Artificial Intelligence (AI) typically involves certain common aspects. This includes, for example, training data, AI training algorithm(s) that use the training data to train an AI model, and predictions and/or classifications as output from the trained AI model. Could a person of ordinary skill in the art (e.g., a computer scientist) find it obvious to combine these common aspects to arrive at any given AI-based invention? The Patent Trial and Appeal Board recently answered “no” in its final written decision in Intel Corporation v. Health Discovery Corporation, IPR2021-00552, Paper No. 38 (September 12, 2022).
The subject patent, U.S. Patent 7,542,959 (the “'959 patent”), describes an AI-related and medical device-related invention that uses Support Vector Machines (SVM) and Recursive Feature Elimination (RFE) for selecting genes capable of accurately distinguishing between medical conditions. Both SVM and RFE are known AI algorithms. An SVM algorithm that finds a “hyperplane” (i.e., a boundary) that distinctly classifies mapped training data. An RFE algorithm selects features (columns) in a training dataset that have an impact on an output prediction or classification.
The '959 patent describes the identification of a determinative subset of features within a large set of features. Such identification is performed by training the SVM to rank the features according to classifier weights and where features are removed to determine how their removal affects the value of the classifier weights. Id. “The features having the smallest weight values are removed, and a new support vector machine is trained with the remaining weights.” '959 Patent, Abstract. “The process is repeated until a relatively small subset of features remain that is capable of accurately separating the data into different patterns or classes.” Id.
Intel petitioned the PTAB to cancel the patent's claims as obvious over printed prior art publications. Even though these publications taught all recited elements of the claims, the PTAB held that Intel failed to show that a skilled artisan would have combined the publication teachings in the manner recited in the '959 patent's claims.
The PTAB based its decision on Personal Web Technologies., LLC v. Apple, Inc., 848 F.3d 987, 993 (Fed. Cir. 2017), where the Federal Circuit determined that even though a skilled artisan may have understood that a set of prior art references could be combined in a specific claimed manner, it is not enough; instead, it must be shown a skilled artisan would have known to pick out the set of prior art references and combine them to arrive at the claimed invention. IPR2021-00552, Final Written Decision at 31.
The PTAB agreed with Intel that the prior art references could be combined. But the PTAB found that Intel nonetheless failed to provide sufficient evidence showing that a skilled artisan would have been motivated to do so:
[W]e are not persuaded by [Intel's] evidence and contention that a skilled artisan would have had a motivation to modify [the prior art] method to rank the SVM features according to their corresponding weight values as [recited by the challenged claims].
Id. at 26-27. In particular, the PTAB found that Intel's evidence and reasoning demonstrated “nothing more than a skilled artisan, once presented with the separate pieces of highlighted information in [the cited references], may have understood that they could be combined in the manner claimed.” Id. at 27.
As to the specific AI technical features, the PTAB found in the evidence no motivation “to modify Kohavi's wrapper method by changing the ranking used in the feature subset selection algorithm from an estimation of the performance of an induction algorithm to classify data properly to a variable–feature weight–used in the algorithm of an SVM to classify data.” Id.
The PTAB's decision was not unanimous, but rather split 2-to-1. The dissenting administrative patent judge stated that “Petitioner explained, with support from its expert … that its proposed addition of Hocking's vector weight ranking criteria 'applies a known technique (Hocking's variable selection) to a known device (Kohavi's RFE method using Boser's SVM) which is ready for improvement to yield predictable results.' ” Id. at 41-42 (citing KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398, 417 (2007)). Accordingly, this judge agreed that the claimed invention was nothing more than an obvious combination of known techniques applied to a known device, yielding only predictable results and thus obvious under KSR's framework. Id. at 42.
The PTAB's decision is the subject of Intel's pending “Request for Rehearing by the Director,” filed October 12, 2022. In its request, the Intel complains, among other things, that the PTAB's decision violates the Administrative Procedure Act because the patent owner never argued that a person skilled in the art would not have been motivated to combine the prior art in the manner the PTAB did. The Director's decision may be expected any day, and the case may be the subject of an appeal to the Federal Circuit. A decision by the Federal Circuit could be important to the application of Personal Web Technologies to AI-related inventions.
A more detailed version of this article can be found at Marshall Gerstein's blog at PatentNext: PTAB finds Artificial Intelligence (AI) Medical Device Patent not so Obvious.
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
© Mondaq® Ltd 1994 – 2022. All Rights Reserved.
Forgot your password?
Free, unlimited access to more than half a million articles (one-article limit removed) from the diverse perspectives of 5,000 leading law, accountancy and advisory firms
Articles tailored to your interests and optional alerts about important changes
Receive priority invitations to relevant webinars and events
You’ll only need to do it once, and readership information is just for authors and is never sold to third parties.
We need this to enable us to match you with other users from the same organisation. It is also part of the information that we share to our content providers (“Contributors”) who contribute Content for free for your use.