PTAB

IPR2021-00550

Intel Corp v. Health Discovery Corp

Log In
Key Events
Petition
petition

1. Case Identification

2. Patent Overview

  • Title: Method for Identifying Patterns in Data Using Support Vector Machines
  • Brief Description: The ’188 patent discloses a computer-implemented method for identifying patterns in data by combining a support vector machine (SVM) classifier with an iterative feature selection technique known as recursive feature elimination (RFE). The method repeatedly trains an SVM, computes a ranking criterion for data features based on classifier weights, eliminates the lowest-ranked feature, and repeats the process to identify an optimal subset of features.

3. Grounds for Unpatentability

Ground 1: Obviousness over Kohavi, Boser, and Hocking - Claims 1, 3-8, 10, 13-17, and 19-23 are obvious over Kohavi in view of Boser and Hocking.

  • Prior Art Relied Upon: Kohavi (a 1997 article, "Wrappers for feature subset selection"), Boser (Patent 5,649,068), and Hocking (a 1967 article, "Selection of the Best Subset in Regression Analysis").
  • Core Argument for this Ground:
    • Prior Art Mapping: Petitioner argued this combination taught every limitation of the challenged claims. Kohavi disclosed a general "wrapper" method for RFE that iteratively removes features to find an optimal subset, using a generic "induction algorithm" as a black box. Boser disclosed a complete SVM system, a known type of induction algorithm that could be used in Kohavi's wrapper. Boser’s SVM trains on data to create a decision function with optimized weights to minimize classification error. Hocking taught a computationally efficient method for feature selection in linear regression models by computing a ranking criterion based on the model's weight vectors and eliminating the variable with the smallest rank.
    • Motivation to Combine: A Person of Ordinary Skill in the Art (POSITA) would combine Kohavi's RFE method with Boser's SVM because Kohavi's method required an induction algorithm, and Boser’s well-known SVM was a suitable choice. A POSITA would have been further motivated to incorporate Hocking's weight-based ranking criterion into the Kohavi/Boser system. This would be a trivial substitution to improve the computational efficiency of the feature selection process, as Hocking’s method of using weight vectors to rank features was directly applicable to the linear model underlying Boser's SVM.
    • Expectation of Success: A POSITA would have had a reasonable expectation of success because Boser's SVM is a type of induction algorithm explicitly contemplated by Kohavi's wrapper method. Furthermore, applying Hocking's established statistical technique for feature ranking to the weights generated by Boser's SVM was a predictable application of a known method to a known system.

Ground 2: Obviousness over Kohavi, Boser, Hocking, and Cristianini - Claims 8 and 13 are obvious.

  • Prior Art Relied Upon: The combination from Ground 1, further in view of Cristianini (a 2000 textbook, "An Introduction to Support Vector Machines").
  • Core Argument for this Ground:
    • Prior Art Mapping: This ground specifically addressed limitations requiring the use of "gene expression data obtained from DNA microarrays." While Petitioner argued Kohavi's disclosure of a "DNA dataset" was sufficient, Cristianini was cited as an alternative. Cristianini explicitly described the application of SVMs to biological data mining, including the "automatic categorisation of gene expression data from DNA microarrays."
    • Motivation to Combine: A POSITA looking to apply the Kohavi/Boser/Hocking SVM-RFE system to biological problems would have consulted a reference like Cristianini, a primary textbook on SVM applications. Cristianini's express teaching of using SVMs for gene expression analysis would directly motivate its application in the primary combination to arrive at the invention of claims 8 and 13.

Ground 3: Obviousness over Kohavi, Boser, Hocking, and Cortes - Claim 2 is obvious.

  • Prior Art Relied Upon: The combination from Ground 1, further in view of Cortes (a 1995 article, "Support-Vector Networks").

  • Core Argument for this Ground:

    • Prior Art Mapping: This ground addressed claim 2, which requires a "soft margin support vector machine." Cortes, a foundational paper on SVMs, explicitly taught the concept of a soft margin SVM. This technique allows for some misclassifications in the training data to achieve better overall generalization, particularly with non-linearly separable data.
    • Motivation to Combine: A POSITA would have been motivated to use the soft margin SVM taught by Cortes to improve the robustness and generalization of the base Kohavi/Boser system. Since avoiding overfitting is a primary goal in machine learning, incorporating the well-known soft margin technique from a seminal work like Cortes would have been an obvious modification to enhance the classifier's performance.

  • Additional Grounds: Petitioner asserted additional obviousness challenges for other dependent claims. These grounds relied on the primary combination of Kohavi, Boser, and Hocking in view of single additional references teaching: pre-processing data for comparable scaling (Saunders), using clustering as a pre- or post-processing step (Castelli), and pre-processing to reduce data skew (Burros).

4. Key Claim Construction Positions

  • "data": Petitioner argued this term should be construed as "biological data." This construction was based on an explicit statement in the ’188 patent specification that the terms are used interchangeably.

5. Key Technical Contentions (Beyond Claim Construction)

  • Priority Date Challenge: A central argument was that the ’188 patent was not entitled to its claimed 1999 priority date and its effective date was no earlier than August 7, 2000. Petitioner alleged the priority chain was broken because an intermediate patent in the chain failed to claim priority to earlier applications and because there was no overlap of inventors between the ’188 patent and earlier patents in the chain. This later priority date is critical for establishing the asserted references as prior art.

6. Arguments Regarding Discretionary Denial

  • Petitioner argued against discretionary denial under Fintiv. The petition asserted that the parallel district court trial was scheduled only two months before the FWD deadline and that trial dates were highly uncertain due to the COVID-19 pandemic. Petitioner also stipulated that it would not pursue the same invalidity grounds in the district court if the IPR was instituted, which it argued eliminated concerns of duplicative efforts and weighed against denial.

7. Relief Requested

  • Petitioner requested institution of an inter partes review and cancellation of claims 1-23 of the ’188 patent as unpatentable under 35 U.S.C. §103.