I released the online version of my number puzzle game Summit today, free to play in browser. Aimed at mathsy types with a few minutes to kill! .

I have been developing a free mobile number puzzle game, and excitingly its now in Early Access on Google Play. If you own an Android, like number games (sudoku, nerdle etc.), and would like to test the game, please get in contact. See the video here for more details and a demo!.

For fans of alt rock, Radiohead etc, some new moody catharsis from my band.

The PROV ontology can be used to model provenance information for data generated from multiple entities and processes. But what if my workflow contains a feedback loop? I.e recursively generating and updating data in light of new information. Is there a way to model this?.

#wordle normal distribution status: complete ✅

RT @PessoaBrain: Problem to solve:.r(a,b)=x; r(b,c)=x; r(a,c)=0, where r is correlation. How large can x be??.

These results highlight the importance of replication/reproduction and the need for further work towards pipeline harmonization. Group-level maps for all workflows shared on Neurovault, scripts shared on Github/OSF. See the manuscript for full details and full results! 5/5.

But we also identify analysis step manipulations that cause minimal differences in the final results. E.g., group-level results were largely unaffected by changes in the software used to model the low-frequency fMRI drifts. 4/5

In this new work, we find that the analysis step manipulation that matters most varies by dataset. E.g. for the BART task, the packages’ different 1st-level signal models induced sizable variability between the final group-level t-stat maps. 3/5

Follows on our previous research (Bowring et al., 2019, HBM), where we discovered sizable inter-software differences when reproducing the same 3 studies’ results in AFNI, FSL + SPM (Dice commonly < 0.5, Correlations between unthresholded t-stat maps commonly < 0.6). 2/5

Isolating the Sources of Pipeline-Variability in Group-Level Task-fMRI results - now out in HBM! w/ @ten_photos & @cmaumet.Reanalyzing 3 task-fMRI studies with workflows that interchange between AFNI, FSL and SPM at different stages of the pipeline. 1/5

Today was my last day in academia. I'm very excited to be joining @smithinst as a mathematical consultant on 22nd November. As an RA, PhD student and postdoc I had so many good times I will remember, thank you to all friends and collaborators that have helped me along the way!.

RT @ten_photos: A new approach for voxelwise modelling longitudinal brain lesion data: Relative risk regression with GEE, applied to white….

RT @fMRI_today: My section on Functional Imaging Methods at the NIH is looking for a post doc to do research advancing fMRI - anything incl….

Really thoughtful and elegant new paper about resting-state models for hypothesis testing, which we were lucky enough to have @R_Liegeois present to our group yesterday.

RT @Rainmaker1973: Anscombe's Quartet, the Datasaurus and the Datasaurus Dozen or why it can be difficult to demonstrate the importance of….

RT @cmaumet: New paper! Built with OHBM Open Science community @OhbmOpen - this one is very close to my heart 💞. We review existing barrier….

RT @nomadj1s: A simple, *hypothetical* example of the Base Rate Fallacy:. Suppose w/o a vaccine, 9/10 get hospitalized. Suppose w/ a vaccin….

These results highlight the importance of replication/reproduction and the need for further work towards pipeline harmonization. Group-level maps for all workflows shared on Neurovault, scripts shared on Github/OSF. See the manuscript for full details and full results! 5/5.

But we also identify analysis step manipulations that cause minimal differences in the final results. E.g., group-level results were largely unaffected by changes in the software used to model the low-frequency fMRI drifts. 4/5