Names Don't Matter: Symbol-Invariant Transformer for Open-Vocabulary Learning
This addresses a fundamental limitation in AI for handling unseen symbols, though it is incremental as it builds on existing Transformer frameworks.
The paper tackled the problem of neural architectures struggling with interchangeable tokens by proposing a Transformer-based mechanism that is provably invariant to token renaming, resulting in substantial performance gains on open-vocabulary tasks.
Current neural architectures lack a principled way to handle interchangeable tokens, i.e., symbols that are semantically equivalent yet distinguishable, such as bound variables. As a result, models trained on fixed vocabularies often struggle to generalize to unseen symbols, even when the underlying semantics remain unchanged. We propose a novel Transformer-based mechanism that is provably invariant to the renaming of interchangeable tokens. Our approach employs parallel embedding streams to isolate the contribution of each interchangeable token in the input, combined with an aggregated attention mechanism that enables structured information sharing across streams. Experimental results confirm the theoretical guarantees of our method and demonstrate substantial performance gains on open-vocabulary tasks that require generalization to novel symbols.