iiinova — where your digital self keeps evolving

Martin Trapp @trappmartin.eurosky.social · 19/10/2026

After a quick look, the definition of an extended abstract is quite odd here. 6 pages (inc. refs), and authors are discouraged from including a thorough literature review. Is this common?

100 1

Martin Trapp @trappmartin.eurosky.social · 25/08/2026

Excited for the WASP(-HS) workshop on Bias, Representation and Transparency in generative models!

000 0

Martin Trapp @trappmartin.eurosky.social · 04/05/2026

Cool work from Helsinki folks on discrete flow matching without actually going discrete but working on the simplex. #AISTATS

160 6

Martin Trapp @trappmartin.eurosky.social · 04/05/2026

Polynomial chaos expansions go deep… cool work by @ropeharz.bsky.social’s group on fusing circuits and PCEs. #AISTATS

291 9

Martin Trapp @trappmartin.eurosky.social · 02/05/2026

The prior strikes back, awesome work by @davidruegamer.bsky.social’s group. Thanks to Julius for the great explanation. #AISTATS

1196 19

Martin Trapp @trappmartin.eurosky.social · 02/05/2026

Super cool work on learning PC parameters efficiently and effectively by @anjiliu.bsky.social! Big congrats!

1101 10

Martin Trapp @trappmartin.eurosky.social · 21/04/2026

There are sooo many sakura 🌸 at the moment in Stockholm. Very nice!

180 8

Martin Trapp @trappmartin.eurosky.social · 31/03/2026

Skiing in northern Lapland is fun!

0130 13

Martin Trapp @trappmartin.eurosky.social · 08/12/2025

I will also add my two cents and strongly agree with what you wrote and with many of the points you raised in your keynote. Picture of the night train I took yesterday below.

040 4

Martin Trapp @trappmartin.eurosky.social · 11/09/2025

I mean, it did start with a rather illuminati-cult like logo.

041 4

Martin Trapp @trappmartin.eurosky.social · 30/08/2025

Swimming in foggy water with friends, what else could one wish for. 🤩

060 6

Martin Trapp @trappmartin.eurosky.social · 18/08/2025

Considering how “warm” I feel now, this was probably the last swim of the year. But what a beautiful swim. 🤩

060 6

Martin Trapp @trappmartin.eurosky.social · 21/07/2025

Moreover, #BitVI can help identify the numerical precision required to represent a target density, a crucial task in the quantisation of neural networks and when working with low-precision regimes.

Entropy of BitVI for varying complexity of the target distribution.

Bayesian neural networks with varying number of numerical precision.

110 1

Martin Trapp @trappmartin.eurosky.social · 21/07/2025

In #BitVI, we exploit the fixed-point representations of numbers and a tractable variational approximation based on #circuits. Thus, enabling efficient ELBO computation and control of the numerical precision. Additionally, BitVI is easily extendable to other number systems, such as floating-point.

Binary decision tree induced by BitVI for fixed-point representations.

Resulting circuit model used in BitVI with fixed-point representations.

110 1

Martin Trapp @trappmartin.eurosky.social · 21/07/2025

Remember that computers use bitstrings to represent numbers? We exploit this in our recent @auai.org paper and introduce #BitVI. #BitVI directly learns an approximation in the space of bitstring representations, thus, capturing complex distributions under varying numerical precision regimes.

2223 22

Martin Trapp @trappmartin.eurosky.social · 10/06/2025

Paris was such a treat! Back to 🇫🇮!

050 5

Martin Trapp @trappmartin.eurosky.social · 28/04/2025

Big tech made it possible to finally have one of the most important problems of mankind solved. I give you… ChickenZilla!!!!

150 5

Martin Trapp @trappmartin.eurosky.social · 15/04/2025

We had a pretty dystopian atmosphere one evening in Jakarta with heavy rain. So I made it look even more dystopian. 😅

2311 31

Martin Trapp @trappmartin.eurosky.social · 02/04/2025

Crazy evening sky today in Helsinki.

1242 24

Martin Trapp @trappmartin.eurosky.social · 20/03/2025

I’m looking for some awesome machines learning or Bayesian stats #memes! Any suggestions?

200 2

Martin Trapp @trappmartin.eurosky.social · 19/03/2025

Instead of increasing the inference (prediction) cost by a factor of ~5 - 70 x (depending on the method), you can now make predictions with almost no additional overhead.

100 1

Martin Trapp @trappmartin.eurosky.social · 19/03/2025

Because we effectively turn the NN into a multilinear function, we can also assess the model's sensitivity to input perturbations. For example, we can estimate sensitivity maps by learning Gaussian input densities for the covariates and visualise their input sensitivity information.

100 1

Martin Trapp @trappmartin.eurosky.social · 19/03/2025

We develop a deterministic propagation path for various NN architecture choices and covariance structures of the approximate posterior. This allows us to make efficient & sampling-free predictions in Bayesian #LLMs and #ViT models.

110 1

Martin Trapp @trappmartin.eurosky.social · 19/03/2025

We ask how we can have a fast and practical approximation of the posterior predictive distribution in BNNs. Our "Streamlining" approach achieves this by approximating the posterior predictive with little computational overhead through local linearisations and local Gaussian approximations.

100 1

Martin Trapp @trappmartin.eurosky.social · 23/02/2025

Are you going to be at #WACV and want to know if “Flatness Improves Backbone Generalisation in Few-shot Classification”? Then join the oral presentation by @ruili-pml.bsky.social of our paper! 🔗 lnkd.in/dBMmN7Vs Done together with @marcusklasson.bsky.social and @arnosolin.bsky.social.

163 6

Martin Trapp @trappmartin.eurosky.social · 16/02/2025

So much new snow tonight in Helsinki. Amazing 🤩

Picture of a snowy pier and snow covered frozen water.

0100 10

Martin Trapp @trappmartin.eurosky.social · 01/02/2025

What a beautiful day today. ☀️

0100 10

Martin Trapp @trappmartin.eurosky.social · 29/12/2024

I’m so excited to test the old telescope that we found in the back of the garage. Night sky awaits.

050 5