Ya Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Ya Wang has a total of 7 co-authors. Here are the top co-authors by number of shared publications:
- Adrian Paschke
Fraunhofer Institute for Open Communication Systems
4 shared publications Daniel Barta
Fraunhofer Institute for Open Communication Systems
2 shared publicationsJulian Hesse
Fraunhofer Institute for Open Communication Systems
2 shared publicationsPhilip Buchwald
Fraunhofer Institute for Open Communication Systems
2 shared publicationsAdrian Paschke
Fraunhofer Institute for Open Communication Systems
1 shared publication- Kumar Manas
Freie Universität Berlin
1 shared publication
Latest Publications
- Open Access
- 28 July 2026
- Published by SAGE Publications in Neurosymbolic Artificial Intelligence
- Vol. 2
Empirical Analysis of Chain-of-Thought and Solver-Augmented Large Language Models for Deductive Reasoning
Large language models (LLMs) have demonstrated remarkable capabilities in multistep reasoning, yet fundamental questions remain about their ability to perform reliable deductive reasoning. Two contrasting paradigms have emerged: chain-of-thought (CoT) prompting, which treats LLMs as self-contained reasoners, and solver-augmented approaches, which use LLMs as semantic parsers to translate problems into formal logic for symbolic execution. We present a systematic comparison of these approaches under strict single-pass evaluation across three established benchmarks (ProofWriter, PrOntoQA, and LogicalDeduction) and controlled synthetic datasets. Our findings reveal three key insights: (1) on standard benchmarks, solver augmentation yields minimal improvements ( < 1 % ) over zero-shot CoT for state-of-the-art models, suggesting diminishing performance margins between paradigms under current benchmark settings; (2) mid-sized open-source models (14B–32B) match proprietary model performance when solver-augmented, with the primary challenge shifting from generating valid syntax (small models) to accurate semantic parsing (larger models); and (3) on synthetic datasets with controlled difficulty variations, solver-augmented methods demonstrate superior robustness, maintaining 76.7% accuracy retention at extreme reasoning depths (14–17 steps) compared to 62.2% for CoT, and showing greater resilience to distracting facts and rules. These results indicate that while CoT-augmented LLMs achieve strong performance on deductive reasoning tasks up to five hops, solver augmentation remains valuable for deeper multihop deduction and for applications requiring robust and verifiable reasoning.
- 28 July 2026
- Published by SAGE Publications in Neurosymbolic Artificial Intelligence
- Vol. 2
Open AccessEmpirical Analysis of Chain-of-Thought and Solver-Augmented Large Language Models for Deductive Reasoning
Large language models (LLMs) have demonstrated remarkable capabilities in multistep reasoning, yet fundamental questions remain about their ability to perform reliable deductive reasoning. Two contrasting paradigms have emerged: chain-of-thought (CoT) prompting, which treats LLMs as self-contained reasoners, and solver-augmented approaches, which use LLMs as semantic parsers to translate problems into formal logic for symbolic execution. We present a systematic comparison of these approaches under strict single-pass evaluation across three established benchmarks (ProofWriter, PrOntoQA, and LogicalDeduction) and controlled synthetic datasets. Our findings reveal three key insights: (1) on standard benchmarks, solver augmentation yields minimal improvements ( < 1 % ) over zero-shot CoT for state-of-the-art models, suggesting diminishing performance margins between paradigms under current benchmark settings; (2) mid-sized open-source models (14B–32B) match proprietary model performance when solver-augmented, with the primary challenge shifting from generating valid syntax (small models) to accurate semantic parsing (larger models); and (3) on synthetic datasets with controlled difficulty variations, solver-augmented methods demonstrate superior robustness, maintaining 76.7% accuracy retention at extreme reasoning depths (14–17 steps) compared to 62.2% for CoT, and showing greater resilience to distracting facts and rules. These results indicate that while CoT-augmented LLMs achieve strong performance on deductive reasoning tasks up to five hops, solver augmentation remains valuable for deeper multihop deduction and for applications requiring robust and verifiable reasoning.
- Open Access
- 4 March 2025
- Published by Springer Nature
- p. 319-347
Optimierung der Entscheidungsfindung in autonomen Fahrsystemen mit neuro-symbolischem Wissen
Zusammenfassung: Im sich rasch entwickelnden Bereich des autonomen Fahrens ist es entscheidend, hohe Leistung, Erklärbarkeit und regelkonforme Entscheidungsfindung gleichzeitig zu erreichen, für Sicherheit, Nutzerakzeptanz und rechtliche Haftung der Hersteller. Ansätze auf Basis von Deep Learning haben ihre Wirksamkeit in verschiedenen Aufgaben unter Beweis gestellt, doch sie scheitern oft, wenn es um Interpretierbarkeit und Legalität geht. Im Gegensatz dazu haben regelbasierte Ansätze, die zusätzliche Erklärbarkeit und Nachvollziehbarkeit bieten, Schwierigkeiten, mit verrauschten, unsicheren und großen Daten in hochdynamischen Verkehrsszenen in einer Vielzahl von Szenarien umzugehen.
- 4 March 2025
- Published by Springer Nature
- p. 319-347
Open AccessOptimierung der Entscheidungsfindung in autonomen Fahrsystemen mit neuro-symbolischem Wissen
Zusammenfassung: Im sich rasch entwickelnden Bereich des autonomen Fahrens ist es entscheidend, hohe Leistung, Erklärbarkeit und regelkonforme Entscheidungsfindung gleichzeitig zu erreichen, für Sicherheit, Nutzerakzeptanz und rechtliche Haftung der Hersteller. Ansätze auf Basis von Deep Learning haben ihre Wirksamkeit in verschiedenen Aufgaben unter Beweis gestellt, doch sie scheitern oft, wenn es um Interpretierbarkeit und Legalität geht. Im Gegensatz dazu haben regelbasierte Ansätze, die zusätzliche Erklärbarkeit und Nachvollziehbarkeit bieten, Schwierigkeiten, mit verrauschten, unsicheren und großen Daten in hochdynamischen Verkehrsszenen in einer Vielzahl von Szenarien umzugehen.
- Open Access
- 11 September 2024
- Published by Springer Nature
- p. C1
Correction to: Legally-Guided Automated Decision-Making System Using Language Model Agents for Autonomous Driving
- 11 September 2024
- Published by Springer Nature
- p. C1
Open AccessCorrection to: Legally-Guided Automated Decision-Making System Using Language Model Agents for Autonomous Driving
- 11 September 2024
- Published by Springer Nature
- p. 234-248
Legally-Guided Automated Decision-Making System Using Language Model Agents for Autonomous Driving
- 11 September 2024
- Published by Springer Nature
- p. 234-248
Legally-Guided Automated Decision-Making System Using Language Model Agents for Autonomous Driving
- 15 October 2023
- Published by Springer Nature
- p. 238-253
Extracting Interpretable Hierarchical Rules from Deep Neural Networks’ Latent Space
- 15 October 2023
- Published by Springer Nature
- p. 238-253
Extracting Interpretable Hierarchical Rules from Deep Neural Networks’ Latent Space