6.1. Reproductibilité

In T 48/24 claim 1 as granted defined a device comprising a "training data generation unit", a "model construction unit" and an "estimation unit". The device was for training and using machine learning for obtaining a "value representing the composition of waste" in a waste pit upon inputting "data of" a captured image of the waste.

The board observed that sufficiency of disclosure had to be assessed for each case individually. The board saw no apparent reason to treat inventions in the field of machine learning differently from other inventions in this regard. According to the board, it went without saying that the implementation of a suitable machine learning model, its training, and whether the trained model can successfully estimate the specified output based on the input parameters as claimed may be important aspects of sufficiency of disclosure of machine learning inventions. However, there were no special requirements and no general rules for assessing whether these aspects were sufficiently disclosed.

With regard to the decisions referred to by the appellant (opponent), the board considered that it was not possible to derive from T 161/18 any general criteria which might be applicable to the present case. Similarly, the board observed that T 1669/21 illustrated the glaring gap between the breadth of the claimed invention and the level of detail in the patent, but did not provide generally applicable criteria required for sufficiently disclosing a machine learning invention. Likewise, the board also pointed out that the considerations in the present case did not imply any universally applicable criteria for assessing sufficiency of disclosure of machine learning inventions.

In the case in hand, the board noted that the patent did not contain a specific example of the claimed invention. That is, it did not disclose any specific combination of certain "data of" captured images and a particular "value representing composition" of the waste in the images, nor did it provide any details on the implementation and training of an exemplary machine learning model, or any information on the achieved accuracy of estimation. In other words, the patent did not contain any concrete, reproducible example of implementation of the invention. Such a specific example was not in itself an absolute requirement for sufficient disclosure, provided that the skilled person was aware of "at least one way" of carrying out the invention in other ways, for example, through the generic disclosure in the patent or the common general knowledge (see R. 42(1)(e) EPC, "using examples where appropriate"). In the present case, however, providing such an example could have demonstrated that the invention was workable at all, at least in this specific case of the example. It could have served as a reference to better understand the claimed invention, its terms and purpose and the achievable or expected level of accuracy.

The board also explained that sufficiency of disclosure required that the invention could be carried out over the whole claimed breadth without undue burden. This requirement had been formulated in decisions across all technical fields (T 149/21). It reflected the general principle that the protection obtained with the patent had to be commensurate with the disclosed teaching. Even if "one way" of performing the invention was disclosed, this would only be sufficient if this disclosure enabled the skilled person to perform the invention over the whole claimed breadth. The patent in suit taught the general idea of using machine learning to infer properties of the waste composition that could be relevant for operating and controlling a waste incineration plant from images of the surface of the waste pit. However, the disclosure was mostly limited to stating a "result to be achieved". The patent left it to the skilled person to select and evaluate combinations of input data and machine learning models for different desired outputs. Each evaluation involved implementing, training and evaluating the selected models. Overall, this resulted in an enormous number of parameter combinations to choose from. Exploring all the possible combinations of these parameters would require a comprehensive research programme and would place an undue burden on the skilled person.

The board concluded that the maintenance of the patent as granted according to the main request was prejudiced by the ground for opposition under Art. 100(b) EPC. Auxiliary requests 1 to 7 were not allowable under Art. 83 EPC either.