AI Times published patent attorney Yongduck Kim's column on AI technology trends and patent protection. He works on AI, smart-factory and blockchain IP matters for clients including LG Electronics, Samsung Electronics, Baidu, SUALAB and MakinaRocks.
Read the column
"What can be protected by patent in the field of artificial intelligence?"
In general, companies that develop AI technologies often develop technologies across a whole area rather than a specific area. However, in terms of securing a patent, securing one patent for a technology across the entire area may result in an excessively narrow scope of rights. Therefore, it is good to have a patent on each of the specific areas.
A filing focused only on where AI is used may overlook the technical contributions made during model development.
However, if you only have a patent for the application, the problem is that the key technologies that take the most time and effort to develop AI models are not properly protected by patents. So, how do you secure an AI patent? As a developer, you need to think about what happens when you develop an AI model.
Potential patent areas include data collection; preprocessing such as deriving data attributes, augmentation, generalization and noise filtering; defining and training models, including training environments, model implementations and hyperparameter optimization; physical implementations such as neural-network devices and circuits; network architecture; and inference, including processing and using model outputs. Protecting technically substantive improvements across these areas can support a coherent patent portfolio and help demonstrate technical capability.
- Excerpt from the column
This article reflects the information available when it was published. Contact us to discuss your circumstances.