Concept潜在分词 / 隐式分词
Latent Tokenization
A text representation paradigm that does not rely on explicit discrete tokens, instead performing representation and modeling in a continuous latent space. It is explored as a potential alternative to conventional tokenizer-based approaches in language modeling.
Timeline (last 90 days)
Sep 30
The survey prospectively discusses new paradigms such as Latent Tokenization and Visual Tokenization that may replace traditional tokenizers.
Unverified50%