Linus Zheng
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Linus Zheng has a total of 9 co-authors. Here are the top co-authors by number of shared publications:
- Isil Dillig
University of Texas at Austin, Austin, USA
2 shared publications - Dixin Tang
University of Texas at Austin, Austin, USA
1 shared publication - Jarrett Holtz
Robert Bosch LLC., Budapest, Hungary
1 shared publication - Jiayi Wei
University of Texas at Austin, Austin, TX, USA
1 shared publication - Jimmy Xin
University of Texas at Austin, Austin, USA
1 shared publication - Joydeep Biswas
University of Texas at Austin
1 shared publication
Latest Publications
- Open Access
- 9 October 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Programming Languages
- Vol. 9 (OOPSLA2) , 615-641
Homomorphism Calculus for User-Defined Aggregations
Data processing frameworks like Apache Spark and Flink provide built-in support for user-defined aggregation functions (UDAFs), enabling the integration of domain-specific logic. However, for these frameworks to support efficient UDAF execution, the function needs to satisfy a homomorphism property , which ensures that partial results from independent computations can be merged correctly. Motivated by this problem, this paper introduces a novel homomorphism calculus that can both verify and refute whether a UDAF is a dataframe homomorphism. If so, our calculus also enables the construction of a corresponding merge operator which can be used for incremental computation and parallel execution. We have implemented an algorithm based on our proposed calculus and evaluate it on real-world UDAFs, demonstrating that our approach significantly outperforms two leading synthesizers.
- 9 October 2025
- Published by Association for Computing Machinery (ACM) in Proceedings of the ACM on Programming Languages
- Vol. 9 (OOPSLA2) , 615-641
Open AccessHomomorphism Calculus for User-Defined Aggregations
Data processing frameworks like Apache Spark and Flink provide built-in support for user-defined aggregation functions (UDAFs), enabling the integration of domain-specific logic. However, for these frameworks to support efficient UDAF execution, the function needs to satisfy a homomorphism property , which ensures that partial results from independent computations can be merged correctly. Motivated by this problem, this paper introduces a novel homomorphism calculus that can both verify and refute whether a UDAF is a dataframe homomorphism. If so, our calculus also enables the construction of a corresponding merge operator which can be used for incremental computation and parallel execution. We have implemented an algorithm based on our proposed calculus and evaluate it on real-world UDAFs, demonstrating that our approach significantly outperforms two leading synthesizers.
- 5 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 9 (6) , 4894-4901
Programmatic Imitation Learning From Unlabeled and Noisy Demonstrations
Imitation Learning (IL) is a promising paradigm for teaching robots to perform novel tasks using demonstrations. Most existing approaches for IL utilize neural networks (NN), however, these methods suffer from several well-known limitations: they 1) require large amounts of training data, 2) are hard to interpret, and 3) are hard to refine and adapt. There is an emerging interest in Programmatic Imitation Learning (PIL), which offers significant promise in addressing the above limitations. In PIL, the learned policy is represented in a programming language, making it amenable to interpretation and adaptation to novel settings. However, state-of-the-art PIL algorithms assume access to action labels and struggle to learn from noisy real-world demonstrations. In this paper, we propose PLUNDER, a novel PIL algorithm that addresses these shortcomings by synthesizing probabilistic programmatic policies that are particularly well-suited for modeling the uncertainties inherent in real-world demonstrations. Our approach leverages an EM loop to simultaneously infer the missing action labels and the most likely probabilistic policy. We benchmark PLUNDER against several established IL techniques, and demonstrate its superiority across five challenging imitation learning tasks under noise. PLUNDER policies outperform the next-best baseline by 19% and 17% in matching the given demonstrations and successfully completing the tasks, respectively.
- 5 April 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Robotics and Automation Letters
- Vol. 9 (6) , 4894-4901
Programmatic Imitation Learning From Unlabeled and Noisy Demonstrations
Imitation Learning (IL) is a promising paradigm for teaching robots to perform novel tasks using demonstrations. Most existing approaches for IL utilize neural networks (NN), however, these methods suffer from several well-known limitations: they 1) require large amounts of training data, 2) are hard to interpret, and 3) are hard to refine and adapt. There is an emerging interest in Programmatic Imitation Learning (PIL), which offers significant promise in addressing the above limitations. In PIL, the learned policy is represented in a programming language, making it amenable to interpretation and adaptation to novel settings. However, state-of-the-art PIL algorithms assume access to action labels and struggle to learn from noisy real-world demonstrations. In this paper, we propose PLUNDER, a novel PIL algorithm that addresses these shortcomings by synthesizing probabilistic programmatic policies that are particularly well-suited for modeling the uncertainties inherent in real-world demonstrations. Our approach leverages an EM loop to simultaneously infer the missing action labels and the most likely probabilistic policy. We benchmark PLUNDER against several established IL techniques, and demonstrate its superiority across five challenging imitation learning tasks under noise. PLUNDER policies outperform the next-best baseline by 19% and 17% in matching the given demonstrations and successfully completing the tasks, respectively.