Monday, May 11, 2020 T 509/18 - Lack of sufficiency in a machine learning application

This examination appeal concerns a driver alertness detection system. Based on imaging of the driver's head and eyes the driver's attention state is determined. Determining the attention state uses a machine learning system. 

The description has a few sections in which the machine learning aspects are explained. Apparently, a matrix of inter-point metrics is used for a look-up-table classification. In principle classification systems are well known, and this suggests some kind of nearest neighbor classification, so I didn't think anything of it.  But the board probes deeper and finds the description insufficient. On closer reading it is indeed unclear what exactly is happening. I'm convinced that one could make a system like this work in many ways; but perhaps not while literally following the description. 

The applicant had argued that the skilled person would know how to enable this based on his common general knowledge; the applicant cited multiple documents in support of this argument. Moreover, in first instance the application was refused for lack of novelty not for lack of sufficiency.  The reasons of the appeal decision do not refer to the common general knowledge though.

In any case, it is a useful reminder not to be too succinct in your description. 

Reasons for the Decision

1. The appeal is admissible.

2. The invention as defined in claim 1 of the third auxiliary request (including all the features of claim 1 of each of the main, first and second auxiliary request) is not disclosed in the European patent application 12 765 209.7 (WO-A) in a manner sufficiently clear and complete for it to be carried out by a person skilled in the art (Article 83 EPC).

The feature (of claim 1) reading "the driver alertness detection system is configured to use a classification training process to register the driver's head position and eye vector for the A-pillars, instrument panel, outside mirrors, rear view mirror, windshield, passenger floor, center console, radial and climate controls within the vehicle, and configured to save a corresponding matrix of inter-point metrics to be used for a look-up-table classification of the driver's attention state, the inter-point metrics being geometric relationships between detected control points and comprising a set of vectors connecting any combination of control points including pupils, nostrils and corners of the mouth" is derived from paragraphs [0027], [0029], [0033] in WO-A and constitutes the central feature on which the driver alertness detection system of the invention is based.

Further, according to paragraph [0029] in WO-A, figure 8A is an image showing a driver in a full-alert state, figure 8B is an image showing a driver in an attention partially diverted state, and figure 8C is an image showing a driver in an attention entirely diverted state.

In the Board's view the definition of claim 1 and the corresponding passages in the patent application (WO-A) do not teach the skilled person how a "look-up-table classification of the driver's attention state" is to be obtained by the skilled person, based on said "matrix of inter-point metrics", the inter-point metrics representing "geometric relationships between detected control points and comprising a set of vectors connecting any combination of control points including pupils, nostrils and corners of the mouth".

In particular, WO-A does not teach how to derive from said "matrix of inter-point metrics" a "look-up-table classification of the driver's attention state", such a "look-up-table classification" permitting to decide on the driver's attention state. A "matrix of inter-point metrics" being a mathematical object representing a set of "geometrical relationships between detected control points" according to WO-A (and to claim 1), a specific mathematical method and corresponding criteria (or algorithms) necessarily have to be determined in order to be able to handle said matrix and to deal with said matrix. No such mathematical methods and corresponding criteria allowing to handle said matrix and obtain a "look-up-table classification" are disclosed or even suggested in the description of WO-A. In addition, the actual specific form and construction of said "matrix of inter-point metrics" is likewise not specified in WO-A. Therefore the skilled person would not know how to construct a "look-up-table classification" and consequently how to decide on the driver's attention state based on the video camera's image of the actual position of driver's head and eyes at a given instant.

In addition, claim 1 and WO-A likewise do not teach how a video camera's image representing the instant position of a driver's head and eyes (as seen e.g. in figures 8A, B or C) should be actually compared with a hypothetical "look-up-table classification" in order to assess the driver's attention state. In effect, this step requires instructions and teaching concerning the kind of information to be extracted from a given video camera image and concerning the method and the criteria (similarly as above) to be applied in order to compare this information with the information included in the hypothetical "look-up-table classification". No such disclosure is to be found in the description of the patent application (WO-A).

The same conclusions apply a fortiori to claim 1 of the main, first and second auxiliary request, since the subject-matter of each of these claims includes only part of the features of claim 1 of the third auxiliary request, thus including even less information than is included in claim 1 of the third auxiliary request.

It ensues that, for the same reasons as indicated in relation to claim 1 of the third auxiliary request, claim 1 of aforesaid auxiliary requests (in conjunction with the description) likewise does not disclose the invention in a manner sufficiently clear and complete for the skilled person to be able to carry out e.g. a "classification of the driver's attention state".

Order

For these reasons it is decided that:

The appeal is dismissed.

This decision T 0509/18 (pdf) has European Case Law Identifier: ECLI:EP:BA:2020:T050918.20200303. The file wrapper can be found here. Photo by Rudy and Peter Skitterians (Skitterphoto) obtained from Pixabay under the Pixabay license.