Institution profile

Max Planck Institute for Intelligent Systems

Academic institutioneurope · de
Official website
Research library414linked papers
Opportunities0open roles
Selected work

Representative Papers

Reinforcement Learning via Self-Distillation

Jan 28, 2026

This work addresses the challenge in reinforcement learning (RL) for verifiable domains such as code generation and mathematical reasoning, where reliance on sparse scalar rewards impedes effective credit assignment despite abundant textual feedback from the environment. To overcome this limitation, the authors propose Self-Distillation Policy Optimization (SDPO), a novel approach that leverages the model’s own predictions of corrective tokens from textual feedback to construct dense supervision signals. These signals are then distilled back into the policy network via a self-teaching mechanism, enabling context-aware self-correction without external teachers or explicit reward models. SDPO is the first method to formally treat textual feedback as an RL signal, significantly outperforming existing RLVR approaches on scientific reasoning, tool usage, and competitive programming tasks in LiveCodeBench v6. At inference time, applying SDPO to a single problem reduces the number of attempts needed to achieve comparable success rates by a factor of three.

6 citationsRead paper

Imagining and building wise machines: The centrality of AI metacognition

Nov 04, 2024arXiv.org

Current AI systems exhibit intelligence but lack human-like wisdom, primarily due to the absence of metacognitive capabilities—such as intellectual humility, perspective-taking, and contextual adaptability—resulting in insufficient robustness, explainability, human-AI collaboration, and goal alignment in novel environments. To address this, the paper formally introduces “AI metacognition” as the cornerstone of artificial wisdom and proposes the first computationally grounded metacognitive capability framework, shifting beyond conventional object-level performance optimization. Methodologically, it integrates cognitive science modeling, explainable AI, value alignment, and novel wisdom-oriented benchmark design. The contributions include: (1) a theoretical foundation for wise AI; (2) a multidimensional evaluation framework; and (3) a principled implementation pathway. This work lays the groundwork for developing next-generation AI systems that are safe, trustworthy, and adaptive.

4 citations1 influentialRead paper

Root Cause Analysis of Outliers with Missing Structural Knowledge

Jun 07, 2024arXiv.org

Real-world root cause analysis (RCA) faces a critical challenge: post-intervention distributions often contain only a few—or even a single—sample, rendering distribution-dependent or low-density-region regression methods statistically ill-posed. This paper proposes a lightweight root cause identification framework that requires neither counterfactual reasoning nor a fully specified structural causal model (SCM). It operates either given a causal DAG or, in the absence of one, solely from an anomaly score ranking. We theoretically prove that low-scoring anomalies rarely trigger high-scoring ones and derive a probabilistic upper bound on non-monotonic propagation paths. By abandoning Shapley-value-based attribution and density-sensitive regression, our method achieves linear time complexity O(n). It eliminates SCM fitting and counterfactual computation while providing rigorous theoretical guarantees and strong empirical performance.

4 citations1 influentialRead paper

Incentivizing Honesty among Competitors in Collaborative Learning and Optimization

May 25, 2023Neural Information Processing Systems

This paper addresses the problem of strategic participants in collaborative learning who submit misleading model updates to gain competitive advantage, thereby degrading others’ performance. Unlike prior work assuming malicious adversaries, we model participants as self-interested yet non-malicious rational agents and propose the first incentive-compatible mechanism for federated learning. Leveraging game-theoretic analysis and mechanism design, we rigorously prove that truthful updates constitute a Nash equilibrium—both in single-round mean estimation and multi-round strongly convex SGD. Theoretically, the mechanism guarantees learning performance approaching that of full cooperation. Empirically, we validate its effectiveness on non-convex federated benchmarks (e.g., FEMNIST): it substantially mitigates strategic manipulation, achieves convergence rates comparable to fully cooperative training, and seamlessly integrates with standard FedAvg without architectural modification.

3 citationsRead paper

Adding Internal Audio Sensing to Internal Vision Enables Human-Like In-Hand Fabric Recognition with Soft Robotic Fingertips

Sep 30, 2025IEEE-RAS International Conference on Humanoid Robots

Distinguishing the feel of smooth silk from coarse cotton is a trivial everyday task for humans. When exploring such fabrics, fingertip skin senses both spatio-temporal force patterns and texture-induced vibrations that are integrated to form a haptic representation of the explored material. It is challenging to reproduce this rich, dynamic perceptual capability in robots because tactile sensors typically cannot achieve both high spatial resolution and high temporal sampling rate. In this work, we present a system that can sense both types of haptic information, and we investigate how each type influences robotic tactile perception of fabrics. Our robotic hand's middle finger and thumb each feature a soft tactile sensor: one is the opensource Minsight sensor that uses an internal camera to measure fingertip deformation and force at 50 Hz, and the other is our new sensor Minsound that captures vibrations through an internal MEMS microphone with a bandwidth from 50 Hz to 15 kHz. Inspired by the movements humans make to evaluate fabrics, our robot actively encloses and rubs folded fabric samples between its two sensitive fingers. Our results test the influence of each sensing modality on overall classification performance, showing high utility for the audio-based sensor. Our transformer-based method achieves a maximum fabric classification accuracy of 97% on a dataset of 20 common fabrics. Incorporating an external microphone away from Minsound increases our method's robustness in loud ambient noise conditions. To show that this audio-visual tactile sensing approach generalizes beyond the training data, we learn general representations of fabric stretchiness, thickness, and roughness.

2 citationsRead paper
Recent publications

Latest Papers