HOME > English > Insights > Patent

Dongyu LI
Patent Attorney

 
I. Introduction

In recent years, technologies such as artificial intelligence, “Internet Plus,” big data, and blockchain have advanced at a breathtaking pace, generating an urgent need to protect innovative achievements in emerging fields and new business formats. To meet this demand, the 2019 revision of the Guidelines for Patent Examination (hereinafter referred to as “Guidelines”) added a new Section 6 to Part II, Chapter 9, entitled “Provisions on the Examination of Invention Patent Applications Involving Algorithmic Features or Business Rules and Methods”, which sets out explicit requirements for subject-matter eligibility, novelty, inventive step, and the drafting of the description and claims for such applications. The 2023 revision of the Guidelines has further refined the examination standards for patent applications involving algorithmic features, thereby strengthening protection for inventions in these new fields and formats.

As the criteria for assessing subject-matter eligibility of claims containing algorithmic features have become clearer, the number of applications that clearly fall outside patent-eligible subject matter has decreased markedly, while objections based on lack of inventiveness have grown more frequent. Consequently, a correct understanding and effective application of the rules for evaluating inventiveness in such patent applications is essential for both applicants and patent attorneys.

II. Provisions of the Guidelines

With regard to the inventive step of patent applications involving algorithmic features, Part II, Chapter 9 of the Guidelines provides:

When examining inventive step for an invention patent application that contains both technical features and algorithmic features, or business rules and methods, the algorithmic features or business rules and methods that, functionally, mutually support and interact with the technical features shall be considered together with the technical features as a whole. “Functionally, mutually support and interact” means that the algorithmic features or business rules and methods are closely integrated with the technical features, jointly constituting the technical means for solving a particular technical problem and achieving the corresponding technical effect.

Regarding the specific criteria for determining “functionally, mutually support and interact”, the Guidelines further stipulate as follows:

If the algorithm in the claims is applied to a specific technical field and can solve a specific technical problem, it can be considered that the algorithmic features and technical features are functionally, mutually support and interact. Such algorithmic features become part of the technical means adopted and their contributions to the technical solution shall be considered during the examination of inventiveness.

If the algorithm in the claims has a specific technical association with the internal structure of a computer system, achieves an improvement in the internal performance of the computer system, and enhances hardware computational efficiency or execution performance, such as reducing data storage, decreasing data transmission, or increasing hardware processing speed, then the algorithmic feature may be regarded as functionally, mutually supporting and interacting with the technical features, and its contribution to the technical solution shall be considered in the inventive-step assessment.

Where the solution of the invention patent application enhances user experience, and such enhancement is brought about either by the technical features alone, or jointly by the technical features together with the algorithmic features or business rules and methods that functionally, mutually support and interact with the technical features, the enhancement shall likewise be taken into account in the inventive-step examination.

Below, the author, by drawing on three practical cases, will illustrate how to evaluate the correlation between algorithmic features and technical features, as well as their technical contributions to the technical solution, in the examination of inventive step for a patent application involving algorithms.

III. Case Introduction

1. Case 1 (201810734681.2)

This case is one of the top ten patent reexamination and invalidation cases in 2024, involving “a method and an apparatus for processing images”.

Claim 1 subject to the Decision of Rejection is as follows:

A method for processing images, comprising:

obtaining a captured image of a target object;

inputting the captured image into a pre-trained key point detection model corresponding to the target object, thereby obtaining a set of position information, wherein the position information comprises position coordinates that indicate, in the captured image, the positions of key points indicated by key-point information in a key-point information set, the key points being predetermined points on the target object, and different target objects corresponding to different key-point detection models;

wherein the position information further comprises visibility information that represents the probability that a key point indicated by the key-point information in the key-point information set is visible in the captured image.

The Decision of Rejection cited D1 and D2 to deny the inventiveness of claim 1.

