Yorgos Pantis

Ph.D. studies @uoaofficial and @ArchimedesUnit.
Ex-research @MIT, @dtutweet, @athenaRICinfo, @studyatctu, @mpiMathSci.
Ex-studies @UniofOxford, @upatras.

[Yorgos Pantis @yorgos_pantis]

(1/4) 🚨 @NeurIPSConf 2025 Announcement Our paper “Teaching Transformers to Solve Combinatorial Problems through Efficient Trial & Error” will be presented in San Diego!

Paper: https://t.co/OHMS524VjX

[Alireza Mousavi @alirezamh_]

I’m at ICML 🇰🇷 presenting a 🔦 spotlight 🔦 poster today. Check it out if you want to know how process rewards help RLVR go beyond the base model support when outcome rewards alone can’t.

📆 Thu. 5-6:45 pm. Hall A. Poster # 1307.

[Zixuan Wang @zzZixuanWang]

A question on synthetic data generation: If we want a language model to solve k-step arithmetic problems (such as a+b*c-d=?), with operands from 1 to 100, which training distribution should we use?

A. Uniform distribution: Sample these k operands uniformly from 1 to 100
B. Power law: randomly shuffle 1-100 and impose an artificial power law. Sample these k operands according to this power law.

⚡Our ICML 2026 (spotlight) paper shows: Option B is better! Surprisingly, the same idea extends far beyond this simple example to many reasoning tasks that require implicit composition of multiple atomic skills, including multi-hop QAs and synthetic GSM problems.

📄Paper: https://t.co/8LVlmCCBz9
📝Blog: https://t.co/t2yhIJ0s1P

[Thea Sommerschield @TSommerschield]

We're making a (epigraphic) habit out of this... 😉 Introducing the Predicting the Past Skill in @antigravity: grounding Gemini directly in our specialised Aeneas and Ithaca models to study Greek + Latin inscriptions interactively.
With @iassael & Zoi Tsangalidou @GoogleDeepMind

[Yilun Du @ ICML @du_yilun]

Check out our work on building EBMs for molecular graphs!

[Yuchen Zeng @yzeng58]

💻Tired of running so many slow, expensive benchmark evals across every checkpoint?

Try ✨BenchPress✨ at https://t.co/uB3gUBwc0Z: provide a few benchmark scores, then get predictions for the remaining ~100 benchmarks, with trust probabilities and calibrated 90% prediction intervals.

How does this work? In his original post (https://t.co/XSuJ44bFp3), @DimitrisPapail first tried the idea as a fun question: collect model-by-benchmark scores into a matrix, find its low-rank structure, and use matrix completion to predict missing benchmark scores from a few observed ones.

We expanded this into a full system: a fully audited 84-model x 133-benchmark score matrix, an optimized matrix-completion predictor, and a reliability layer for trust probabilities and 90% prediction intervals.

Beyond predicting missing scores, we also suggest practical seed benchmark sets. The five-probe set {GPQA-D, HLE, Codeforces, MMLU-Pro, ARC-AGI-1} recovers the rest of a model's public score profile with a MedAE of 3.93 points. A lower-cost set {GPQA-D, MMLU-Pro, Aider Polyglot, MATH-500, AIME 2026} reaches 4.55 points.

See more details below 🧵1/7

This work is with @DimitrisPapail at AI Frontiers, a boutique research lab inside @MSFTResearch.

[Bill Psomas @bill_psomas]

🎉 REGLUE accepted at #ECCV2026 🎉

🎨 A unified framework jointly modeling VAE latents ➕ global ➕ local VFM semantics for faster, higher-fidelity diffusion image generation.

💨 Matches 1M-step SOTA in just 700k iterations (~30% fewer steps).

More info below👇

[Efstathios Karypidis @K_Sta8is]

🎉Re2Pix accepted at #ECCV2026!

💡Should a world model predict future dynamics and render pixels simultaneously? Re2Pix says no. Forecast in VFM semantic space first 🧠, synthesize pixels second 🎨

Updated Paper and code coming soon. Details👇

[Giannis Daras @ ICML 2026 @giannis_daras]

Training robots typically requires a lot of teleop data.

Ambient Diffusion Policy allows you to learn from suboptimal data in robotics 🤖, like different robots performing entirely different tasks, in real life or in sim.

The result? A significant decrease in the number of demonstrations you need.

