Heuristic search
Bidirectional, bounded-suboptimal, memory- and parallel-aware algorithms, built on theory about which nodes any admissible search must expand, including the asymptotic cost of search itself.
Shahaf S. Shperberg
Senior Lecturer (Assistant Professor)
Stein Faculty of Computer and Information Science · Ben-Gurion University of the Negev
- Heuristic search
- Automated planning
- Metareasoning
- Reinforcement learning
- Multi-agent systems
I work on the decision-making abilities of autonomous agents: how an agent should search, plan, and learn when time, memory, and computation are all in short supply, and how it should reason about its own reasoning to spend those resources well.
Research
Read more
What we work on
Learning for search
We study reinforcement learning methods for constructing heuristics, the bias introduced by their training rules, and their behavior on instances beyond the training distribution.
Learning planning models
Classical planners need a domain model. We learn action models, including numeric effects, from observation, online, and alongside reinforcement learning.
Planning while the clock ticks
Metareasoning decides where the next unit of computation goes, when to commit to an action, and when to start executing a plan that is not finished yet.
Many agents, many objectives
Pareto-front search and the dominance checks that dominate its runtime, pathfinding under alternative cost measures, meeting problems, and multi-directional search.
Safe and fair RL
Shields learned from catastrophic effects, constrained policies, cost-aware world models, and fairness stated as a specification rather than a penalty weight.
LLM-based systems
Language models as an interface to planning tools and as a search problem in their own right. We also study what these systems retain after you ask them to forget.
Selected publications
All publications
Recent and representative work
-
2025
-
2024
-
2026
-
2026
-
2025
-
2024
-
2024
-
2022
-
2019
News
Recently
-
Elected co-program chair of ICAPS 2027, the International Conference on Automated Planning and Scheduling.
-
Serving as co-conference chair of SoCS 2026 and co-chair of PRL 2026 at ICAPS and IJCAI.
-
Named a Distinguished Senior Program Committee member of AAMAS 2026.
-
Two papers accepted to AAAI 2026, on bidirectional bounded-suboptimal search and on reinforcement learning of heuristics with limited-horizon search.
-
Joined the Stein Faculty of Computer and Information Science at Ben-Gurion University as a Senior Lecturer.
-
"Bridging theory and practice in bidirectional heuristic search with front-to-end consistent heuristics" published in the Artificial Intelligence Journal.
-
Received the BGU Rector's Glatt Award for excellence in teaching.
Contact
Get in touch
- shperbsh@bgu.ac.il
- Where
- Ben-Gurion University of the Negev
Be'er Sheva, Israel - Code
- SPL@BGU, our search, planning, and learning lab on GitHub
Students interested in search, planning, or reinforcement learning are welcome to write to me. See the group page for what my students are working on.