Thanks to the interest generated, we have been closely following the progress and providing our opinion on Google’s Batch Normalization Layers patent application: Google has started registering patents for their deep learning techniques, including DQN and Dropout along with BatchNorm that we discuss in this article.
To read our insights on Google’s patents, please refer to our “Google’s AI Patent Series”.
In the previous article, we introduced the reason for the first rejection of Google’s Batch Normalization Layers patent.
After the first rejection reason was disclosed on 14 November, 2018, Google submitted a response to the office action (OA) in late January. And on January 30th, the records of the interview with the examiner were uploaded.
What is an Examiner Interview?
In the previous article regarding the first rejection reason, we explained that after an OA is issued, the applicant, Google Inc. in this case, has an opportunity to respond against the rejection reason. Generally, the applicant submits their arguments to the patent office, and such action leaves records. Matters such as what features the applicant had emphasized in order to register his or her patent, and in what aspect did the applicant comply with the examiner are all left on record.
Such records can present a potential disadvantage for the patentee in future litigation trials. There may be the case in which there may exist logical discrepancy between the arguments and content claimed to the examiners during the examination process and the arguments set forth during the patent infringement lawsuit. Thus, if possible, it may be preferable not to leave a record, however, here is when Examiner’s Interviews can be useful.
What is an interview?
According to the USPTO, “Interviews provide an opportunity to discuss the application with the examiner and can be useful to clarify positions, resolve issues, and provide a better understanding of the application to both parties.”
Many patent applicants in the U.S. utilize the examiner interview to strategically refrain from leaving their argument on record and quote the examiner’s words on the statements. Instead of stating every argument, is it possible to write as “I will argue as I said in the interview”, which would reduce the amount of arguments left on record.
Furthermore, interacting directly with the patent office examiners can reduce miscommunication and increase efficiency than arguing through written statements. In South Korea, due to additional costs, applicants tend to avoid utilizing their interviews with the examiners. However, American patent applicants actively take advantage of the system.
After such interviews, a brief summary of the interview is left on record. Not every conversation is recorded word by word but a brief summary is left on record. The recorded document on January 30th is a summary of this interview.
The interview includes every conversation regarding the reasons for the first rejection.
Infringement on the Invention’s Novelty and Inventiveness ( In regards to US5479576)
It seems that this interview failed to bring the issue of the invention’s novelty and inventiveness to the negotiation table. Google explained how its invention is different from US5479576 and the examiner acknowledged the differences. However, as the differences were not reflected in the original claims, the examiner argued that it would be impossible to grant a patent unless the claims have been revised.
Please refer to our previous article for a brief explanation regarding US5479576.
The patent application process is similar to a game of tug-of-war because it constantly involves the process of negotiating the scope of the patent between the patent office and the applicant. The claims of the application play the most important role in deciding the patent’s scope of rights.
It may be easier to understand if you consider the claims of a patent as a stake that marks your territory. No matter what kind of invention you have made, the final scope of the patent right is established by where the stakes that determine the scope of rights are embedded. Therefore, the patent examiner decides whether or not to grant a patent right based on the claims. The detailed description and drawings in the application are only supporting data to aid the claims.
That is, upon following up with Google’s technical explanation, although the examiners at the USPTO did agree with the differences between US5479576 and Google’s Batch Normalization Layers patent, they confirmed that if such differences are not reflected in the claims, they cannot grant the subject patent to Google.
Then, what should Google do?
Yes, the patent claims must be amended so that the differences between the existing inventions and their invention are also reflected in the patent claims.
However, at the end, the examiner left the following comment:
“The examiner suggested that if the concepts and language of Claim 8 and/or 9 were incorporated into at least Claim 1, prosecution could potentially be accelerated”
What does this mean?
To me, it seems that the examiner is hinting that they are willing to accept the patentability of claims 8 and 9 to a certain degree.
It sounds like a suggestion that if you apply for a review including the contents of claims 8 or 9, you will ease the concerns about novelty and inventiveness issues, and the review may be proceeded more quickly.
With that said, let’s take a look at claims 8 and 9 of the Batch normalization layers patent.
Here is claim 8:
“The neural network system of claim 7, wherein new neural network inputs processed by the neural network system after the neural network system has been trained are a different type of input than the training examples used to train the neural network system.”
Claim 8 may look simple, but is actually quite intricate. This is because claim 8 cites claim 7, while claim 7 cites claim 5, claim 5 cites 4, claim 4 cites 2, and claim 2 cites 1.
The citations of the claims are complicated, but basically, it means that after the training is performed in the first place, if a new type of input data is fed, the previously calculated normalization components will be used again.
Here is claim 9:
“The neural network system of claim 1, wherein the first neural network layer is a convolutional layer, wherein the plurality of components of the first layer output are indexed by feature index and spatial location index, and wherein computing a plurality of normalization statistics for the first layer outputs comprises:
computing, for each combination of feature index and spatial location index, a mean of the components of the first layer outputs having the feature index and spatial location index;
computing, for each feature index, an average of the means for combinations that include the feature index;
computing, for each combination of feature index and spatial location index, a variance of the components of the first layer outputs having the feature index and spatial location index; and
computing, for each feature index, an average of the variances for combinations that include the feature index.”
The contents of claim 9 look like the output of the convolutional layer for the batch normalization.
It seems highly unlikely for Google to give up on securing this part of their patent.
We assume that within the next 1 to 2 months, we will be able to find amendments in the original claims in Google’s response and amendments to the USPTO, which will most likely be on incorporating some of the technical features in claim 9 into claim 1.
Overall, even though Google has tried its best to deliver the technical aspect of the invention to the examiner, I doubt they effectively convinced the examiner on the inventiveness of the BatchNorm layers compared to the prior art.
However, from what I see, what Google gained through this interview is the clue to overcome the issue of patent eligibility without sacrificing too much of their potential right.
Regarding patent eligibility issues, it is necessary to understand the background, including the previously mentioned US Supreme Court precedents.
I’ll deal with that and explain in more detail in the subsequent article.
![Patent Eligibility Issues: Updates on Google’s Batch Normalization Layers Patent [2]](https://storage.googleapis.com/piip-intra-content-production/article-images/1607499732036/10754058-b708-4bec-858d-a9902f4999f8/what would google do with batch norm layers patent.jpg)