When filing the request for reexamination, the applicant did not amend the claims. During the reexamination procedure, the applicant amended claim 1. Compared with the examination text addressed by the Decision of Rejection, the following features were added to claim 1: “the target object comprises at least one of a designated object and any object captured by the image-capturing device within a preset time period, the target object including a sports court” and “the key-point detection model is trained by using captured images of the target object together with the corresponding sets of position information for those images”.

Based on the above amendments, the collegial panel held that the amended claim 1 possesses inventiveness over D1 and D2.

In the Decision of Reexamination, the collegial panel conceded that most features of claim 1 were not disclosed by D1. Regarding the inventiveness of claim 1, the collegial panel analyzed as follows:

Based on the distinguishing features, the actual technical problem solved by claim 1 is how to determine the corresponding relationship of the same target object in different captured images.

The collegial panel held that the above distinguishing features limit the target object in claim 1 to a designated or captured specific court object. For each such specific court object, a key-point detection model is trained exclusively for that object by using its own captured images together with the corresponding sets of position information. When multiple frames of images of the same specific court object (e.g., images taken from different angles or at different times) are input into the corresponding key-point detection model, the model outputs the varying position coordinates and visibility information of the court object’s key points across those images. From this output, the geometric transformation relationships of the same key points across the images can be determined, thereby enabling match-data analysis based on the multi-frame court images.

Although D1 mentions a key-point positioning model, the target objects processed are only facial images and do not involve court images; all facial images in D1 usually correspond to the same face key-point positioning model, which is not in one-to-one correspondence with a specific target object, and the face recognition function involved in D1 has no such requirement. Therefore, a person skilled in the art cannot obtain technical suggestion from it to set a key-point positioning model in one-to-one correspondence with each target object; the face key-point positioning model in D1 only targets key points of a single frame of face image and its image quality, not involving the corresponding relationship of key points of the same target object in a plurality of frames of images. Therefore, D1 does not disclose the above distinguishing features, nor does it provide a suggestion for adopting the above distinguishing features to solve the technical problem of determining the corresponding relationship of the same target object in different captured images.

The processing objects of D2 also do not involve the court captured images as defined by the aforementioned distinguishing features, and do not provide a technical suggestion for establishing a key point detection model in one-to-one correspondence with the target object as in the aforementioned distinguishing features, let alone face the technical problem of determining the corresponding relationship of key points of the same court object in different captured images. Therefore, D2 does not disclose the aforementioned distinguishing features, nor does it provide a suggestion for combining with D1 and applying it to court objects to solve the technical problem of determining the corresponding relationship of the same target object in different captured images.

The China National Intellectual Property Administration (CNIPA) remarked on the typical significance of this case as below:

This case has a demonstration effect on the examination of inventiveness of technical solutions combining algorithms or models with application scenario features, and encourages scenario innovation to drive technological innovation. The Decision points out that under the “three-step method” framework for inventiveness examination, full consideration should be given to whether application scenario features lead to substantial adjustments or changes in algorithms or models. If, compared with the prior art, the application scenario of the algorithm or model in the patent application in question is different, resulting in substantial differences in the structure of the algorithm or model, the training data, input data, output data processed by the algorithm or model, the selection of the algorithm or model, etc., which are non-obvious to a person skilled in the art, and corresponding technical effects are achieved, then the technical solution possesses inventiveness. If the technical contribution of the invention lies in solving a specific technical problem in a certain application scenario through the improvement of the algorithm or model, the innovator should fully reflect the above content when drafting the application document, making a response, and amending the claims.

The technical contribution of Case 1 lies in solving a specific technical problem in a specific application scenario through algorithm improvement. In the present application, by adding features related to the specific application scenario in claim 1 during the reexamination phase: “the target object comprises at least one of a designated object and any object captured by the image-capturing device within a preset time period, the target object including a sports court” and “the key-point detection model is trained by using captured images of the target object together with the corresponding sets of position information for those images”, the amended claim 1 fully reflects the specific application scenario applicable to the algorithm and the specific technical problem, thereby enabling the technical features and algorithmic features to support each other, and the contribution of the algorithmic features to the technical solution is identified by the collegial panel.