[Antonis Anastasopoulos @anas_ant]

Ιστορία Δύο Ινστιτούτων - Story of Two Institutes
Άρθρο από το Χρίστο Παπαδημητρίου για τα εμπόδια που βάζει η ελληνική κυβέρνηση στο Ινστιτούτο Τεχνητής Νοημοσύνης Αρχιμήδης

Article in Greek about the financial woes of the Archimedes AI Institute
https://t.co/ldPpa2vpSa

[Theo Kypraios @tkypraios]

The 16th edition of the #GreekStochastics meeting will be taking place in #Zakynthos, Greece, on 28-31 July 2026 with the focus being on Scalable Inference with 3 short courses by Lyudmila Grigoryeva, Panayioti Mertikopoulo & Chris Nemeth. More details: https://t.co/ogF0D2Wi64

[Greeks in AI @greeksinai]

🇬🇷 Reminder: the Call for Papers for #GreeksInAI 2026 is open!

📍Eugenides Foundation, Athens
📅15–17 July 2026
📝Submission deadline: 15 May 2026

We invite submissions on recent/ongoing/emerging work in AI/ML worldwide.

Submit via: https://t.co/QpSR4KsC4x

[Nikos Tsikouras @nikos_tsikouras]

#ICLR2026 🇧🇷

If you are interested in finding out why weights of neural networks discover the hidden signal structure, come see us at Pavilion 3 #815.

This is joint work with @yorgos_pantis, @bouzoukipunks and @ChristosTzamos.

[Yorgos Pantis @yorgos_pantis]

@dvruette I recommend taking a look at our recent NeurIPS paper; in this work, we train a GPT-2 model from scratch and reach SOTA performance, achieving 99.8% accuracy on randomly generated Sudoku. Link:
https://t.co/OHMS524nup

[Yorgos Pantis @yorgos_pantis]

@logic_int I suggested you read our recent NeurIPS paper, “Teaching Transformers to Solve Combinatorial Problems Through Efficient Trial and Error.” We train a GPT-2 model from scratch and achieve state-of-the-art 99.8% accuracy on randomly generated Sudoku.

Link: https://t.co/OHMS524VjX

[Yorgos Pantis @yorgos_pantis]

Very happy to see our work accepted at @iclr_conf 2026. Congrats to the team! 💪

[Nikos Tsikouras @nikos_tsikouras]

1/4 Excited to share that our paper has been accepted at ICLR 2026! 🎉

Preprint available here: https://t.co/Syycx5xYnQ

[Constantine Caramanis @cmcaram]

Μαθητές Γυμνασίου/Λυκείου στην Ελλάδα: άνοιξε ο Πανελλήνιος Διαγωνισμός Τεχνητής Νοημοσύνης (online).
Προθεσμία: 1 Μαρτίου 2026.
Διάλεξε 1 από τα προβλήματα και κάνε την 1η σου υποβολή. Links στο 1ο σχόλιο 👇
#ΤεχνητήΝοημοσύνη #AI #Πληροφορική

[Yorgos Pantis @yorgos_pantis]

Announcing the 2nd Greek Olympiad in AI!

Site: https://t.co/q65gVn5TNn

[Athena RC @athenaRICinfo]

🇬🇷 The Greek Olympiad in Artificial Intelligence (PDTN) returns for a second year! Students from across Greece are invited to explore AI through hands-on challenges in ML, computer vision & NLP.

ℹ️ More info: https://t.co/hSAuZ25ZUK

[Christos Tzamos @ChristosTzamos]

Announcing the 2nd Greek Olympiad in AI!

The contest is open to middle and high school students in Greece aiming to form the national team for @IOAIOfficial

Even if you do not meet these criteria, you may participate unofficially solving interesting AI tasks.

Spread the word!

[Christos Tzamos @ChristosTzamos]

At #NeurIPS2025 today? Check out our poster at 3403. Teaching transformers to solve combinatorial problems.

[Yorgos Pantis @yorgos_pantis]

If you’re in San Diego, come by to discuss our paper on Transformer-based reasoning in combinatorial problems, co-presented with @ChristosTzamos.

🗓️ Poster presentation: Thursday, Dec 4, 11:00 – 14:00.

📍 Location: Exhibit Hall C, D, E.

[Yorgos Pantis @yorgos_pantis]

(1/4) 🚨 @NeurIPSConf 2025 Announcement Our paper “Teaching Transformers to Solve Combinatorial Problems through Efficient Trial & Error” will be presented in San Diego!

Paper: https://t.co/OHMS524VjX