Ali Taheri
PhD Student at MPI-SWS
I am a PhD student in the HyCoDeV group at the Max Planck Institute for Software Systems (MPI-SWS), Kaiserslautern, Germany, under the supervision of Prof. Sadegh Soudjani. I hold a B.Sc. in Computer Engineering from Isfahan University of Technology (IUT), where I ranked 1st in Software Engineering Specialization and 3rd among all Computer Engineering peers.
Previously, I worked with the SPACE Lab at Nanyang Technological University (NTU) under Prof. Wanhua Li; the HyCoDeV group at MPI-SWS as a Research Intern under Prof. Sadegh Soudjani, in close collaboration with Prof. Ashutosh Trivedi; the Structured Machine Learning Lab at Simon Fraser University under Prof. Oliver Schulte; and the TMLR Group at Hong Kong Baptist University under Prof. Bo Han and Dr. Qizhou Wang.
Research Interests
Trustworthy Machine Learning, AI Safety, Formal Methods
Honors and Awards
- J2C Certification - Transactions on Machine Learning Research (TMLR) - 2026
- B.Sc. Rank: 1st/20 in Software Engineering Specialization, 3rd/89 among all CE students - 2025
- Highest Grade for Undergraduate Thesis Project - Isfahan University of Technology, 2025
- 1st Place in ICPC - Isfahan University of Technology, 2023
- 10th Place in the 46th ICPC West Asia Regional Contest - 2023
- 3rd Place in National Machine Learning Competition - 2022
- 1st Place in ICPC - Isfahan University of Technology, 2022
PhD in Computer Science [2026–Present]
Max Planck Institute for Software Systems (MPI-SWS)
Supervisor: Prof. Sadegh Soudjani
Bachelor of Computer Engineering [2021–2025]
Isfahan University of Technology (IUT)
GPA: 18.63/20, Ranked 1st in Software Engineering Specialization and 3rd overall
Research Experience
Developed LLM-guided agentic frameworks and benchmarks for barrier certificate synthesis in the safety verification of dynamical systems, and contributed to other work on formal methods and temporal logic.
Worked on multi-label event detection for the ROAD-R autonomous driving dataset, focusing on improving label relationship modeling through graph-based correlation learning
Proposed a forgetting mechanism for LLM fine-tuning that categorizes tokens into positive/negative sets via influence scores, learning from positive while unlearning negative tokens to improve generalization
Explored and analyzed the capabilities and challenges of both Large Language Models and Vision-Language Models. The work contributed to understanding how these models can be improved for trustworthy and reliable AI applications