The EPO has revealed its Guidelines to offer steerage on varied
proceedings earlier than the EPO, comparable to for the search and examination
of European patent functions and for the put up-grant opposition
of European patents. Helpfully, the Guidelines have a section devoted to the EPO’s method to
innovations that contain machine studying or synthetic
intelligence. So, what do the Guidelines say?
The Guidelines use the phrases “machine studying” and
“synthetic intelligence” interchangeably, drawing no
distinction between the 2.
The foremost level made within the Guidelines on machine studying
algorithms is that the EPO considers such algorithms, when considered
in isolation as summary processes or as a core expertise, to be
purely mathematical strategies. That is no matter the kind of
algorithm concerned, comparable to neural networks, genetic algorithms,
assist vector machines, okay-means, kernel regression and
discriminant evaluation. This helps clarify why there can usually be
difficulties in acquiring safety on the EPO for innovations
directed purely at machine studying algorithms (i.e. when divorced
from any particular implementation, software, knowledge sorts or
use-case), since purely mathematical strategies are excluded from
patentability beneath the European Patent Conference. The section within the Guidelines that units out the
EPO’s method to mathematical strategies sheds extra mild on
this, and might subsequently be useful when searching for additional perception
on how the EPO views machine studying algorithms.
Typically, innovations that contain machine studying have
supporting context (e.g. being configured to function on particular
knowledge to realize sure targets), and this may usually transfer the
invention away from being thought-about as merely a mathematical
methodology. With such innovations, making it clear that the claimed
invention is focused at reaching a technical impact can usually be
key to success on the EPO. Usefully, the Guidelines present
quite a few examples of what could also be considered as a technical impact.
Examples of this embody using a neural community in a coronary heart
monitoring equipment for the aim of figuring out an irregular
heartbeat; the classification of digital photos, movies, audio or
speech alerts based mostly on low-stage options (e.g. edges or pixel
attributes for photos); or optimising load distribution in a
pc community. Based mostly on the Guidelines, such innovations are, in
precept, patentable. Likewise, the Guidelines present examples of
what is usually considered as non-technical, comparable to classifying textual content
paperwork solely in respect of their textual content material (which is
thought-about by the EPO as a non-technical linguistic drawback). Of
course, there are sometimes shades of gray inside this, and having the ability
to initially solid your invention in the most effective mild, to make sure that
the EPO views the aim and context of the invention as reaching
a technical impact and subsequently as doubtlessly patentable, could be
extraordinarily useful – these sections of the Guidelines can
present a helpful steer on this.
There are, in fact, features of machine studying algorithms
different than simply how the algorithms are used. For instance, for a
classification algorithm, a set of coaching knowledge and a set of take a look at
knowledge could should be created and an appropriate process for the
coaching section could should be devised (e.g. to make the entire
course of sensible, environment friendly and efficient). The Guidelines make it
clear that the creation of coaching knowledge and take a look at knowledge and the
particular coaching course of are features of the invention which may be
individually protectable. Particularly, the Guidelines explicitly
state that, the place a classification methodology serves a technical
goal, the steps of producing the coaching set and coaching the
classifier may additionally contribute to the technical character of the
invention in the event that they assist reaching that technical goal.
The Guidelines additionally make it clear that, even when a machine
studying algorithm shouldn’t be focused at a selected technical
software, then patentability can lie in how the algorithm is
carried out, particularly if the implementation/design is
motivated by technical issues of the interior functioning
of the pc system or community on which the algorithm will run.
A particular instance given within the Guidelines is that of
“assigning the execution of knowledge-intensive coaching steps of a
machine-studying algorithm to a graphical processing unit (GPU) and
preparatory steps to a regular central processing unit (CPU) to
benefit from the parallel structure of the computing
platform”. Thus, the Guidelines make it clear that features of
the invention which might be tailor-made to particular {hardware}/community
issues could themselves be patentable, even within the absence of
a technical goal for the algorithm per se.
A remaining level of EPO observe that’s usually of explicit
curiosity for machine studying algorithms additionally mentioned within the
Guidelines is that an algorithm could also be extra environment friendly compared
to earlier algorithms (e.g. by way of useful resource utilization, execution
velocity, and so forth.). Nevertheless, the Guidelines make it clear that this, in
itself, shouldn’t be essentially sufficient to make sure that the brand new algorithm
is patentable on the EPO. Particularly, addressing a non-technical
drawback (such because the linguistic drawback talked about above) extra
effectively nonetheless quantities to addressing a non-technical drawback in
the eyes of the EPO. That is subsequently unlikely to be patentable.
If the invention is considered as reaching a technical impact (e.g.
figuring out an irregular heartbeat as talked about above), nevertheless,
then options of the algorithm that enhance effectivity could then
contribute to the evaluation of patentability. For instance, if the
algorithm for figuring out an irregular heartbeat has been designed
to make use of much less battery energy, then this will make it extra appropriate for
implanted units.
While the section within the Guidelines on machine studying
algorithms is comparatively brief, the steerage and examples set out
there are helpful. When learn together with the section on mathematical strategies, the
Guidelines give a transparent overview of how the EPO approaches the
varied completely different features of machine studying algorithms, together with
coaching, implementation points, deployment, core expertise and
actual-world use circumstances.
The content material of this text is meant to offer a common
information to the subject material. Specialist recommendation must be sought
about your particular circumstances.