Explainable AI and Data Science Lab

Root Detection

In collaboration with Prof. Shimon Rachmilevitch’s lab, we build deep-learning models to automatically detect and analyze plant roots from images and infer species and growth patterns, motivated by evidence that plants adjust below-ground strategies based on neighboring roots. We apply XAI to identify which root patterns drive predictions and compare these explanations with expert manual analyses to assess their alignment with established biological knowledge.

Forecasting Cruise Routes

The project develops vessel-route prediction models by building a novel dataset that fuses AIS trajectories with weather data, addressing a key limitation of current AIS-only approaches. We evaluate how much weather improves prediction accuracy and identify the minimal data volume and feature set needed to reach state-of-the-art performance for safer and more efficient maritime operations.

Explainable AI for Combinatorial Problems

The project uses explainable AI, centered on PCA-based dimension reduction, to uncover the low-dimensional concepts encoded in GNN latent spaces when solving NP-hard combinatorial problems such as SAT, graph coloring, and max-clique. By extracting and interpreting these emergent concepts, we compare the networks’ learned strategies to classical human heuristics (e.g., degree and support) to understand how and why GNNs make optimization decisions. 

Shoham, E., Cohen, H., Wattad, K., Rika, H., & Vilenchik, D. (2025). Concept learning for algorithmic reasoning: Insights from SAT-solving GNNs. Information Sciences, 122754.

Shoham, E., Rika, H., & Vilenchik, D. (2025). From Black Box to Algorithmic Insight: Explainable AI in Graph Neural Networks for Graph Coloring. AAAI 2025 Workshop NeurMAD.

Smart Social Junction Traffic Control

Our work develops a data-driven smart-junction system that applies machine learning and reinforcement learning to embed social priorities into traffic monitoring and adaptive signal control, enabling prioritized fast-lane flow while sustaining efficient service for all vehicles.

Barzilai, O., Rika, H., & Hassine, Y. (2025). Smart social junction traffic control using reinforcement learning on real data. Journal of Computational Social Science, 8(4), 83.

Barzilai, O., Rika, H., Voloch, N., Hajaj, M. M., Steiner, O. L., & Ahituv, N. (2023). Using Machine Learning Techniques to Incorporate Social Priorities in Traffic Monitoring in a Junction with a Fast Lane. Transport and Telecommunication, 24(1), 1-12.