Methods reminiscent of Gmail, YouTube and Google Play depend on textual content classification fashions to establish dangerous content material together with phishing assaults, inappropriate feedback, and scams. Some of these texts are more durable for machine studying fashions to categorise as a result of dangerous actors depend on adversarial textual content manipulations to actively try and evade the classifiers. For instance, they may use homoglyphs, invisible characters, and key phrase stuffing to bypass defenses. 

To assist make textual content classifiers extra sturdy and environment friendly, we’ve developed a novel, multilingual textual content vectorizer known as RETVec (Resilient & Environment friendly Textual content Vectorizer) that helps fashions obtain state-of-the-art classification efficiency and drastically reduces computational value. As we speak, we’re sharing how RETVec has been used to assist defend Gmail inboxes.

Strengthening the Gmail Spam Classifier with RETVec

Determine 1. RETVec-based Gmail Spam filter enhancements.

Over the previous yr, we battle-tested RETVec extensively inside Google to guage its usefulness and located it to be extremely efficient for safety and anti-abuse purposes. Particularly, changing the Gmail spam classifier’s earlier textual content vectorizer with RETVec allowed us to enhance the spam detection fee over the baseline by 38% and scale back the false optimistic fee by 19.4%. Moreover, utilizing RETVec lowered the TPU utilization of the mannequin by 83%, making the RETVec deployment one of many largest protection upgrades lately. RETVec achieves these enhancements by sporting a really light-weight phrase embedding mannequin (~200k parameters), permitting us to cut back the Transformer mannequin’s measurement at equal or higher efficiency, and being able to separate the computation between the host and TPU in a community and reminiscence environment friendly method.

RETVec Advantages

RETVec achieves these enhancements by combining a novel, highly-compact character encoder, an augmentation-driven coaching regime, and using metric studying. The structure particulars and benchmark evaluations can be found in our NeurIPS 2023 paper and we open-source RETVec on Github.

As a result of its novel structure, RETVec works out-of-the-box on each language and all UTF-8 characters with out the necessity for textual content preprocessing, making it the best candidate for on-device, internet, and large-scale textual content classification deployments. Fashions skilled with RETVec exhibit quicker inference pace as a result of its compact illustration. Having smaller fashions reduces computational prices and reduces latency, which is essential for large-scale purposes and on-device fashions.

Determine 1. RETVec structure diagram.

Fashions skilled with RETVec might be seamlessly transformed to TFLite for cellular and edge gadgets, because of a local implementation in TensorFlow Textual content. For internet software mannequin deployment, we offer a TensorflowJS layer implementation that’s accessible on Github and you may try a demo internet web page working a RETVec-based mannequin.

Determine 2.  Typo resilience of textual content classification fashions skilled from scratch utilizing totally different vectorizers.

RETVec is a novel open-source textual content vectorizer that means that you can construct extra resilient and environment friendly server-side and on-device textual content classifiers. The Gmail spam filter makes use of it to assist defend Gmail inboxes towards malicious emails.

If you want to make use of RETVec to your personal use circumstances or analysis, we created a tutorial that can assist you get began.

This analysis was carried out by Elie Bursztein, Marina Zhang, Owen Vallis, Xinyu Jia, and Alexey Kurakin. We wish to thank Gengxin Miao, Brunno Attorre, Venkat Sreepati, Lidor Avigad, Dan Givol, Rishabh Seth and Melvin Montenegro and all of the Googlers who contributed to the venture.