Sie befinden sich hier:

EPO Board of Appeal (BoA) comments on sufficiency of disclosure of training data for an artificial neural network

17. August 2020

WINTER · BRANDL Partnerschaft mbB, Patent Attorney Michael Schüller

(1) 特に人工ニューラルネットワークに関連した機械学習に基づく本発明は、発明に関する人工ニューラルネットワークのトレーニングが開示不足により実施可能でないため、十分に開示されていない。

(2) 本件では、クレームされた方法は、人工ニューラルネットワークによってのみ先行技術と異なるが、そのトレーニング方法が詳細に開示されていないため、人工ニューラルネットワークの使用は、進歩性を生じさせるような特定の技術的効果をもたらすものではない。

  1. The present invention, which is based on machine learning in particular in connection with an artificial neural network, is not sufficiently disclosed, since the training of the artificial neural network in accordance with the invention cannot be carried out for lack of disclosure
  2. Since, in the present case, the claimed method differs from the prior art only by an artificial neural network, the training of which is not disclosed in detail, the use of the artificial neural network does not lead to a specific technical effect which could give rise to inventive step. (Guidelines of the BoA)

BoA, Decision from May 12, 2020 – T 0161/18 – Method for determining cardiac output / EPC Art. 83, 56

The subject matter of the patent application in suit is the use of an artificial neural network to transform a blood pressure curve

The appellant is an applicant for a European patent application (No. 06804383.5). The invention concerns, inter alia, a method for determining cardiac output. For this purpose, the blood pressure of a person is first recorded as a blood pressure curve. The person’s blood pressure is taken from an upper arm of the person via a cuff 2, see below copied Fig. 1. Via a line 3, the blood pressure is led from the cuff 2 to a device 1 with a computing unit.

The copied Fig. 2 above shows the process sequence on the computing unit. It shows the blood pressure curve 7 which was recorded via the cuff 2. With the aid of an artificial neural network 8, the blood pressure curve 7 is transformed into an equivalent aortic pressure 9. From this, the cardiac output 11 is then calculated with the aid of an optimization model 10. The neural network 8 has weight values. These are determined by learning.

The Examining Division rejected the present patent application, since the above explained method is not inventive. The appellant objects to this in her appeal.

The BoA considers the invention not sufficiently disclosed

The BoA takes a position with regard to the claimed training of the neural network 8, see Fig. 2. According to the BoA, the patent application merely discloses that the input data for the training should cover a wide range of patients of different age, sex, etc. However, according to the BoA, the patent application does not disclose which input data are suitable for training the neural network 8. Neither does the patent application disclose at least one suitable set of training data. Thus, the training of the neural network 8 cannot be post-processed by the skilled person. The invention is thus not sufficiently disclosed.

The appeal is rejected by the BoA on the grounds of insufficient disclosure, among others.

Training data should at least under certain circumstances be disclosed in the patent application

In the present patent application, the process and device claim claimed that the weight values of the neural network are determined by learning. In order to avoid the ground for refusal on insufficiency of disclosure in this case, the input data for the training and at least one training data set should be disclosed in the patent description.

Copyright © 2021 Winter • Brandl – Partnerschaft mbB

Alois-Steinecker-Str. 22, 85354 Freising | T +49 8161 930-0 | F +49 8161 930-100 |

Scroll to Top