Jaron Maene (@jjcmoon) header
Jaron Maene (@jjcmoon) profile picture

PhD Machine Learning @KU_Leuven | #neurosymbolic learning & #probabilistic reasoning

Joined January 2022
Photo by Jaron Maene (@jjcmoon)Photo by Jaron Maene (@jjcmoon)Photo by Jaron Maene (@jjcmoon)Photo by Jaron Maene (@jjcmoon)Photo by Jaron Maene (@jjcmoon)Photo by Jaron Maene (@jjcmoon)Photo by Jaron Maene (@jjcmoon)Photo by Jaron Maene (@jjcmoon)
An internal model at OpenAI has produced a claimed refutation of ETH! Russell Impagliazzo, Mohan Paturi, and I have been asked to help organize an open and independent examination of this claim. 1/4
304
792
544
7,422
3,755,747
I let gpt-6 astra review 250 random ICLR submissions. It didn't recommend clear accept for a single paper.
25
15
5
203
32,336
Reinforcement learning from reviewer feedback
7
208
“we sandboxed the agent” meanwhile the agent:
258
3,410
247
35,501
1,780,140
TIL I'm a professional cyclist...
Not bad but a bit outdated indeed.
2
1
3
721
Jaron Maene retweeted
Just a little under 5 days left to submit your work on #tractable #probabilistic #models to TPM [email protected] (June 12, AoE)! If you have ongoing projects or recently accepted papers that are #tractable, #causal or #neurosymbolic, then check the CfP at tractable-probabilistic-mode…
1
1
2
106
Jaron Maene retweeted
When you prompt an LLM for code, you get one deterministic program. However, the LLM actually defines a distribution over many programs, and existing methods discard it‼️ PPoT uses this distribution to extract free performance and efficiency gains. 🧵👇
1
17
1
68
12,969
Jaron Maene retweeted
How far you can travel by train in 3 hours from Brussels
132
366
433
10,325
6,262,895
the next-gen EUV machine from ASML got delivered yesterday at @imec_int in Leuven. as a Belgian, it's amazing to see such unique machines being deployed in my alma mater's city
1
12
675
Only 10 days left to submit your work to ProbML (formerly AABI)! 👉 probml.cc/call/
7
8
1,695
Jaron Maene retweeted
Too many papers fail to provide a reasonable related literature comparison. I'm not even talking about a good comparison...it’s simply part of doing research to place your result within the existing literature.
2
1
34
5,658
Jaron Maene retweeted
If you care about enforcing constraints over time without breaking your computational resources, then read our new blog post over at @aihuborg! It focuses on showing how our neurosymbolic Markov models beat the SOTA in out-of-distribution generalisation and so much more.
5
6
498
all the LLM bros are obsessed with counting model weights, but me? I do weighted model counting 😎
Replying to @xuanalogue
Especially glad I started the PhD after the Russell-Domingos-Tenenbaum-etc synthesis of logic and symbolic representations with probability theory. We actually can have the best of both worlds!
2
2
24
2,296
1/5 Tomorrow I’ll talk about the 𝐩𝐫𝐨𝐛𝐚𝐛𝐢𝐥𝐢𝐬𝐭𝐢𝐜 𝐩𝐫𝐨𝐠𝐫𝐚𝐦𝐦𝐢𝐧𝐠 𝐬𝐞𝐦𝐚𝐧𝐭𝐢𝐜𝐬 𝐨𝐟 𝐝𝐢𝐟𝐟𝐞𝐫𝐞𝐧𝐭𝐢𝐚𝐛𝐥𝐞 𝐩𝐫𝐨𝐯𝐢𝐧𝐠 at #NeurIPS San Diego (poster #614 11am). 📃 openreview.net/pdf?id=rEUbDh… 📺 youtube.com/watch?v=sOTXBsmG…
1
4
8
1,845
4/5 We show how to factorize this equivalence distribution using latent categorical variables—a kind of embeddings. Plus we develop exact (via singularization + magic sets) and approximate (Monte Carlo) inference methods.
1
1
99
5/5 Joint work with Efthymia Tsamoura. Paper: openreview.net/pdf?id=rEUbDh…
1
87
Jaron Maene retweeted
New paper from Basis' Project MARA team and collabs. The ability to learn and use world models is a key aspect of human intelligence, but evaluating this ability remains elusive. In this work we propose WorldTest, a representation-agnostic, behavior-based agent eval framework.
1
11
2
20
4,282