2. Case 2 (201780089483.9)

Case 2 involves a system and a method for generating a unified machine-learning model using a neural network.

The targeted claim 1 in the Decision of Rejection is as follows:

A computer-implemented method for generating a unified machine -learning computation model on a data-processing apparatus using a neural network, the method comprising:

determining, by the data processing apparatus and for the neural network, a respective learning target for each of a plurality of object vertices, wherein each object vertex defines a different category of objects belonging to the vertex, and wherein each learning target is based on two or more embedding outputs of at least one other neural network; wherein determining the respective learning targets for the neural network further comprises:

training, by the data processing apparatus and based on a second loss function, a plurality of dedicated models, wherein each of the plurality of dedicated models is trained to identify a different object vertex; and

generating, by the data processing apparatus, one or more embedding outputs using each of the plurality of dedicated models, wherein the respective learning targets are determined based on the embedding outputs;

training, by the data processing apparatus and based on a first loss function, the neural network to identify data associated with each of the plurality of object vertices, wherein the neural network is trained using the respective learning targets; and

generating, by the data processing apparatus and using the neural network trained based on the first loss function, a unified machine-learning model configured to identify items included in data associated with each of the plurality of object vertices,

wherein the first loss function is an L2 loss function and generating the unified machine-learning model comprises:

generating a specific unified machine-learning model that minimizes a computational output associated with the L2 loss function.

The Decision of Rejection cited D1 to deny the inventiveness of claim 1.

When filing the request for reexamination, the applicant incorporated the following feature to claim 1: “wherein the plurality of dedicated models are first trained, based on the second loss function, to achieve a desired threshold accuracy level in distinguishing objects included in the image data, and the embedding outputs of each trained dedicated model then serve as learning targets for training the unified machine-learning model”.

In the phase of collegial examination, the panel directly recognized the inventiveness of the claims amended in the request for reexamination, and issued a Decision of Reexamination revoking the Decision of Rejection without issuing a reexamination notification.

In the Decision of Reexamination, the panel acknowledged that most features of claim 1 were not disclosed in D1. Regarding the inventiveness of claim 1, the collegial panel analyzed as follows:

Based on the distinguishing technical features, it can be determined that the actual technical problem solved by claim 1 is how to expand the categorical recognition scope of a model without reducing the distinguishing accuracy.

To solve the above technical problem, the technical solution sought to be protected by 1 in the present application, in classification and recognition, after training a plurality of dedicated models, obtains the embedding outputs of each dedicated model as the learning target of the unified learning model to train the unified machine-learning model. However, D1, in an image recommendation or visual recommendation, based on deep learning, uses a dedicated model or a unified machine-learning model to provide users with ranked similar product images. The dedicated model and the unified machine-learning model in D1 are trained separately, and the trained dedicated model and unified machine-learning model can be used to process input images and output processing results. It can be seen that in the training process of the dedicated model and the unified machine-learning model in D1, there is no need to use the output of one model as the learning target of another model, that is, D1 cannot provide a suggestion for using the embedding outputs of the dedicated model as the learning target of the unified learning model. Therefore, based on the disclosure of D1, a person skilled in the art would have no motivation to make improvements to obtain the distinguishing technical features.

At the same time, there is no evidence in the case proving that the above distinguishing technical features as a whole belong to the common knowledge in the art.

The above distinguishing technical features enable the technical solution sought to be protected by 1 to have the following beneficial technical effects: the classification and recognition range of the model can be expanded without reducing the distinguishing accuracy.

The technical contribution of Case 2 lies in optimising the scheduling of hardware resources within a computer system through the execution of an algorithm, thereby enhancing the performance of classification and recognition tasks and effecting an improvement in the internal performance of the computer system.

Specifically, the technical solution sought to be protected by 1 trains a plurality of dedicated models based on a second loss function so as to achieve a desired threshold accuracy level for identifying objects present in image data. The embedding outputs of each trained specialised model are then used as learning targets for training the unified machine-learning model, allowing the scope of classification and recognition to be expanded without degrading accuracy. Consequently, the algorithmic features for training the unified machine-learning model and the technical features functionally support and interact with one another. As an integrated whole, these features must therefore be taken into account in the inventive-step assessment. The collegial panel found that this combined set of features is neither disclosed in D1 nor derivable from common knowledge. Accordingly, claim 1 involves an inventive step.

