T 0702/20: patentability of a sparsely connected neural network

Blog published by Appleyard Lees IP LLP, under Artificial Intelligence / Machine learning

In T 0702/20, the Board of Appeal considered the patentability of a sparsely connected neural network in EP 14882049.1 to Mitsubishi Electric Corporation. This is one of the first cases to consider the patentability of core AI at the EPO.

In T 0702/20, the Board of Appeal considered the patentability of a sparsely connected neural network in EP 14882049.1 to Mitsubishi Electric Corporation. 

This case is interesting because it is one of the first cases to consider the patentability of core AI at the EPO. Anecdotally, the European Patent Office (EPO) is willing to grant patents for inventive technical applications and technical implementations of artificial intelligence (AI) and machine learning (ML) algorithms but not for new core AI and ML innovations. As you will see below, it is clear form this decision that care should be taken when drafting the application to ensure implementation and application details are included when filing the application. 

In this case, the preamble of Claim 1 of the Main Request recited, “A hierarchical neural network apparatus (1) implemented on a computer …” The remainder of Claim 1 went on to recite how the neural network apparatus learns weights between nodes, solves a generic regression or classification problem, and updates the weights. 

There are two steps to consider when the EPO assesses the patentability of claims involving algorithms. Firstly, does the claimed subject-matter as a whole fall within the exclusions in Article 52(2) EPC? Secondly, is there an inventive step when considering only features that contribute to the technical character of the claim? 

The EPO correctly decided that the first step was satisfied when applying the commonly used “any hardware” approach. 

When arguing in favour of the second step, the Appellant stated that the novel algorithm in Claim 1 resulted in the technical effect of reducing resources (e.g. storage) when implemented on a computer. The Board of Appeal rejected this argument by pointing out that the claim is to a new algorithm, not a new technical implementation of the algorithm on a computer. In the Board’s view, it is not possible to compare the implementation of a new algorithm with an implementation of a different algorithm because they do different things. This argument is largely aligned with the anecdote that the EPO will not grant patents to core AI innovations (i.e. modifications to algorithms), but will grant patents to new technical implementations of algorithms, if they are considered to be inventive.  

When considering the inventiveness of the claimed algorithm as a technical application of AI, the Board concluded that the claimed specificity of the task performed by the AI algorithm was too general. Performing a generic regression or classification task covered cases that are both technical (e.g. cryptography) and non-technical (e.g. forecasting stock market evolution). The Board concluded that this level of generality could not confer technical character on the algorithm to an extent that would warrant the claimed subject-matter involving an inventive step. 

In conclusion, this case largely reinforces the views of the EPO in the Guidelines for Examination, G, II, 3.3, which describes that: 

“When assessing the contribution made by a mathematical method to the technical character of an invention, it must be taken into account whether the method, in the context of the invention, produces a technical effect serving a technical purpose.” 


“A mathematical method may also contribute to the technical character of the invention independently of any technical application when the claim is directed to a specific technical implementation of the mathematical method and the mathematical method is particularly adapted for that implementation in that its design is motivated by technical considerations of the internal functioning of the computer system or network (T 1358/09G 1/19).”  

This decision largely confirms the view that the EPO does not look favourably on claims whose subject-matter is directed to improvements in core AI, rather than to novel and inventive technical applications and implementations of AI. We can only speculate regarding what level of detail Claim 1 would require regarding its technical application and/or technical implementation for this application to proceed to grant. It is important to note that details relating to application field or technical implementation of an AI algorithm need to be included when filing the EP application. Therefore, it is advisable to consult a European patent attorney who specialises in this area at the drafting stage in case there are any details that can be included that may improve the chances of success when prosecuting this type of application at the EPO.  

About the author: 

Paul Roscoe: Paul is a senior associate and patent attorney. Paul has experience of subject-matter including engineering, electronics, and computer-implemented inventions, and in particular, artificial intelligence applications.

Subscribe to the CW newsletter

This site uses cookies.

We use cookies to help us to improve our site and they enable us to deliver the best possible service and customer experience. By clicking accept or continuing to use this site you are agreeing to our cookies policy. Learn more

Start typing and press enter or the magnifying glass to search

Sign up to our newsletter
Stay in touch with CW

Choosing to join an existing organisation means that you'll need to be approved before your registration is complete. You'll be notified by email when your request has been accepted.

Your password must be at least 8 characters long and contain at least 1 uppercase character, 1 lowercase character and at least 1 number.

I would like to subscribe to

Select at least one option*