All our email inboxes receive many emails daily. A part of them, however, can be configured as spam: therefore, it is necessary that i dedicated services are able to find this unwanted content to better filter the incoming mail. This is a necessary step, especially to guarantee users greater security.
In this sense, Gmail has received a major update. Google, in fact, declared that the technology used favored a 38% improvement in text identification. Specifically, the Mountain View giant has focused on RETVec (Resilient & Efficient Text Vectorizer), a new approach to text vectorization, which guarantees better performance while reducing calculation costs (created by Google Research).
Going into the details of the news, RETVec allowed a 38% more effective spam recognition, reducing the false positive rate by 19.4% and the use of the Tensor Processing Unit (TPU) by as much as 83%. Google said that these improvements were made possible by its lighter word embedding model – with 200,000 parameters – which allowed a reduction in the size of the Transformer model, which guarantees the same, if not better, performance.
Among other things, there is the possibility of dividing the computing task between host and TPU in an efficient manner. Finally, Google specified that it supports everything “every language and every UTF-8 character without the need for text preprocessing.” Therefore, the system can be used both on mobile devices and via the web.
