SondreWold / ltn-experiments

Approaching the GLUE Benchmark using Logic Tensor Networks

Approaching the GLUE benchmark with Logic Tensor Networks

Comparing standard CrossEntropy based training on the GLUE tasks to a LTN based approach using the SatAgg objective.

Why?

Why not?

Datasets

MNLI, QNLI, SST-2, WNLI, RTE, CoLA

Representing the tasks as First Order Logic

TOOD

Comparison with baseline models

TODO

Lessons learned

TODO