Recently, our firm won the first instance of an administrative litigation case for invalidation of an invention patent, in which it is ruled that the invention patent should be comp...
Introduction In the chemical field, it is difficult to seek invalidation of a patent claiming to have achieved unexpected technical effect. This case provides a strategy for success...
Discussion on Drafting of Patent Applications Involving Algorithms of Artificial Intelligence from the Perspective of Subject Matter Eligibility
Wendy ZHANG
Chinese patent attorney
Electrical Engineering Department
Linda Liu & Partners
The CNIPA issued the latest amended Guidelines for Patent Examination on December 31, 2019 which was to take effect as of February 1, 2020. In the amended Guidelines for Patent Examination, a new Section 6 "Relevant Provisions on the Examination of Patent Applications for Inventions Containing Algorithm Features or Business Rules and Method Features" was added to Chapter 9 of Part II, aiming to stipulate the examination of patent applications for inventions involving AI, "Internet +", big data, block chain, etc. which generally contain algorithms or business rules and methods, according to the requirements of the Chinese Patent Law and the Implementation Regulations of the Chinese Patent Law.
In fact, during and long before the period of soliciting public opinions on the amended Guidelines for Patent Examination, the CNIPA has already examined patent applications for inventions involving algorithms in the field of AI in accordance with the relevant provisions of Section 6 in Chapter 9 of Part II of the amended Guidelines for Patent Examination under the relevant requirements of the Chinese Patent Law and the Implementation Regulations of the Chinese Patent Law. Herein, the provision of the amended Guidelines for Patent Examination with regard to patent examination under Paragraph 2 of Article 2 of the Chinese Patent Law is discussed with reference to cases of patent applications for AI algorithms handled by the author in the latest two years, in hope of being informative and inspiring for the readers.
The amended Guidelines for Patent Examination formulates the examination on the subject matter eligibility under Paragraph 2 of Article 2 of the Chinese Patent Law as below:
In examining whether or not a claim containing algorithm features or business rules and method features is a technical solution, the examiner shall consider all the features of the claim as a whole. If the claim recites using technical means utilizing the law of nature to address the technical problem to be solved and thereby achieves a technical effect conforming to the law of nature, the solution defined in the claim shall be determined as a technical solution prescribed in Paragraph 2 of Article 2 of the Chinese Patent Law. For example, if the steps of the algorithm involved in the claim are closely related to the technical problem to be solved, say the data processed by the algorithm is data with specific technical meaning in the technical field, the execution of the algorithm can directly reflect the process of using law of nature to solve a certain problem and achieves a technical effect, the solution defined in the claim should be usually determined as a technical solution prescribed in Paragraph 2 of Article 2 of the Chinese Patent Law.
This provision further explains and clarifies how to draft claims related to algorithm features to meet the requirements of the "technical solution" specified in Paragraph 2 of Article 2 of the Patent Law. Hence, the amended Guidelines for Patent Examination provides applicants and patent attorneys in the field of AI algorithms with more detailed guidance in drafting invention patent applications containing algorithm features and clarifies the ambiguity in the algorithm steps in the claims and the technical field of application. Moreover, it effectively eliminates the possibility that an applicant might get a granted patent by filing extensive patent applications for a general algorithm. Under the above provision, the claims of a patent application involving an AI algorithm should clearly show that each step of the algorithm is to the technical problem to be solved.
In a nutshell, the "closely related" relationship includes the following aspects: 1. The data processed by the algorithm must have specific technical meaning, instead of being an abstract data concept; 2. The processing should reflect that the data is processed in accordance with the laws of nature; 3. The output data of the processing by the algorithm must have specific technical meaning rather than being an abstract data concept; and 4. The execution of the algorithm can solve a certain technical problem and achieve corresponding technical effects. Subsequently, three cases are described to illustrate how it is determined whether or not an algorithm step in the claim is "closely related" to the technical problem to be solved.
Case 1- an example of "closely related"
Subject matter: a processing method of convolutional neural network (CNN) features
Addressed technical problem: how to process CNN features to improve the accuracy of image recognition and search
Claim:
A processing method of CNN features, for processing feature images obtained by inputting a plurality of image groups into a convolutional neural network, comprising:
inputting each image group in the plurality of image groups into a corresponding sub-neural network in a neural network model to obtain a visual feature vector of each image group;
calculating distribution of each feature image on an original image based on the obtained visual feature vector;
extracting a feature vector corresponding to each element in the original image according to the distribution of each feature image on the original image;
for a region of interest in the original image, adding feature vector of all elements in the region of interest to obtain a feature vector corresponding to the region of interest, so as to obtain an image feature extraction model.
Analysis and conclusion
This solution is a processing method of convolutional neural network features. It clarifies that the data processed in each step is image data or a certain feature representation obtained from image data, and that the execution of each step reflects how to obtain the feature vector of the region of interest in the original image step by step based on original image data (the input data, the intermediate processing data, and the output data being underlined). Hence, the algorithm for processing convolutional neural network features is closely related to image information processing. This solution solves the technical problem of improving image recognition and search accuracy, uses a technical method that follows the laws of nature, and achieves corresponding technical effects. The claim is drafted in full conformity to requirements of the “closely related” relationship in the afore-mentioned four aspects. Therefore, the solution of the patent application for invention belongs to the technical solution specified in Paragraph 2 of Article 2 of the Chinese Patent Law and is eligible.
