Summary of The Appeals Review Panel Decision in Ex Parte Desjardins (September 26, 2025)

DEREK FAHEY, ESQ.

In late September 2025, in Ex Parte Desjardins, the Director of the United States Patent and Trademark Office convened an Appeals Review Panel that vacated a §101 rejection concerning claims directed to methods of training machine-learning models. A Director-initiated vacatur of a PTAB decision is unusual, and the use of this procedure signals heightened attention to the effect of §101 jurisprudence on AI and software-related inventions.

The Claims and The PTAB’s Earlier Decision

The application, filed by DeepMind Technologies Limited, disclosed a method for training a machine-learning model on sequential tasks. A key limitation involved computing an approximation of a posterior distribution over possible parameter values and using it to preserve performance on a first task while training on a second.

The Patent Trial and Appeal Board had affirmed a §103 rejection and entered a new ground of rejection under §101, treating the claims as reciting a mathematical concept and therefore an abstract idea.

Practical Application and Technical Improvement

On rehearing, the Appeals Review Panel focused on Alice Step 2A, Prong Two: whether the claims integrated the alleged abstract idea into a practical application. The panel found that the claims reflected specific improvements in the operation of the machine-learning model, including continual learning across tasks while reducing storage requirements and system complexity.

The panel emphasized that these improvements were described in the specification and supported by the claim language. It criticized the PTAB’s broad characterization of machine learning as an unpatentable “algorithm” and its treatment of additional elements as merely generic computer components.

Broader Significance For AI and Software Applicants

The decision warned that generalized reasoning of this kind risks excluding artificial-intelligence innovations from patent protection. It also reaffirmed that §§102, 103, and 112—not §101—are the proper statutory tools for determining the appropriate scope of patent protection.

In practice, the decision reinforces the importance of clearly articulating concrete technical improvements and practical applications in both the claims and the specification. Applicants should explain how the claimed technology improves computer operation, model performance, efficiency, memory usage, or system complexity rather than relying on generalized assertions.

Conclusion

The Appeals Review Panel vacated the §101 rejection while leaving the §103 rejection intact. The decision suggests a potentially more favorable path for AI and software inventions when applicants can demonstrate a concrete improvement to underlying technology.