Mitsui Chemicals Co., Ltd. (Headquarters: Chuo-ku, Tokyo; President: Osamu Hashimoto), in collaboration with blueqat (Blue Cat) Co., Ltd. (Headquarters: Shibuya-ku, Tokyo; President: Yuichiro Minato), has begun an initiative combining advanced technologies of natural language processing and quantum computation to accelerate the discovery of new applications for Mitsui Chemicals products.
This initiative utilizes a natural language processing model called Simple Contrastive Learning of Sentence Embeddings (simCSE). SimCSE models are vector expressions that capture the meaning of documents and sentences*1It is a high-performance model that generates highly accurate semantic similarity between documents. Applications to information retrieval and question answering systems are expected.
This time, both companies carried out preliminary learning of the SimCSE model based on the patent database, and furthermore, tensor networks*2By applying technology, we succeeded in efficiently compressing SimCSE model information. Examples of applications of this model include application to patent searches and new application search fields.
[Examples of use]
Furthermore, machine learning models compressed with tensor networks are compatible with quantum circuits, and can be expected to be executed on quantum computers in the future.
Going forward, Mitsui Chemicals will continue to take on the challenge of promoting DX by combining machine learning technology such as natural language models and quantum technology for the utilization of large-scale language models.
■blueqat official website:https://blueqat.com/
*1 Vector expression: A method for expressing the meaning of words and sentences in natural language processing with numeric vectors (multi-dimensional arrays).
*2 Tensor network: A method for efficiently representing huge tensors (multidimensional data) by decomposing them. It is a technology that has been used for information compression and quantum simulation of quantum circuits in the quantum computation field.
over