Jon Walsh

Co-founder @unlearnai. I try to make beautiful, real things.

[Jon Walsh @AnotherJonWalsh]

@charleskfisher I think it's the salt from preservation. If you wanna get the equivalent to 1 tablespoon of olive oil from whole olives, you need to eat about 1/2 cup, which is about 1/3 of your RDA for sodium 😬

[Unlearn.AI @UnlearnAI]

🎉 Thrilled to announce that we’ve secured a $50M Series C funding round led by Altimeter, joined by returning investors @radicalvcfund, @WittingtonVC, @MubadalaCapital, @epic_ventures, and @NecessaryVC. This investment in our talented team, our data and engineering capabilities, and our longer-term R&D initiatives will help propel us into a new phase of growth and innovation as we continue advancing #AI to eliminate trial and error in #medicine. We’d also like to thank our existing investors, @insightpartners, @8vc, @DCVC, and @DCVCBio, for their support along the way. Onward! https://t.co/43qk6dK7op

[Unlearn.AI @UnlearnAI]

Not all data-driven covariate adjustment methods are created equal! Our prognostic covariate adjustment method, PROCOVA™, adjusts for AI-derived forecasts of participants’ clinical outcomes to increase the efficiency of #clinicaltrials.

[Jon Walsh @AnotherJonWalsh]

@charleskfisher @DataCooker @DataCooker Yes, we're using the NACC dataset. Do you recommend any other data that we might use?

[Jon Walsh @AnotherJonWalsh]

@charleskfisher grug brain boss see big picture

make clear goal, clear needs for tribe

track path, steer away from new shiny rock

tribe make good model, good code, give on time

simple goal, simple path, big success

[Jon Walsh @AnotherJonWalsh]

Once you can do things, what you do is one of the most important decisions you will make.

This has to be in the top 3 of advice I've ever gotten, and I think about it more and more often. What you work on is worth deep thought, even daily.

[Jon Walsh @AnotherJonWalsh]

@famulare_mike My take is that populations often have a mode where subjects are pretty similar, so ranking error tends to be higher even while absolute error is low. Unless you can predict the extremes well (which can have higher rank error), you're kind of stuck.

[Jon Walsh @AnotherJonWalsh]

@famulare_mike I can understand ranking being easier for scenario counterfactuals. In subject-level prediction, ranking algorithms don't seem to be very popular. I think they work about as well as absolute prediction but are less featured. But definitely domain-specific.

[Jon Walsh @AnotherJonWalsh]

@famulare_mike Spearman correlation is great for a metric because once you rank, each member's contribution is independent. And it's close to Pearson if you don't have a lot of outliers.

[Jon Walsh @AnotherJonWalsh]

@famulare_mike My experience has been that models usually don't agree on ranks, though I'm working on problems where there the number of things to be ranked is >100 or >1000. Especially true if you're ensembling weaker learners where you do get mileage on absolute or rank performance.

[Jon Walsh @AnotherJonWalsh]

@famulare_mike Can you explain? I don't see the tie to the law of total variance. If I have models that have similar MSE, then ones with more bias have lower variance which means better ranking. But how is learning ranks an easier problem?

[Jon Walsh @AnotherJonWalsh]

@UnibusPluram This only works if you work alone and you don't suck, make sure you tell your students that.

[Jon Walsh @AnotherJonWalsh]

@famulare_mike @cochranecollab Also what we do @UnlearnAI in clinical trials boils down to metastudies for better study design. Take data from lots of studies -> build models to predict outcomes -> apply them intelligently in trials to create better designs. Fewer patients needed, better decision making!

[Jon Walsh @AnotherJonWalsh]

@famulare_mike Got any opinion on @cochranecollab? Trying to do rigorous metastudies for better decision making on interventions.

[Jon Walsh @AnotherJonWalsh]

@charleskfisher generative AI: models that can create things as they really are

[Jon Walsh @AnotherJonWalsh]

@famulare_mike No, but did you hear they added an r to tumeric? It's turmeric now.

[Unlearn.AI @UnlearnAI]

Unlearn is hiring for a Head of Biostatistics Research and will be reviewing applications and scheduling interviews during JSM 2022. All registered applicants can apply through the JSM Career Portal, here: https://t.co/5W19JdlhnY @Amstatnews

[Jon Walsh @AnotherJonWalsh]

George Nakashima was an amazing woodworker with an equally amazing story who had a profound impact on woodworking and the American craft movement. The Conoid bench is striking.

[Jon Walsh @AnotherJonWalsh]

A project born of the right piece of wood, a "Nakashima tray". I had a small piece of live edge black acacia, and decided to make a book-matched tray with a narrow divide in the middle. The handles are maple, and there are maple feet dovetailed into the bottom. About 21"x12".

[Jon Walsh @AnotherJonWalsh]

Looking for handle ideas, I found George Nakashima designed a tray with a split middle and beautiful flared handles, so I used his design. The handles really bring it together, shaping the ends. Finished with Rubio monocoat.