To even the most casual observer of the tech startup space, machine learning (ML) is an incredibly promising and fast-growing sector. If you can’t see it, you’re either the ultimate contrarian or you’re not in the sector. ML seemingly offers endless opportunities for tech entrepreneurs, VC investors, and established innovators alike. However, protecting the inventions that stem from these innovative efforts with patents poses unique challenges, particularly in the crafting of machine learning patent claims.
The Anatomy of Patent Claims
At the heart of a patent are the patent claims, which are legally enforceable parts of the patent that define the patent owner’s rights. Patent claims function like property lines, marking the territory of the invention that you believe to be new and non-obvious. Crafting these claims is a delicate endeavor, and each word, comma, or symbol can impact the strength and scope of your patent.
The specification and claims of a patent, particularly if the invention be at all complicated, constitute one of the most difficult legal instruments to draw with accuracy.
The Supreme Court of the United States (Topliff v. Topliff, 145 U.S. 156, 171 (1892)).
When drafting patent claims, you need to ask a crucial question: “Who can infringe this claim?” Note the difference between analyzing whether someone might infringe the claim, and asking if it’s even possible for them to infringe the claim. This question is not trivial. It’s a tactical question at the core of furthering a patent strategy. It influences the scope of your patent and your potential to successfully assert it against alleged infringers (e.g., your competition).
Overclaiming is a Trap in the Machine Learning Patent Space
While drafting patent claims in the ML field, many inventors—and even some patent attorneys—fall into a common trap: overclaiming. It’s tempting to claim both (A) the method of training the ML model and (B) the functioning of the model once deployed. This approach, while aiming for a comprehensive protection (usually in good faith), can inadvertently weaken your patent due to the fundamentals of establishing and deploying ML models, and the complexities associated with ML patents.
Why is this problematic? The answer lies in the tenets of direct patent infringement. To establish a case of direct infringement, a single entity must perform all the steps of a patented process. So if your patent claims encompass both the training and deployment of an ML model, your potential infringer needs to engage in both of these activities to be liable for direct infringement.
Of course, there are times and strategies that incorporate both training and deployment, but the important note is that such instances should be intentional and not accidental.
Direct Infringement in Machine Learning Patent Context
Direct infringement, according to 35 U.S. Code § 271(a), occurs when a party “without authority makes, uses, offers to sell, or sells any patented invention, within the United States or imports into the United States any patented invention during the term of the patent therefor.” In the context of ML, it’s often the case that the party training the model and the party deploying the model—or operating the model—are separate entities.
This fragmentation of processes complicates direct infringement cases. It may leave patent owners with a challenging task of proving indirect infringement forms such as induced or contributory infringement. These infringement modes, as outlined by 35 U.S. Code § 271(b) and (c), require proving the accused party wanted induce an infringement or provided a substantial part of an invention knowing it would lead to infringement.
This is one of the reasons why selecting a patent attorney with an understanding of patent litigation is important for drafting effective patent claims. Such attorneys consider not only the requirements of claims to pass muster in examination, but also how claims may be interpreted and challenged in court. They use precise language with the goal of withstanding scrutiny. Such a patent attorney considers potential legal objections and builds strategies for both defense and offense in potential lawsuits right into your patent application. Being informed about current patent laws and court decisions, they draft claims that align with these legal frameworks, including intricate laws like the Alice/Mayo framework for patent eligibility. This lens enriches the patent drafting process, with the goal of providing robust claims geared toward long-term enforcement and validity.
At Stake, we consider this litigation lens on the patenting process with every patent claim. What this looks like in practice is our asking about potential competitors and design-arounds right at the start of working together on a patent.
The First Step Toward Machine Learning Patent Claims
Given these complexities, it becomes crucial to rethink the strategy of claiming both the training and deployment aspects of ML models in a single patent claim. Instead, in most circumstances it’s preferable (see supra) to focus your patent claims either from the perspective of the party training the model or the party deploying the trained model.
If your claim focuses on the training process of the ML model, any party using your specific method to train a similar model could potentially infringe your claim. This strategy allows you to cast a wide net over potential infringers who might be using your unique training method, regardless of how they use the trained model.
On the other hand, if your patent claim hones in on how the trained model is used, then any entity deploying a model that functions similarly to your patented model might be infringing upon your patent. This approach places emphasis on the end-use of the model rather than the specific training method employed.
Capturing Your Machine Learning Inventions
How do you find out what inventions your ML startup should focus on protecting? Like other tech startups, an IP capture plan is essential to secure your startup’s competitive edge and potential value. This plan helps startups systematically identify, protect, and manage their IP assets, including patents, trade secrets, copyrights, and trademarks. It can provide a clear roadmap for turning innovative ideas into protectable assets, increasing the moat around your startup’s market position. Furthermore, an effective IP capture plan enhances a startup’s appeal to potential investors and acquirers, who often consider the strength and scope of a company’s IP portfolio when evaluating its potential for growth and return on investment. Thus, an effective IP capture plan can fuel a startup’s growth, fundraising, and overall success.
Stake works with startups every day to help build out and work IP capture plans.
Take a Strategic Approach to Machine Learning Patent Claims
In the high-stakes world of tech startups, protecting your innovative ideas is crucial for long-term success. Top startups show this plainly. Patents provide this much-needed protection, serving as a legal safeguard for your inventions. However, the effectiveness of this safeguard is directly tied to how effectively your patent claims are crafted.
In the context of drafting claims for a machine learning patent application, this means being mindful of the trap of overclaiming. By focusing your patent claims on either the training or deployment of the ML model, rather than both, you can avoid diluting the strength of your patent. This targeted approach simplifies the task of proving infringement, expands the scope of potential infringers, and ultimately fortifies the protection offered to your ML inventions.
Navigating the patenting process can be challenging, and it may be beneficial to seek help from a patent attorney. Additionally, resources like the USPTO’s Manual of Patent Examining Procedure can provide valuable guidance.
In conclusion, while capturing a machine learning invention in a patent can seem daunting, a thorough understanding of the patent claiming process and a eye toward litigation can empower you to navigate this journey confidently. Protecting your invention isn’t just about filing a patent; it’s about doing so in the most effective manner to ensure robust, enforceable protection that furthers your intellectual property strategy.
