosity.org/vizgp/ - would love to hear your suggestions for improvements! Made in #svelte

covariance vs. mutual information

Dr Ellie Murray, ScD It used to be called The Inverse Method. The Bayesian epithet is fairly new. Bayes has priority over Laplace. Dale’s book is important springer.com/gp/book/978038… Andy Grieve @[email protected] will know.

New paper is online arguing that multiple corrections don't work in practice. "...p-values are not the problem, it’s the effect sizes. Adjusting p-values is like treating the symptoms of a disease instead of the cause." arxiv.org/abs/2108.04752

It's time to stop making t-SNE & UMAP plots. In a new preprint w/ Tara Chari we show that while they display some correlation with the underlying high-dimension data, they don't preserve local or global structure & are misleading. They're also arbitrary.🧵biorxiv.org/content/10.110…

A draft of the sequel, "Probabilistic Machine Learning: Advanced Topics", will be released by late Fall 2021. Together these two new books amount to about ~2000 pages of content. So my 2012 book ("Machine Learning: a Probabilistic Perspective") has finally been surpassed!

There is also lots of python code at github.com/probml/pyprobml using #sklearn, #JAX, etc. Thanks to Mahmoud Soliman, Gerardo Duran-Martin and the github community for contributing, as well as #GSoC and #Outreachy for sponsoring 5 ​summer students who also helped (see ​goo.gle/3jfd0BN).

Fresh-off-the-press in SLAS Discovery (preprint available): changepoint GPs can help reduce time, cost and ensure reproducibility of the compound-ranking procedure for novel drug modalities (don't miss the Supplement) Stanley E. Lazic Avid #DrugDiscovery pubmed.ncbi.nlm.nih.gov/34543136/

Nature does open access

Prioris.ai is hiring for a fully remote position! Open to candidates world wide, but to align time zones, a base in North America is preferred. Get in touch if you have questions or are interested. linkedin.com/jobs/view/3128…

The Hannon lab CRUK Cambridge Institute Cambridge University is recruiting a bioinformatician to apply large-scale genomics and machine learning methods to elucidate molecular mechanisms behind the piRNA pathway. Join us to discover new and exciting RNA and chromatin biology! jobs.cam.ac.uk/job/35837

✨NEW PUBLICATION✨ We identified opportunities for transboundary conservation in Africa - findings can help achieve positive outcomes for nature & people Article link: frontiersin.org/articles/10.33… UNEP-WCMC WWF US University of British Columbia University of Copenhagen Research Research summary (+📸from East Africa) below: 1/n

Here, in full directly on Twitter, is "A Hackers' Guide to Language Models". This 90 minute tutorial is designed to be the one place I point coders at when they ask "hey, tell me everything I need to know about LLMs!" It covers both OpenAI models and open source ones in depth.

A new(ish) organisation involved with research reproducibility that might be of interest: brokenscience.org I like their emphasis on prediction (and they dig Bayes...)