Implicit Training of Energy Model for Structure Prediction
This work addresses a bottleneck in structure prediction for machine learning researchers, offering an incremental improvement over existing inference network methods.
The paper tackles the problem of training objectives not aligning with evaluation metrics in complex structured output prediction by proposing implicit-gradient techniques to learn dynamic loss objectives parameterized by an energy model, resulting in improved model performance.
Most deep learning research has focused on developing new model and training procedures. On the other hand the training objective has usually been restricted to combinations of standard losses. When the objective aligns well with the evaluation metric, this is not a major issue. However when dealing with complex structured outputs, the ideal objective can be hard to optimize and the efficacy of usual objectives as a proxy for the true objective can be questionable. In this work, we argue that the existing inference network based structure prediction methods ( Tu and Gimpel 2018; Tu, Pang, and Gimpel 2020) are indirectly learning to optimize a dynamic loss objective parameterized by the energy model. We then explore using implicit-gradient based technique to learn the corresponding dynamic objectives. Our experiments show that implicitly learning a dynamic loss landscape is an effective method for improving model performance in structure prediction.