Case 2- an example of not “closely related"
Subject matter: a feature selection method based on artificial neural network
Addressed technical problem: how to improve the efficiency of feature selection
Claim:
A feature selection method based on artificial neural network, comprising:
constructing an artificial neural network with an input layer, an intermediate layer and an output layer according to a feature to be selected and an output target;
training the artificial neural network with a training set to determine a connection relationship between each layer in the artificial neural network and its upper and lower layers, wherein optimization functions used for training includes items for selection of a respective layer, so as to pick out the feature to be selected according to the connection relationship.
Analysis and conclusion:
This solution is a feature selection method based on artificial neural network. Feature selection is a data processing method widely applied in many technical fields. Feature selection itself is a mathematical algorithm because the data it processes can be any data. Only when feature selection is applied to a specific technical field will a specific technical solution be formed to solve a corresponding technical problem and achieve a corresponding technical effect. The input data processed by this method is "feature to be selected", which is an abstract concept. It does not have a specific technical meaning when not applied to a specific technical field. The processing of the "feature to be selected" by the algorithm as defined in the claim is essentially a description of the general algorithm and does not include processing data with specific technical meanings in accordance with the laws of nature. Moreover, the claimed technical effect of “improving the efficiency of feature selection” in the specification is actually more an effect of improvement to the existing feature selection algorithm made by the algorithm according to the application than a technical effect. Obviously, the drafting of this claim does not meet the requirements for the “closely-related” in the above-mentioned four aspects. Therefore, the solution proposed the patent application for invention is not a technical solution prescribed in Paragraph 2 of Article 2 of the Chinese Patent Law and thus is ineligible.
Case 3- an example of not sufficiently "closely related"
Subject matter: a processing method for recovering spatio-temporal data sequence with missing data component
Addressed technical problem: how to improve the accuracy of recovering missing data of spatio-temporal sequence data
Claim:
A processing method for recovering missing data of component data, configured to perform supplementation processing of missing component data in a data sequence of spatial and temporal dimensions to obtain a complete spatio-temporal data sequence, comprising:
calculating spatial dimension data of a point to be calculated according to multiple spatial peripheral points of the point to be calculated;
calculating time dimension data of a point to be calculated according to multiple time peripheral points of the point to be calculated;
calculating component data of the point to be calculated based on the spatial dimension data and the temporal dimension data.
Analysis and conclusion
This solution is a processing method for recovering spatio-temporal data sequence with missing data component. According to the specification, the addressed problem is the filling in missing spatio-temporal data, not a technical problem in a specific field. The adopted method is to estimate the missing data at the point to be calculated based on the spatial dimension data and time dimension data of the point to be calculated, which is not applied to a specific technical field and is not technical means conforming to the law of nature. In addition, it only achieves the effect of obtaining a missing data value, which is not a technical effect.
Although the applicant tried to limit the missing data to be processed to "missing data of component data" to clarify the technical field of application, in terms of the overall technical solution, it does not change the fact that the data processed in the steps of the algorithm does not have specific technical meanings in the field of component analysis. In fact, the data processing by the algorithm does not use the features of the component data in the respective field, but uses the spatial dimension data and the time dimension data of the point to be calculated where the component data is missing. Since other fields may also have the common data features of spatial dimension data and time dimension data, the data processed by the algorithm has no specific technical meaning, and the algorithm is a general algorithm in essence. Moreover, the effect of "improving the accuracy of recovering missing data of spatio-temporal sequence data" mentioned in the specification is more an effect of the improvement to the existing algorithm for filling in missing spatio-temporal data by the algorithm proposed by this application , than a technical effect.
To say the least, even if the requirements of the “closely related” relationship in the aspects 1 and 3 are far-fetchingly satisfied by limiting the data processed in the steps of the algorithm to "missing data of component data" and limiting the output data to "component data of the point to be calculated" as submitted by the applicant, the processing by the algorithm does not involve intermediate data related to the technical meaning of the component data per se, nor does it reflect the processing of the component data in accordance with the laws of nature. Therefore, at least the requirement of the “closely related” relationship according to aspect 2 is not satisfied. as a result, the solution is not so "closely related" to the technical problem to be solved.
Therefore, the above solution of the patent application for invention is not a technical solution prescribed in Paragraph 2 of Article 2 of the Chinese Patent Law and thus is ineligible..
From the above three cases, it can be easily seen that an intuitive way to determine whether or not a solution of an invention patent application containing algorithm features is a technical solution prescribed in Paragraph 2 of Article 2 of the Chinese Patent Law is to check whether the steps of the algorithm in the technical solution is closely combined with a specific technical field, i.e. whether the data processed by the algorithm is data with specific technical meaning in the technical field and whether the execution of the algorithm can directly reflect the process of solving a technical problem using the laws of nature and achieves a technical effect. If the result is negative, the patent application for invention containing algorithmic features will not possibly be considered as a technical solution prescribed in Paragraph 2 of Article 2 of the Chinese Patent Law and will possibly be determined as ineligible.