3. Case 3 (201810476385.7)

Case 3 involves a blockchain-based offline transaction method and apparatus.

The targeted claim 1 in Decision of Rejection is as follows:

A blockchain-based offline transaction method, comprising:

obtaining a designated identifier and determining whether a corresponding record of the designated identifier exists on the blockchain;

verifying whether the designated identifier is in an activated state when the record exists;

when the designated identifier is in the activated state, verifying whether the current transaction satisfies transaction restrictions corresponding to the designated identifier on the blockchain, where the transaction restrictions are set when a payment account delegates the designated identifier for payment;

when the transaction restrictions are satisfied, generating offline transaction information according to the payment account corresponding to the designated identifier and recording the offline transaction information on the blockchain.

The Decision of Rejection cited D1 and D3 to evaluate the inventive step of claim 1.

When filing the request for reexamination, the applicant amended claim 1 by adding “the offline transaction refers to an electronic payment conducted when the wireless network is out of service or the terminal is powered off”, “wherein the payment function of the paying account is delegated to the designated identifier via the blockchain”, and “while in the activated state, offline payment using the designated identifier is permitted”.
 
At the stage of collegial examination, the panel accepted the inventive step of the claims as amended in the request for reexamination and issued a Decision of Reexamination revoking the Decision of Rejection without issuing a Re-examination Notification.

In the Decision of Reexamination, the collegial panel acknowledged that most features of claim 1 were not disclosed in D1. Regarding the inventiveness of claim 1, the collegial panel analyzed as follows:

Based on the above distinguishing features, it can be determined that the actual technical problem solved by the claims is how to ensure the continuity of electronic payment when the wireless network is out of service or the terminal is powered off.

Although D3 discloses the technical features of “activated state” and “transaction restrictions”, D3 does not involve the offline transaction demand when the wireless network is out of service or the terminal is powered off, and there is no need to set a designated identifier for the payment account.

There is also no evidence to prove that the above distinguishing features belong to common approach in the art.

By means of the distinguishing features, the payment function of the payment account is delegated to the designated identifier through the blockchain, so that in the offline situation where the wireless network is out of service or the terminal is powered off, the designated identifier is obtained, it is determined whether a corresponding record of the designated identifier exists on the blockchain, it is verified whether the designated identifier is in an activated state when the record exists, offline payment using the designated identifier is permitted in the activated state, when the designated identifier is in the activated state, it is verified whether the current transaction satisfies the transaction restrictions corresponding to the designated identifier on the blockchain, where the transaction restrictions are set when the payment account delegates the designated identifier for payment; when the transaction restrictions are satisfied, transaction information is generated according to the payment account corresponding to the designated identifier, thereby continuing electronic payment in the offline state, improving user experience, and achieving beneficial technical effects.

The technical contribution of Case 3 resides in the combined effect of the technical features and the algorithmic features that functionally, mutually support and interact with them, thereby achieving an enhanced user experience.

IV. Conclusion

As illustrated by the above examples, the criteria applied by the examiners when assessing the inventive step of patent applications involving algorithmic improvements are fully consistent with the provisions of the Guidelines. The decisive factor is whether the algorithmic features “functionally, mutually support and interact” the technical features. When this functional interdependence exists, the algorithmic features must be taken into account in evaluating inventive step, specifically, by determining whether those features have been disclosed or suggested, or whether they are common knowledge.

Therefore, a thorough grasp of the provisions set forth in the Guidelines is essential for both applicants and patent attorneys practicing in the field of algorithms. Based on the above three re-examination cases, this article has briefly discussed the assessment of inventive step in patent applications involving algorithms. Should any inaccuracies be found, comments and corrections from fellow practitioners are sincerely welcome.