Google patents method to improve AI search accuracy
This filing describes a computer system that uses two linked neural networks to match questions with answers. It improves the training process by using a specific mathematical rule to help the system better distinguish between correct and incorrect matches.
In the filing’s own words
The technology provides neural network models having dual encoder architectures. For instance, a token embedder layer section of a dual encoder is associated with a first input and a second input and generates token embeddings ( 2002). An encoder layer section receives the token embeddings from the token embedder layer and generates encodings based on the token embeddings ( 2004). A projection layer receives the encodings from the encoder section and generates a set of projections ( 2006). The projection layer is shared by the dual encoders (FIG. 4 C). An embedding space generates, based on the set of projections, a question embedding and a response embedding ( 2008). These question and response embeddings are used in identifying a set of candidate responses to an input question ( 2008). The embedding space is configured during training according to a contrastive loss that applies same tower negatives to generate the question and response embeddings ( 2010).
What exactly they claim (claim 1)
1. A computer-implemented machine learning dual encoder system, comprising: a token embedder layer section associated with a first input and a second input and configured to generate token embeddings; an encoder layer section configured to receive the token embeddings from the token embedder layer and configured to generate encodings based on the token embeddings; a projection layer configured to receive the encodings from the encoder section and to generate a set of projections, wherein the projection layer is shared by the dual encoder system; and an embedding space configured, based on the set of projections, to generate a question embedding and a response embedding, the question and response embeddings for use in identifying a set of candidate responses to an input question; wherein the embedding space is configured during training according to a contrastive loss that applies same tower negatives to generate the question and response embeddings.
Facts
- Company
- Google (Google LLC)
- Official title
- IMPROVING CONTRASTIVE LOSS FOR DUAL ENCODER RETRIEVAL MODELS WITH SAME TOWER NEGATIVES
- Publication number
- US20260259794A1
- Application number
- 19162123
- Filed
- May 30, 2023
- Published
- Sep 3, 2026
- Inventors
- 7
- Claims
- 20
Classified under
G06F 11/1004 Physics & computingMore from Google in AI & software
- Google patents video search that answers questions using AI descriptions Sep 15, 2026
- Google patents photo tool that removes people and their gear Sep 15, 2026
- Google patents AI that writes live meeting notes and summaries Sep 15, 2026
- Google patents AI tool that turns messy emergency notes into clear messages Sep 3, 2026
- Google patents AI that clicks through apps to finish tasks for you Sep 24, 2026
Summaries are simplified from the filing’s own abstract and first claim. A patent filing is not a product announcement.
Patent data and drawings: U.S. Patent and Trademark Office (USPTO). Nebula is not affiliated with or endorsed by the USPTO.