Patent Protection On AI Inventions – Intellectual Property – United States – Mondaq News Alerts

31 August 2021

Sheppard Mullin Richter & Hampton

To print this article, all you need is to be registered or login on Mondaq.com.

In recent years, AI patent activity has exponentially increased.The figure below shows the volume of public AI patent applicationscategorized by AI component in the U.S. from 1990-2018. The eightAI components in FIG. 1 are defined inan article published in 2020by theUSPTO. Most of the AI components have experienced explosive growthin the past decade, especially in the areas of planning/control andknowledge processing (e.g., using big data in automatedsystems).

Figure 1. AI patent activities byyear

AI technology is complex and includes different parts acrossdifferent fields. Inventors and patent attorneys often face thechallenge of effectively protecting new AI technology development.The rule of thumb is to focus the patent protection on what theinventors improve over the conventional technology. However,inventors often need to improve various aspects of an existing AIsystem to make it fit and work for their applications. In thefollowing sections, we will discuss an illustrative list of subjectareas that may offer patentable AI inventions.

The training phase of an AI system includes most of the excitingtechnical aspects of machine learning algorithms exploring thelatent patterns embedded in the training data. A typical trainingprocess includes preparing training data, transforming the trainingdata to facilitate the training process, feeding the training datato a machine learning model, fitting (training) the machinelearning model based on the training data, testing the trainedmachine learning model, and so on. Different AI models or machinelearning models may have different training processes, such assupervised training based on labeled training data, unsupervisedtraining that infers a function to describe a hidden structure fromunlabeled training data, semi-supervised training based onpartially-labeled training data, reinforcement learning (RL), etc.Common areas in the training phase that may yieldpatent-protectable ideas include:

The application phase of an AI system includes applying thetrained models to make predictions, inferences, classifications,etc. This phase generally covers the real application of the AIsystem. It can provide easier infringement detectability and thusvaluable patent protection for the AI system. In this digital era,AI systems can be applied to almost every aspect of our life. Forexample, an AI patent can claim or describe how the AI system helpsthe user to make better decisions or perform previously impossibletasks. These applications may be deemed as practical applicationsthat are powerful in overcoming potential "abstract idea"rejections during the prosecution of the AI patent.

On the other hand, simply claiming an AI system as a magicalblack box that generates accurate predictions based on input datawill likely trigger rejections during prosecution, such aspatentable subject matter rejections (e.g., a simple application ofthe black box may be categorized as human activities). There arevarious ways to reduce the chances of getting such rejections. Forexample, adding a brief description of the training process or themachine learning model structure helps overcome U.S.C. 101rejections.

Another flavor of AI patents is related to accelerators,hardware pieces with built-in software logic accelerating trainingand/or inferencing process. These AI patents may be claimed fromeither a software perspective or hardware perspective. Someexamples include specially designed hardware to improve trainingefficiency by working with GPU/TPU/NPU/xPU (e.g., by reducing datamigrations among different components/units), memory layout changesto improve the computational efficiency of computing-intensivesteps, arrangement of processing units for easy data sharing, andefficient parallel training (e.g., segmenting tensors to evenlydistribute workloads to processors), an architecture that fullyexploits the sparsity of tensors to improve computationefficiency.

The state-of-art AI systems are far from perfection. Robustness,safety, reliability, data privacy, are just some of the mostnoticeable pain points in training and deploying AI systems. Forexample, an AI model trained from a first domain may havenear-perfect accuracy for inferencing in the first domain, butgenerate disastrous inferences when being deployed in a seconddomain, even though the domains share some similarities. Therefore,how to train an AI model efficiently and adaptively so that it isrobust when being deployed in all domains of interest is bothchallenging and intriguing.

As another example, AI systems trained based on training datamay be easily fooled by adversarial attacks. For instance, asecond deep neural network may be designed to compete against thefirst one to identify its weaknesses. The safety and reliability ofsuch AI systems will be critical in the coming years and may beimportant patentable subject matters.

As yet another example, training data in many cases may includesensitive data (e.g., customer data), directly using such trainingdata may result in serious data privacy breaches. This problembecomes more alarming when a plurality of entities collectivelytrain a model using their own training data. Accordingly,researchers and engineers have been exploring differential privacyprotection and federated learning to address these issues.

The content of this article is intended to provide a generalguide to the subject matter. Specialist advice should be soughtabout your specific circumstances.

POPULAR ARTICLES ON: Intellectual Property from United States

Obhan & Associates

Trademarks Comparative Guide for the jurisdiction of India, check out our comparative guides section to compare across multiple countries

Haug Partners

In a June 11, 2021 decision, Yu v. Apple Inc., a Federal Circuit panel issued a precedential decision, with a dissent, upholding the invalidation of patent claims to a digital camera on a motion to dismiss.

Brown Rudnick LLP

There are two dates that must be specified in every trademark/service mark application. They are, the date of first use anywhere and the date of first use in commerce.

Original post:

Patent Protection On AI Inventions - Intellectual Property - United States - Mondaq News Alerts