Adversarial Testing as a Tool for Interpretability: Length-based Overfitting of Elementary Functions in Transformers
This work addresses interpretability challenges in Transformers for sequence-to-sequence tasks, though it is incremental in nature.
The study investigated how Transformers overfit to sequence length and structure when learning elementary string edit functions, finding that generalization to shorter sequences is often possible but longer sequences are problematic, with models sometimes preferring structural aspects over algorithmic ones.
The Transformer model has a tendency to overfit various aspects of the training data, such as the overall sequence length. We study elementary string edit functions using a defined set of error indicators to interpret the behaviour of the sequence-to-sequence Transformer. We show that generalization to shorter sequences is often possible, but confirm that longer sequences are highly problematic, although partially correct answers are often obtained. Additionally, we find that other structural characteristics of the sequences, such as subsegment length, may be equally important. We hypothesize that the models learn algorithmic aspects of the tasks simultaneously with structural aspects but adhering to the structural aspects is unfortunately often preferred by Transformer when they come into conflict.