Return to site

Lack of Inventive Step in Machine Learning System for Reinforcement Learning

T 1952/21

· EPO,MachineLearningPatents,InventiveStep,EPCArticle56,AIandPatents

Introduction

The Board of Appeal decision T 1952/21, examined a patent application from Robert Bosch GmbH concerning a machine learning system for reinforcement learning. The Board dismissed the appeal on the grounds of lack of inventive step, finding that the claimed invention did not solve a technical problem as required by the European Patent Convention (EPC). This case highlights the challenges of patenting computer-implemented inventions, particularly in fields like machine learning, where technical contributions must be clearly demonstrated.

Summary of the Invention

The patent application (EP 18174351.9) pertains to a machine learning system designed for reinforcement learning, involving various neural networks to process input data and generate value and policy output data. The system incorporates both deterministic and stochastic units to improve the exploration of possible actions and optimize the learning process, which aims to control technical systems such as engines, valves, electrical circuits, or even robotic arms.

The key feature of the invention is the use of stochastic units within the neural networks, which allow the system to make non-deterministic decisions, enabling broader exploration and better adaptation in scenarios with incomplete data or imperfect decisions.

Summary of the Decision

The appeal arose from the Examining Division’s decision to reject the application, which was based on a lack of inventive step. The Board of Appeal confirmed the rejection, stating that the claimed invention did not sufficiently contribute to the technical character of the system.

The Examining Division argued that the claimed machine learning system merely processed abstract data through mathematical models without contributing to a technical solution. While the system included stochastic units, the Division maintained that this did not address a technical problem, as required under Article 52 EPC.

The Appellant (Bosch) countered that the use of stochastic units improved the technical capabilities of reinforcement learning systems by allowing the system to explore different decision pathways more effectively. They argued that this feature should qualify as a technical improvement, citing case law related to cryptography and simulation systems (G 1/19).

However, the Board of Appeal disagreed, noting that while stochastic units might improve the functioning of machine learning models, they did not result in a concrete technical effect when applied to the claimed system. The Board also referenced the G 1/19 decision, emphasizing that machine learning systems must serve a specific technical purpose to be patentable. Since the claimed invention could be applied in both technical and non-technical fields, the Board ruled that it did not meet the criteria for inventive step under Article 56 EPC.

The Board further pointed out that the technical advantages described by the Appellant were not supported by sufficient experimental evidence, particularly for real-world technical applications beyond video games.

Lessons to be Learned

This decision underscores the difficulties of obtaining patent protection for machine learning systems. Inventions involving mathematical models and algorithms, such as those used in reinforcement learning, must clearly demonstrate a technical contribution to be patentable. The claimed system must serve a specific technical purpose, and general improvements in the operation of machine learning models are not sufficient unless they solve a concrete technical problem.

Applicants should ensure that their patent applications provide clear evidence of technical effects, especially in fields like AI and machine learning, where the line between technical and abstract inventions is closely scrutinized.

 

Contact

If you have any questions concerning intellectual property issuesor need assistance with patent applications, oppositions, or appeals, please do not hesitate to contact us at Novitech IP. Our team of experienced professionals is here to provide you with expert guidance and support. Reach out to us today to discuss how we can help protect your innovations and navigate the complexities of IP law.

To stay informed about the latest reviewsand updates in IP law, subscribe to our blog. Join our community and receive notifications whenever we publish new reviews and insights on important case law and developments in the field of intellectual property.

Legal Disclaimer

The information provided in this blog post is for generalinformational purposes only and does not constitute legal advice. The summary and analysis of the EPO case are based on publicly available information and are intended to offer insights into the decision and its implications. This content should not be used as a substitute for professional legal advice tailored to your specific circumstances. For advice related to any specific legal matters, you should consult a qualified attorney.