US Patent Application 18351397. Contrastive Sequence-to-Sequence Data Selector simplified abstract
Contents
Contrastive Sequence-to-Sequence Data Selector
Organization Name
Inventor(s)
Bowen Liang of Mountain View CA (US)
Macduff Hughes of Los Gatos CA (US)
Taro Watanabe of Mountain View CA (US)
Tetsuji Nakagawa of Tokyo (JP)
Alexander Rudnick of Mountain View CA (US)
Contrastive Sequence-to-Sequence Data Selector - A simplified explanation of the abstract
This abstract first appeared for US patent application 18351397 titled 'Contrastive Sequence-to-Sequence Data Selector
Simplified Explanation
The patent application describes a method for training a target model using a combination of data pairs and contrastive scores. Here are the key points:
- The method starts by generating a base model using a first dataset of data pairs.
- An adapted model is then generated by training the base model on a second dataset of data pairs.
- A third dataset of data pairs is used to determine a contrastive score for each pair.
- The contrastive score reflects the probability of quality for each data pair.
- Finally, a target model is trained using the data pairs from the third dataset and the corresponding contrastive scores.
Original Abstract Submitted
A method includes generating a base model by training with a first dataset of data pairs and generating an adapted model by training the base model on a second dataset of data pairs. The method also includes determining a contrastive score for each data pair of a third dataset of data pairs using the base model and the adapted model. The contrastive score is indicative of a probability of quality of the respective data pair. The method also includes training a target model using the data pairs of the third dataset and the contrastive scores.
