iiinova — where your digital self keeps evolving

iiinova is part of the Atmosphere — an open, public network where you own your identity.

Adrián Javaloy

@javaloyml.bsky.social

1.7K followers 423 following 94 posts

Lecturer at the University of Bath working on Machine Learning. Previously in Edinburgh, Saarbrücken, Tübingen, and Murcia. 🌐 adrianjav.github.io

Posts Replies Media

Adrián Javaloy @javaloyml.bsky.social · 16/09/2026

It has been a while since I have been active here, and lots of things have changed: different city, different role. I have been swamped, but I finally managed to update my website. More news sooner than later! (I accept suggestions for lab names) adrianjav.github.io

1163 16

Adrián Javaloy @javaloyml.bsky.social · 12/02/2026

More importantly, we show that this comes at no cost in performance, and we can even train non-structured-decomposable squared circuits* (* That is, circuits which cannot be efficiently squared)

120 2

Adrián Javaloy @javaloyml.bsky.social · 12/02/2026

As a result, we can train really large squared circuits while saving both time and memory! At 357M parameters, we (⟂) use: - 12 GiB vs 18 GiB (33% reduction!) - 0.29ms vs 0.52ms per iteration (44% faster!)

110 1

Adrián Javaloy @javaloyml.bsky.social · 30/04/2025

Today we have @lennertds.bsky.social from KU Leuven teaching us how to adapt NeSy methods to deal with sequential problems 🚀 Super interesting topic combining DL + NeSy + HMMs! Keep an eye on Lennert's future works!

093 9

Adrián Javaloy @javaloyml.bsky.social · 30/04/2025

This year I am co-organizing the 8th iteration of the Tractable Probabilistic Modeling #TPM workshop at #UAI2025 🌴 Rio de Janeiro edition 🌴 🌐 lnkd.in/dDK8T5Au ⏰ Submission deadline: May 23th AoE 🌴 Date: July 15th 🧵👇

1159 15

Adrián Javaloy @javaloyml.bsky.social · 29/03/2025

Why doing that when you could go full bash tho? Apologies too! 👒 For refs: superuser.com/questions/15...

240 4

Adrián Javaloy @javaloyml.bsky.social · 13/02/2025

If you prefer to read a practical example, you can find a basic tutorial here: causal-flows.readthedocs.io/en/latest/tu... Using a triangular SCM as an example (x1 -> x2, x3 and x2 -> x3), I show how to fit a Causal NF on observational data (left), and use it for causal inference (right).

Pairplot where a Causal NF perfectly matches the joint distribution of the observational data.

The same pair plot, this time simulating an intervention in x1 = 2.5, showing that the Causal NF accurately estimates the true interventional distribution.

130 3

Adrián Javaloy @javaloyml.bsky.social · 13/02/2025

With this new library, defining and training a Causal NF can be as simple as using any other predefined NF. If you feel fancy, you can also build your own Causal NF composing any transformation and distribution from the Zuko library for regular NFs (github.com/probabilists...) 👷

Snippet to define a train a Causal NF, just as you would do with any other predefined PyTorch module.

Snippet where a CausalFlow is built using different transformations from the Zuko library.

100 1

Adrián Javaloy @javaloyml.bsky.social · 19/12/2024

👏 every 👏 single 👏 damn 👏 time 👏 Note to self: why do I even keep trying it?

A screenshot of a conversation with ChatGPT where I asked it to find the general pattern of a sequence, and showed things "two divided by two equals 1.5"

050 5

Adrián Javaloy @javaloyml.bsky.social · 26/11/2024

Thanks for sharing Alex! It looks super interesting Also, wow... 🫠

110 1

Adrián Javaloy @javaloyml.bsky.social · 13/11/2024

Over the years, I have developed this bad habit of typing twitter on the address bar when I don't know what to do, which made me never really quit it. This time, I am redirecting that habit somewhere else 🙄 PS: The add-on is called Redirector: addons.mozilla.org/en-GB/firefo...

Screenshot of a rule in Redirector (a plugin on Firefox), to redirect twitter to bluesky.

040 4