iiinova — where your digital self keeps evolving

Gustav Markkula @markkula.bsky.social · 14/08/2026

I love the little art surprises you get sometimes when plotting 2D data

A 2D histogram / heat map making accidental abstract art

Gustav Markkula @markkula.bsky.social · 29/06/2026

Dr. Xian Liu (not on Bluesky) has led this work, showing that the conflicts generated by a state of the art ML traffic simulation model could be analysed using Extreme Value Theory to accurately predict historical crash rates at five intersections in Leeds.

Example conflict (time to collision) heatmap from a conventional rule-based model and the machine-learned model, showing a more stereotyped and sparse conflict distribution for the rule-based model.

100

Gustav Markkula @markkula.bsky.social · 18/11/2025

🥳 Cool new paper by Siwei Ma, where evidence acc models and EEG data tell the same story: the effects of being cognitively loaded while deciding whether to cross the street can be understood as making us less sensitive to the available perceptual evidence 🧪🧠📈 www.sciencedirect.com/science/arti...

Stimulus and response locked centroparietal EEG are both flatter when under cognitive load than otherwise, this includes the ”CPP” style peak at response

061

Gustav Markkula @markkula.bsky.social · 19/05/2025

Great post, and can definitely be read as film noir voiceover in places

Film noir detective looking out a window, and a screenshot from the linked post: “Moving forward, I would have to be more skeptical, even (or perhaps especially) of high-impact papers with impressive results.”

020

Gustav Markkula @markkula.bsky.social · 11/05/2025

CleanRL was a revelation for me, and I noticed that this paper (which Daphne pointed me to 😄) lauded the single file approach in general. It’s an interesting question I think - readability of a finished research product vs modularity and robustness of a codebase under dev? arxiv.org/abs/2502.20349

R3. Adopt the “single file” philosophy. Ideally, key workflows (e.g. model or evaluation definititions) should be defined in a single file. This has emerged as a powerful force for reproducibility in ML research (see Table 1 for examples). While this approach contradicts traditional software engineering principles of modularity and abstraction, it offers distinct advantages for research code: (1) complete understanding of an important component (e.g. how a model works) requires reading only one file, (2) minimal abstraction layers and dependencies, and (3) easy adaptation of either complete implementations or specific components. One can simply copy this single file into another codebase. Removing friction from future reuse also benefits us—as we can see from Table 1, these methods get a large number of citations.

230

Gustav Markkula @markkula.bsky.social · 30/04/2025

Amazing days at Champalimaud Neuroscience @champalimaudr.bsky.social in Lisbon, discussing how animals, humans, and human road users do sensorimotor control. Super smart super friendly people, and my reading list has exploded. 😅Thanks so much for the invite Alfonso Renart and colleagues!

Champalimaud Foundation, right next to the sea

Me in front of the Champalimaud building, complete with indoors palm trees and midair tubular glass corridor

A custard pasteis, perfectly flaky and still warm from the oven, and a cafe pingado

230

Gustav Markkula @markkula.bsky.social · 02/02/2025

A recipe for pain-free(?) academic writing gmarkkula.github.io/writing-reci... I’m a big fan of “top-down iterative writing”, for all kinds of academic texts. This is my take on why it works and one way to do it, in a series of (individually pain-minimised) steps from blank screen to submission 🧪

0286

Gustav Markkula @markkula.bsky.social · 02/02/2025

A recipe for pain-free(?) academic writing gmarkkula.github.io/writing-reci... I’m a big fan of “top-down iterative writing”, for all kinds of academic texts. This is my take on why it helps, and one way to do it, in a series of (individually pain-minimised) steps from blank screen to submission 🧪

4232

Gustav Markkula @markkula.bsky.social · 12/12/2024

NYT were a little playful with their Connections game today, using pictograms instead of words, but the normally soft spoken puzzlers in the comments section were clearly not in a playful mood 😂

Abby
Sydney
Dec. 11
I absolutely HATED this. I like Connections because it’s a word puzzle, not a visual puzzle. It’s a massive “No” from me.
31 Replies792 RecommendShareFlag

Caz commented December 11
C
Caz
Australia
Dec. 11
I can see this working as a completely separate puzzle but please, please don’t ruin my favourite puzzle by adding in this version. It changes it from being a word puzzle to an interpretation of a picture puzzle.
4 Replies649 RecommendShareFlag

タイタイ commented December 11
タ
タイタイ
日本
Dec. 11
I didn't like this one. The text version is less ambiguous.

A drawing of a mouth? Maybe lips? Teeth? Hard to tell from the image what the focus is supposed to be.

PaisleyP
Australia
Dec. 11
Detested this format. Not all of us associate the pictures with the words they were assigned. I think this is too personal to the experience of the puzzle maker, not universal enough to be enjoyable. Please don't do it again, or make it a different puzzle. Connections is much more fun for me as a word game.
5 Replies430 RecommendShareFlag

H commented December 11
H
H
NZ
Dec. 11
I sure love trying to identify the ambiguous ice cream topping from a monochromatic icon. Good thing it couldn't be one of literal hundreds of options 😠
11 Replies341 RecommendShareFlag

Bryan commented December 11
B
Bryan
Brisbane
Dec. 11
Add me to the list of folks who didn’t like this version of the puzzle. Please go back to the words.

Felix
Göttingen
Dec. 11
Please never do this again.
Reply105 RecommendShare

020

Gustav Markkula @markkula.bsky.social · 26/11/2024

In other news it’s great to be back in lovely Delft, looking forward to look my most serious (?) tomorrow in the Dutch academic paraphernalia

000

Gustav Markkula @markkula.bsky.social · 25/11/2024

Was wondering how global inequality has changed over time and found this great read. Improvement in the last 20 years, but you still need to go back to ca 1880 to find a more equal world, and that elephant curve behind the improvement resonates a lot with current events… wir2022.wid.world/chapter-2/

Graph showing Gini coefficient going up from 0.60 in 1820 to 0.72 in 1910 then fluctuating thereabouts until 2000 from where it drops to 0.67 in 2020.

Graph showing the global growth from 1980 to 2020 in per capita real income for the different percentiles of that income. There is one peak around the 40-60 percentile, a trough at around the 80-90 percentile and then another peak for the very highest percentiles - overall showing how the lower and middle income classes of rich countries have lagged behind in growth

040

Gustav Markkula @markkula.bsky.social · 24/11/2024

Name a game with a 10/10 soundtrack Still gives me the shivers

Promotional artwork for the game Outer Wilds

350

Gustav Markkula @markkula.bsky.social · 03/10/2024

Leading to this little 1964 gem for example www.science.org/doi/10.1126/...

000

Gustav Markkula @markkula.bsky.social · 18/05/2024

I don’t care if it was a reference manager misclick or a misunderstanding of my weird name, I appreciate the personal touch regardless 🧪

Screenshot from paper: “Accordingly, Telban [16] proposed a visual–vestibular interac-
interaction model. Later, Gustav [17] confirmed that during driving, drivers integrate visual and vestibular cues rather than relying solely on the vestibular system for perceptual information.”

050