Considerations In Protecting Inventions Relating to Modeling, Neural Networks, and Artificial Intelligence
Use of modeling, neural networks, and artificial intelligence today is exploding. Modeling relates to the use of a statistical (software) tool that analyzes historical data to forecast future outcomes based on patterns and trends with the data. Generally, the steps of creating a predictive model include gathering the data, preparing/preprocessing the data (cleanse and prepare it for use), selecting/determining a particular modeling technique, building/configuring and training the model (e.g., select variables, select outputs), validating the model, evaluate the model, deploying it for use, and then monitoring and optimizing its performance. Machine-learning algorithms are often used to improve the accuracy of the model.[i]
Our office has helped many global clients in protecting their valuable rights to new inventions and products relating to this exciting technology. Some of the more frequent questions we get from inventors in this area is “…I know my idea is new and valuable but I am having trouble identifying what is patentable about my new invention to modeling and how I can protect my rights in it?” This article addresses some of the issues and answers to these questions.
The good news is that inventions relating to modeling, neural networks, and AI are patentable and can be protected in various and often, multiple ways. The remaining sections of this article below, discuss different ways that an invention in this area can be protected.
HOW TO PROTECT (CLAIM) INVENTIONS RELATED TO MODELING, NEURAL NETWORKS, OR ARTIFICIAL INTELLIGENCE (AI):
Based on our experience, there are four (4) main ways to protect an invention relating to modeling/AI: by 1. protecting a new way of creating a predictive model or neural network; 2. protecting a new and specific function or practical application of using a model or neural network; 3. protecting a new way of generating a particular graphical user interface (GUI) to implement the input/output of a model or neural network to efficiently present data to a user; or, 4. use a combination of the above ways to protect the modeling invention if more ways apply to the particular invention. Each of these ways is discussed in more detail in the sections below.
1. Protect new ways of creating a predictive model or neural network:
a. Protect new ways of preprocessing data before it is input into a general deep learning model or neural network (the model is only good as the data going into it). For example, protect new ways to clean, standardize, normalize, extract, transform or encode data so that it may be more easily parsed by the model or for the purpose of obtaining better or more accurate results. See CN112800053B (Chinese Patent Publication to method and device for preprocessing data to extract particular sample data for input into a deep neural network).
b. Protect new ways of training the model or neural network. See e.g., U.S. Patent Publication No. 2019/0385055 to Sim; CN113837380A (Chinese patent publication to a method of training a neural network based on biological self-organization back propagation).
c. Protect new ways of selecting a predictive modeling technique. For example, protect a new process of testing and determining the appropriate model technique to deploy in a particular circumstance. See e.g., CN101782976B (Chinese patent publication to an automatic selection method for machine learning in a cloud computer environment); US Patent No. 11,775,850 to Campos et al.
d. Protect new ways of building/configuring and training the model. See e.g. US Patent No. 12,014,268 to Oh et al. (patenting a batch normalization layer training method used in a neural network learning system having limited operational processing capability and storage space).
e. Protect new ways of optimizing the performance of the model. CN113837380A (Chinese patent publication to optimizing the neural network once according to a first optimized weight parameter and a second optimized weight parameter, and repeating the optimization for multiple times until the loss function of the neural network converges to a preset value).
and/or
f. or protect a combination of any of the above.
2. Protect a new and specific practical application of using a model or neural network. Another way of protecting modeling inventions is to protect a novel way in which a model is practically applied. For example, a patent claim can be drafted describing the details of “how” the application works in a particular way to obtain a practical result, as opposed to just reciting general, desired outcomes. In particular, for example, specific inputs and outputs of the model may be claimed. The following example claim recites a specific application of a neural network (detecting network anomalies to determine and drop a malicious network packet) generated using a particular training algorithm.
3. Protect a new way of generating a particular graphical user interface (GUI) to implement the input/output of a model or neural network. The hypothetical example below was approved as patent-eligible subject matter by the US Patent Office.
4. Use a combination of the above ways to protect the modeling invention if more ways apply to the particular invention. For example, Example 47 discussed above recites a novel way of training a neural network combined with a new practical application of the same neural network.
Once an inventor develops the details of the invention, it will become more apparent on how to best protect (claim) it. For example, a discussion should be conducted between the inventors, marketing and legal departments to identify all the new features of the model that provide advantages/improvements over the prior art technologies. By identifying the improvements/advantages in the context of each of the four main ways of protecting modeling inventions discussed above, smart decisions can be made as to where the improvements are, what is the value of those identified improvements, and the best ways to move forward to protect rights in those identified improvements.
Our office has worked closely with many inventors, engineers, and computer programmers to identify and protect their valuable inventions in this space. If you have any questions, please contact any of the attorneys at our firm and we would be happy to help you.
[i] A neural network is a type of machine learning model that uses interconnected nodes to process data and make decisions. Modeling and neural networks have been applied in many fields and to many applications, including image recognition, speech recognition, natural language processing, etc.
