Learning methods

Abstract

Reduce the risk of rights infringement. [Solution] The learning method includes, in pre-training of a large-scale language model, an input step of inputting text data with clearly defined sources into the large-scale language model; an acquisition step of acquiring the values ​​of the intermediate layer of the large-scale language model when the large-scale language model performs pre-training with text data as input; and a learning step of generating a learning model by supervised learning using the acquired intermediate layer values ​​as input data and source information indicating the source of the text data as ground truth data.

Claims

1. In the pre-training of a large-scale language model, the input process involves inputting text data with clearly defined sources into the large-scale language model, The acquisition step of obtaining the values ​​of the intermediate layer of the large-scale language model when the large-scale language model performs pre-training using the text data as input, A learning process in which a learning model is generated by supervised learning using the acquired intermediate layer values ​​as input data and source information indicating the source of the text data as ground truth data, Learning methods that include this.

2. The aforementioned learning model is a learning model that estimates the source of the text data used by the aforementioned large-scale language model when generating the response. The learning method according to claim 1.

3. The aforementioned learning model is a learning model for a multi-label classifier. The learning method according to claim 1.

Images