Research
Putting graph neural networks inside exact solvers since 2017, so results stay provably optimal rather than merely good enough.
Learning-guided exact solvers
2017 – presentOptimal planners are exact but scale badly; heuristics are fast but abandon guarantees. My doctoral work put a graph neural network inside the search, predicting which branches matter in constrained path-planning and in disjunctive temporal networks with uncertainty. This preserved optimality while cutting the explored tree substantially.
That verification machinery, generating instances at a chosen difficulty with a certified optimum, was as valuable as the speedup itself.
AAAI 2022Tree search and GNNs for DTNU controllability
IROS 2019GCNs and optimized tree search for constrained path-planning
Thesis 2021Learning off-road maneuver plans for autonomous vehicles
PatentMethod for defining a path, EP 3953662 / US 11846513