logic_net_toy
A useful first reproduction for examining the rule-to-teacher-to-student interface and keeping the distillation step explicit.
Reproduction Projects
Small-scale implementations used to test and refine my understanding of paper-level mechanisms
Each project page isolates one mechanism, one suggested reading order, and one implementation question that is worth tracking in code.
semantic_loss_toy
A direct constraint-on-distribution view in which knowledge is interpreted as probability mass over satisfying worlds rather than through a teacher model.
Open project pageknowledge_landmarks_toy
A softer structured-prior setting in which sparse local data may be regularized by global knowledge landmarks rather than precise symbolic rules.
Open project page