A Perspective on Artificial Intelligence Patent Applications


Whether or not you have a patent application in the works, you should be aware of the implications of this technology. In this article, we will discuss some of the most common issues that you will face when developing and implementing Artificial Intelligence (AI) applications. We will also explore the various ways in which you can protect your intellectual property and your business.

Artificial Intelligence Patent Applications

Whether you are developing an AI/ML-based invention or trying to obtain patent protection for an existing application, there are a number of important considerations you should keep in mind. These include separating the training phase from the execution phase, providing a patentable hook, and identifying the elements of the structure that are patentable.

For a machine learning model to be patented, it must meet the technical character test. This requires a substantive correlation between the abstract conceptual elements and the technical elements. For example, an unsupervised learning model may require a processing unit to perform complex calculations. It also must be deployed in a commercial setting.

The United States Supreme Court has determined that AI-based inventions that employ computer processes are eligible for patent protection. However, this determination is only applicable to those inventions that meet certain requirements.

For instance, a neural network classifier system uses physical data to train a classifier. It can be trained using a training dataset or through back propagation. The training process is typically carried out by a computer program.

Although the patent specification is generally broken into major components of the model, you may need to break it down further. This can help to prevent split infringement issues.

In addition to describing the method or procedure in detail, you should also provide a detailed description of the main embodiment. This may include a physical device such as a computer, or a special hardware used in the training and execution phases.

In addition, you may want to incorporate as much physical structure as possible into your invention. This can obviate some of the issues associated with 35 U.S.C. SS101.

Detecting Images on an Ongoing Basis

Detecting Images on an Ongoing Basis in Artificial Intelligence Patent Applications is a growing trend. The advent of digital pathology and AI have aided in the development of faster and more accurate diagnoses. However, machine interpretation of medical images requires an enormous amount of data-derived knowledge. Fortunately, some companies have come up with systems to detect life-threatening abnormalities.

The best way to identify the good from the bad is to use a computer to compare two images. This can be done on a local network, or downloaded to a cloud-based computer. In the latter case, an intelligent system may automatically add indicators to subsequent acquired images based on earlier acquired annotated images.

The CWRU method uses handcrafted features to calculate the likelihood of mitosis patches in breast cancer pathology images. Similarly, a pixel classifier in Definiens GmbH’s patent is trained to recognize blurred regions in digital images of stained tissue. This pixel classifier is configured to compare pixel values with nearby pixels at offsets.

The Ohio State Innovation Foundation suggested a small, simple, inexpensive technique called thresholding. A set of images was divided into zones based on a scale factor, and each zone was used to determine the most important features. This resulting data set can be used to augment the original set to address poorly performing areas.

A similar set of images was also resampled to create a tessellation. In addition, the Regents of University of California came up with a clever and nifty bit of kit to find the optimal image. This tessellation was based on the same sliding window concept as mentioned above. The tessellation is then used to create a 3D registration of examinations.

Machine Learning

Creating a patentable AI/ML-based invention can be expensive and time-consuming. A company that wants to protect its key products may want to consider structuring the application with special considerations. The goal is to come up with an inventive and commercially viable invention that will stand the test of time.

A machine learning model can be an effective way to reduce the amount of prior art search required. It also can help improve task performance. However, the most important challenge is explainability. While a human domain expert would probably be able to tell you exactly how to do the above, a machine may not.

A machine-learning-based invention is also an excellent way to identify areas that need improvement. A system can automatically identify these areas and augment the specimen data set to address them. This can be done without interfering with a company’s ability to prevent direct infringement.

A more fundamental AI invention would be to develop improved models. These may be the real key to obtaining patent protection.

The best way to demonstrate the benefits of your core-AI invention is to include a complete description of your invention. This includes the structure, functions, and details required to obtain commercially useful protection.

The most important part of your AI/ML-based invention is to describe in detail what it does and what it does not do. It should also provide a concise and informative explanation of the hardware requirements. This is especially true for a machine-learning-based invention.

In the end, an AI/ML-based invention should be a customer driver for the company. It should be a novel approach within the industry. The invention must also be useful to customers and internally at the company.

Searching the Prior Art

Performing an effective prior art search is an important element of patent prosecution. There are several factors that can impact the success of such a task.

The most important is defining the search statement. This involves a clear understanding of the scope of the application and the novelty of the claimed invention.

There are many techniques and strategies to achieve this goal. One common method is the use of an Examiner Automated Search Tool. It is a web-based system that is used to search for patents.

Another approach is to apply machine learning to the prior art search. This strategy can be self-validating, which is helpful in a time-sensitive search.

AI-based prior art search assistance may also reduce the cost of prosecution. This strategy could also help overcome the limitations of traditional keyword searches.

An experimental protocol and platform for testing these techniques is also proposed. Interestingly, these techniques have not been formally vetted by the Patent Office. A significant amount of empirical evidence indicates that invention occurs through recombination of the prior art. However, this method is likely to be flawed.

A new AI-based patent track model is needed. This model would not only revolutionize many ambiguous elements of patent law, it could maintain the incentive for innovative technologies to be patented. The model would include expedited examination, protection for creative AI systems, and shorter patent lifetimes. It will also incorporate a depository requirement for AI working models.

The new model will also require the Patent Office to take advantage of machine learning. Specifically, it will be required to depose AI working models as part of the patent filing process. This will increase the transparency of AI-generated inventions and provide a basis for evaluating the utility of an AI solution.

Inventing AI

Obtaining a patent for artificial intelligence is a difficult process. Many factors have to be considered. A key aspect is the structure of your invention.

AI/ML-based inventions are becoming more common in every area of technology. They include robotics, integrated circuit design, health care and other applications. They can also be directed to mathematical theories.

An AI/ML-based invention can be a useful tool for companies to use internally and to help provide recommendations and other valuable services to customers. Whether you’re building your own prototype or licensing a third-party program, it’s important to understand how to draft your application to protect your intellectual property.

To be a good patent applicant, you’ll need to describe your invention in detail. The patent specification should be broken into two main sections: the training phase and the execution phase. This will help you avoid split infringement issues.

The training phase consists of several steps, including training neural networks. These are computationally expensive models. They involve vast amounts of data and are often black-boxed. To avoid splitting infringement issues, separate the training phase from the execution phase.

In addition to the processing structure of a model, you’ll need to discuss the storage and input structures. This is particularly important with unsupervised learning models, which are used for anomaly detection, association rule learning and other tasks.

An AI/ML-based invention should incorporate as much physical structure as possible. This can obviate problems with 35 U.S.C. SS 101. It can also address issues with court rulings that have held that a computer’s structure can be a part of the functions performed by the device.

You’ll need to identify a use case to describe your core-AI invention. This will ensure that you get commercially useful protection for your invention.