The keys to QML patent success

In this co-published article, Haseltine Lake Kempner’s Laura Compton takes a practical look at how to formulate claims and draft applications for quantum machine learning inventions in view of the EPO’s patent eligibility requirements

In quantum machine learning (QML), classical machine learning algorithms, or expensive subroutines of them, are typically adapted to run on a quantum computing device. QML utilises quantum resources to improve the execution time and/or the performance of classical machine learning algorithms.

Aspects of QML that may be patentable include the utilisation of a quantum computing device to execute more efficiently all or part of a classical machine learning algorithm (for example, using a quantum computer to calculate classical distances more efficiently for nearest neighbour, kernel and clustering methods), or to execute a model itself (for example, reformulating a stochastic model as a quantum system). Other related aspects include the reformulation of an optimisation problem such that it may be solved using a quantum computing device.

Another aspect of QML that may be patentable includes improvements to existing QML algorithms or models (for example, an improvement that reduces the depth of the quantum circuit required to execute the algorithm or model, and/or uses gates that are less complex, and/or avoids repetition of certain subroutines of the algorithm). Some improvements may be specific to the problem being solved itself (for example, modifying the operations applied to a quantum computing device such that a more limited space of potential solutions to an optimisation problem is then searched over by the device).

Inventions relating to these aspects will be considered patentable subject matter at the EPO when the quantum computing device is an integral part of the invention.

For such inventions, the independent claims are likely to make some reference to the quantum computing device and the manner in which the algorithm has been adapted to be implemented on it. The dependent claims, if not the independent claim itself, should:

  1. specify the initial state of the qubits of the quantum computing device;
  2. the variables this initial state represents;
  3. how the qubits are manipulated in accordance with the algorithm;
  4. the output obtained by measurement; and
  5. what the output represents.

In view of the EPO’s “technicality” requirements, having a dependent claim that specifies how the output of the quantum computing device and/or the output of the machine learning model, is then used in some technical process, is recommended.

Where the invention relates to more general QML methods, or improvements to such methods (which could be applied to a wide range of problems across a wide range of fields), it is also recommended to provide a number of different use cases that demonstrate how the invention can be applied to different practical problems in the dependent claims, or the description,.

Quantum computing generally, as well as QML, is a rapidly evolving and complex field. As such, drafting applications which meet the EPO’s sufficiency and clarity requirements can be a challenge. Therefore, when drafting patent specifications, it is best practice to include a full mathematical description of the quantum implementation of the algorithm or model, alongside how each operation being applied to the qubits relates to the algorithm or model being implemented (for example, describing how a series of operations applied to the qubits are representative of an objective function that is to be minimised).

For inventions which relate to improving existing QML algorithms or models, detailed description on how the changes to the quantum circuit enable the improvement to be realised should be included. As with any rapidly evolving field where there is a lack of universally accepted terminologies, for applications relating to quantum computing generally, the terms used in the claims of the application should be defined in the description.

Finally, experimental data can be particularly useful in terms of demonstrating an improvement in speed or accuracy over the prior art and can be useful for supporting inventive step arguments in later prosecution. It is also worth considering setting up the technical problem the invention solves in terms of why classical processes suffer from disadvantages that make them commercially or technically non-viable (for example, too slow for real time deployment).

To summarise, the points above can be used to assist in drafting QML inventions suitable for submission to the EPO and can be used to provide the applicant with the best possible chance of obtaining a commercially useful patent.

Laura Compton is a patent attorney in the Bristol offices of Haseltine Lake Kempner

Previous articles by Haseltine Lake Kempner authors in this series can be accessed here:

How to secure AI patents in Europe

Drafting AI patent applications for success at the EPO – eligibility and claim formulation

Drafting AI patent applications for success at the EPO – drafting the full specification

Technology trends – why patent your hidden AI?

Google and Samsung top the list of applicants for AI-related patents at the EPO

The EPO and UKIPO approaches to AI and patentable subject matter

How revised EPO guidelines affect treatment of AI inventions

Monetising data, machine learning’s most valuable asset

Unlock unlimited access to all IAM content