Sascha Marton

Machine Learning Researcher & PhD Candidate at University of Mannheim

[Sascha Marton @__smarton]

🚀Exciting news! Our paper "Mitigating Information Loss in Tree-Based Reinforcement Learning via Direct Optimization" got accepted at #ICLR2025 ! 🎉🌳🤖
We introduce SYMPOL, a method that directly optimizes hard, axis-aligned decision trees with policy gradients! 🔥

🔽 Thread 👇

[Patrick Knab @p_knab]

What’s in the Bottle? A Survey and Roadmap of Concept Bottleneck Models

CBMs are a rapidly growing direction in interpretable-by-design machine learning. However, the field has become increasingly fragmented.

Preprint Link:
https://t.co/SHspTEW8P2

[Patrick Knab @p_knab]

We took a quick look at the new PaperDecision results for ICLR 2026 and asked a simple question: do we really need LLMs for this task?

We did the same but with tabular data:
https://t.co/rx1P8BU03G

#ICLR2026

[Patrick Knab @p_knab]

@dyhTHU Super cool. We have built a similar tool, but based on tabular data. Let's see which of the two works best ;)
https://t.co/rx1P8BU03G

[Patrick Knab @p_knab]

🚀 Preprint out: Concepts in Motion: Temporal Bottlenecks for Interpretable Video Classification

📄 Preprint: https://t.co/mR0SoYefs2
💻 Code: https://t.co/qNJJuHw7Jz

[ICLR @iclr_conf]

The calm before the storm #ICLR2025 🔥🔥🔥

[Sascha Marton @__smarton]

📝 Full paper: https://t.co/WqaPmlFciW
💻 Code: https://t.co/FsY8qZUBic

#ICLR2025 #Spotlight #ReinforcementLearning #SymbolicAI #DecisionTrees #ExplainableAI #MachineLearning

[Sascha Marton @__smarton]

🌟 Spotlight at #ICLR2025!
Step by our poster #425 this Thu (Apr 24), 10:00–12:30 SGT — we’d love to chat if you’re around! 🇸🇬

📍 “Mitigating Information Loss in Tree-Based RL via Direct Optimization”
With SYMPOL, we directly optimize axis-aligned DTs using policy gradients! 🌳

[Sascha Marton @__smarton]

🔍 Why SYMPOL?
✅ Direct optimization of discrete trees — no post-hoc conversion
✅ Fully interpretable and symbolic policies
✅ No soft trees — just clear, axis-aligned splits
✅ Strong results on classic RL benchmarks

[Patrick Knab @p_knab]

Which LIME should I trust? Our new paper, accepted at XAI 2025 in Istanbul, answers this question!

LIME is a go-to for post-hoc explanations—but with so many variants, which one should you use? 🧵👇

📝 Paper: https://t.co/IeFk5q8cpw

#XAI2025 #XAI #ML

[Sascha Marton @__smarton]

Excited to see our paper on ReMeDe Trees featured! 🚀
We propose a novel recurrent decision tree architecture with an internal memory to capture dependencies in sequential data. 🌳🧠
Efficient, interpretable, and optimized via gradient descent.
Thanks @gm8xx8 for the shoutout! 🙌

[𝚐𝔪𝟾𝚡𝚡𝟾 @gm8xx8]

Decision Trees That Remember: Gradient-Based Learning of Recurrent Decision Trees with Memory

paper: https://t.co/cpaiRFCxZ4

ReMeDe Trees is a decision tree architecture designed to handle sequential data by integrating an internal memory mechanism. It learns long-term dependencies through hard, axis-aligned decision rules optimized via gradient descent. Synthetic benchmark tests demonstrate its effectiveness in capturing temporal patterns.

[fly51fly @fly51fly]

[LG] Decision Trees That Remember: Gradient-Based Learning of Recurrent Decision Trees with Memory
S Marton, M Schneider [University of Mannheim & Boehringer Ingelheim] (2025)
https://t.co/v72EBeCF2K

[Sascha Marton @__smarton]

Excited to present SYMPOL @iclr_conf! See you in Singapore 🇸🇬🏝️
Let's discuss interpretable RL, symbolic policies & decision trees! Looking forward to meeting everyone there! 💬🔥

#ICLR2025 #ReinforcementLearning #SymbolicRL #InterpretableAI #DecisionTrees

[Sascha Marton @__smarton]

🚀Exciting news! Our paper "Mitigating Information Loss in Tree-Based Reinforcement Learning via Direct Optimization" got accepted at #ICLR2025 ! 🎉🌳🤖
We introduce SYMPOL, a method that directly optimizes hard, axis-aligned decision trees with policy gradients! 🔥

🔽 Thread 👇

[Sascha Marton @__smarton]

Want to dive deeper? 🤓
📄 Paper: https://t.co/WqaPmlFciW
💻 Code: https://t.co/FsY8qZUBic
Integration with libraries like SB3 coming soon—follow for updates! 🚀

[Jannik Brinkmann @jannikbrinkmann]

Why do LLMs trained on over 90% English text perform so well in non-English languages?

We find that they learn to share highly abstract grammatical concept representations, even across unrelated languages!

New paper w/ @wendlerch and @amuuueller

[David Holzmüller @ ICML 2026 @DHolzmueller]

PyTabKit 1.1 is out!

- Includes TabM and provides a scikit-learn interface
- some baseline NN parameter names are renamed (removed double-underscores)
- other small changes, see the readme.

[Sascha Marton @__smarton]

@puhsuuu @alexUnder_sky Fingers crossed for you guys! Your work deserves to get accepted!