Learned Search PruningQuadtree DecompositionMixture of Experts
Can a learned model shrink the area a grid search has to explore without giving up optimality? A small CNN over a quadtree decomposition predicts, level by level, which tiles can contain the optimal path, using one specialist network per map type and batched inference. A hybrid mode verifies the learned corridor with algebraic transfer matrices so the result stays optimal. Training uses teacher forcing followed by adversarial hard-example mining. Across five map families (random scatter, DFS maze, spiral, recursive division, rooms and corridors) the predicted corridor typically covers 8–15% of the grid while still containing the optimal path. Neural-only mode is fast but not guaranteed optimal; hybrid and matrix-only modes are. The mixture of experts needs the map type to pick a specialist, and a full runtime comparison against classical search is still open.
IMPACT plans through clutter by considering both which objects tolerate contact and how pushing them may affect surrounding objects. It combines VLM-derived object costs with sampled push outcomes to construct a direction-aware costmap, then uses contact-aware A* to plan through changes in object positions. My contribution focused on formulating the directional-contact representation and implementing the planner.
A generative deep learning approach that optimizes spectral response functions for material classification, moving beyond the conventional three-cone human vision model. Inspired by animal vision, it adapts spectral channels to environmental factors for more accurate, efficient material discrimination. Published patent application EP4339901A1.
A deep learning approach to methane leak detection from hyperspectral imagery. Meta-learning with one-shot learning compares unknown spectral signatures against known methane profiles; evaluation on AVIRIS data from the Aliso Canyon incident shows clear gains in accuracy and speed over traditional methods.
A 5x5 Go agent built for USC’s Foundations of Artificial Intelligence course, combining minimax search with alpha-beta pruning and a Q-learning component, with a tournament simulator and a small GUI for manual and automated play. Scored 95.2/100, in the top 5 of about 300 students.
