Vedran Sekara

Asst professor @ITUkbh. Previously @UNICEFinnovate & @sony
Working on data representativeness, AI & ML bias, and crappy algorithms | Learning about this mastodon stuff

Vedran Sekara boosted:

@robysinatra @vedransekara These PhDs will proceed in #Copenhagen. In Denmark, a PhDs is considered a full-time paid working position (37h) with excellent work-life balance: denmark.dk/society-and-busines. Here also an interesting infographic about salary from a few years ago:

Bar chart of average PhD stipend versus living costs for different countries/cities, ordered by stipend to living cost ratio. Denmark is on the left with 1.6.
2023-03-13

Are you interested in doing a PhD at the intersection of computational social science, AI & ML, and network science then look at this 👇

I'm looking for a PhD student for a project (w. @robysinatra) on understanding the predictive limits of ML systems

nerds.itu.dk/2023/03/08/two-ne

Vedran Sekara boosted:

🚨Two new PHD calls! Deadline April 1st. With @robysinatra or @vedransekara. Topics: Science of Science / Algorithmic fairness, and Predictability of Social Systems

👉 nerds.itu.dk/2023/03/08/two-ne

Photo of Copenhagen's Nyhavn harbor with colorful houses and boats.
Vedran Sekara boosted:
2022-12-22

The Pioneer Centre for AI now has a Mastodon account: @AiCentreDK

2022-12-20

For scenarios involving luck and hidden unknowns, AI models are not better than simple models. In these situations, simple models are just as accurate, far more transparent, and worth considering

/FIN

2022-12-20

Final summary, the simple benchmark was equally good at predicting the group stage resutls, and outperformed the sophisticated AI in the knock-out stages, and correctly predicted the winner of the final!

The moral of the story:
AIs excel at some tasks, but cannot predict everything. We should beware of the hype that surrounds them. Often, AI models are not substantially better than simple models. We must resist the commercial interests that aim to obfuscate this fact

#AIHype

3/

2022-12-20

For the knock-out stages the prediction task is easier since games cant end in a draw (so predictive accuracies are higher)

Overall, the AI model predicted FRA 🇫🇷🏆❌ to win and had an overall accuracy of 75%

The benchmark model, uses FIFA's ranking of teams (from before the world cup), and predicts the highest ranked team to always win.

This model predicted ARG to win the final 🇦🇷🏆✅
Its overall accuracy was 81.25% 👀

2/

2022-12-20

The world cup is over (what a great game between ARG 🇦🇷 - FRA 🇫🇷), congrats to Argentina on the fantastic win!! ⚽️🏟🏆

Its time to evaluate the last AI predictions for the world cup and settle the question can "AI predict the world cup?" 🤔

We will look at results from the knock-out games (i.e. from round of 16, quarter-finals, semi-finals, ets.)

But first lets summarize our findings from the group stage, here the accuracy was

AI model: 54.17%
Benchmark model: 54.17%

#FIFAWorldCup2022 #AI

1/

Client Info

Server: https://mastodon.social
Version: 2025.07
Repository: https://github.com/cyevgeniy/lmst