5 papers
Breaking Predictions Is Not Enough: Specified-Foil Counterfactuals for Temporal Graphs
Minwoo Yu, Young-guk Ha
Temporal graph counterfactual explanations typically change past events to change or invalidate an original prediction, while leaving its replacement unspecified. Yet a user facing…
LiFTER: A Grounded Neuro-Symbolic Microscope for Continuous-Time Dynamic Graph Forecasting
Minwoo Yu, Young-guk Ha
Continuous-time dynamic graph models predict future links by compressing past interactions into neural states. Although effective for forecasting, this computation obscures which e…
Back to All-Entity Ranking: Sampler-Dependent Evaluation in Continuous-Time Dynamic Graphs
Minwoo Yu, Young-guk Ha
Next-destination prediction in continuous-time dynamic graphs (CTDGs) commonly ranks an observed interaction against sampled negative destinations. The resulting score is condition…
What Softmax Throws Away: Mass-Aware Attention for Evidence Accumulation
Minwoo Yu, Young-guk Ha
High task performance does not show whether a model retains prediction-relevant structural information in its internal representation. Temporal graph models, for example, can achie…
Relevance Is Not Permission: Localizing and Controlling Metric-Facing Attention Contributions
Minwoo Yu, Young-guk Ha
Attention identifies items relevant to a current query, but does not separately determine whether their value contributions support the prediction. We propose Warrant, a unified me…