Health Discovery Corp. v. Intel Corp.
| Decision Date | 27 December 2021 |
| Docket Number | 6:20-cv-666-ADA |
| Citation | 577 F.Supp.3d 570 |
| Parties | HEALTH DISCOVERY CORPORATION, Plaintiff, v. INTEL CORPORATION, Defendant. |
| Court | U.S. District Court — Western District of Texas |
Charles E. Cantine, Pro Hac Vice, Joseph Diamante, Pro Hac Vice, Dunlap Bennett & Ludwig PLLC, New York, NY, Eleanor Musick, Pro Hac Vice, Musick Davison LLP, San Diego, CA, Erick Scott Robinson, Spencer Fane LLP, Houston, TX, for Plaintiff.
Allison H. Altersohn, King & Spalding LLP, New York, NY, Brent P. Ray, King & Spalding LLP, Chicago, IL, Bryan S. Banks, Perkins Coie LLP, Phoenix, AZ, Christopher Marth, Pro Hac Vice, James S. Miller, Pro Hac Vice, Perkins Coie LLP, Seattle, WA, Lauren May Eaton, Sarah E. Piepmeier, Perkins Coie, LLP, San Francisco, CA, Lori A. Gordon, King & Spalding LLP, 1700 Pennsylvania Avenue, NW, Jose Carlos Villarreal, Perkins Coie LLP, Steven Mark Zager, King & Spalding LLP, Austin, TX, Kourtney Mueller Merrill, Perkins Coie LLP, Denver, CO, for Defendant.
Came on for consideration this date is Intel Corporation's Motion to Dismiss (the "Motion"), filed October 19, 2020. ECF No. 12. Health Discovery Corporation ("Plaintiff" or "HDC") filed a response on November 23, 2020, ECF No. 21, to which Intel Corporation ("Defendant" or "Intel") replied on December 7, 2020, ECF No. 25. The Court held a hearing on the Motion on September 28, 2021. See ECF No. 57. After careful consideration of the Motion, the Parties’ briefs, oral arguments, and the applicable law, the Court GRANTS-IN-PART and DENIES-AS-MOOT-IN-PART Intel's Motion to Dismiss. The Court GRANTS Intel's Motion to the extent it moves to dismiss under 35 U.S.C. § 101 and DENIES-AS-MOOT Intel's Motion to the extent it moves to dismiss for a failure to sufficiently plead direct and indirect infringement under Rule 12(b)(6).
On July 23, 2020, HDC filed a complaint accusing Intel of infringing U.S. Patent Nos. 7,117,188 (the "’188 patent"), 7,542,959, 8,095,483, and 10,402,685 (collectively, the "Asserted Patents"). See ECF No. 1 ¶¶ 15–18. (HDC states that these patents share a "substantially common specification," ECF No. 21 at 1 n.1, so this Order's reference to the ’188 patent ’s written description refers to that shared specification.) The complaint states that each of HDC's asserted patents "relate[s] to innovative technology for using learning machines (e.g. , Support Vector Machines) to identify relevant patterns in datasets, and more specifically, to a selection of features within the datasets that best enable classification of the data (e.g. , Recursive Feature Elimination)." ECF No. 1 ¶ 27. HDC accuses Intel of infringing its patents directly and indirectly through, for example, the use, sale, and marketing of "Intel AI-optimizing/machine learning processors," field programmable gate arrays, system on chips, and software deployed on "Intel/Intel-partnered computers" and other architectures. ECF No. 1 ¶¶ 51, 74, 78.
On October 19, 2020, Intel moved to dismiss HDC's complaint with prejudice under Federal Rule of Civil Procedure 12(b)(6) for asserting claims that are invalid under 35 U.S.C. § 101 and failing to sufficiently plead direct and indirect infringement. See ECF No. 12 at 1–2. That Motion is now fully briefed and ripe for judgment.
The inventors of the Asserted Patents, Dr. Isabelle Guyon and Dr. Jason Weston, "are widely recognized as being among the most influential scholars in the field" of machine learning. ECF No. 1 ¶ 22. At the time the common specification was drafted, genomic sequencing produced a daunting amount of data—"regarding the sequence, regulation, activation, binding sites and internal coding signals." ’188 patent at 2:14–16. But isolating valuable data presented a challenge. Id. at 2:16–17. To be sure, traditional methods of data analysis could generate interesting and relevant information, but they could not "intelligently and automatically assist humans in analyzing and finding patterns of useful knowledge." Id. at 3:21–23. Human researchers turned to more advanced technology—machine learning algorithms like neural networks, to identify relevant patterns. Id. at 3:30–43. Even these produced "crude models of the underlying processes," id. at 2:18–22, and were limited by the "curse of dimensionality"—as the dimensions of the data set increased, the processing time and power increased disproportionately, id. at 3:65–4:3.
More advanced machine learning technology, like support vector machines ("SVM"), avoided those issues. An SVM:
maps input vectors into high dimensional feature space through non-linear mapping function, chosen a priori. In this high dimensional feature space, an optimal separating hyperplane is constructed. The optimal hyperplane is then used to determine things such as class separations, regression fit, or accuracy in density estimation.
Id. at 4:5–11. SVMs can process high-dimensionality data sets without concern for the curse of dimensionality. Id. at 4:12–20.
But SVMs are not perfect. When a machine learning algorithm like an SVM is trained with only a few training profiles—for example, the gene profiles of a few dozen patients—but each training profile includes a high number of features—"thousands of genes studied in a microarray"—the algorithm risks "overfitting." Id. at 25:29–43. That is to say, the SVM will accurately predict patterns for its training profiles but fails to do so when presented with new profiles. See id. Addressing this issue requires a reduction in feature size, which may be pursued by ranking features, and eliminating the lowest ranked features. Id. at 25:56–26:9. "Previous attempts to address this problem used correlation techniques, i.e., assigning a coefficient to the strength of association between variables." Id. at 24:34–37. One specific example using correlation coefficients referred to throughout the patents is that of T.K. Golub. See, e.g. , id. at 26:20–62.
Recursive feature elimination ("RFE") may be used to reduce features. "RFE methods comprise iteratively 1) training the classifier, 2) computing the ranking criterion for all features, and 3) removing the feature having the smallest ranking criterion." Id. at 27:62–66. This iterative process eventually produces nested subsets of features "of increasing informative density." Id. at 53:50–60. And these subsets can then be put into an SVM for pattern recognition. Id. at 53:61–66.
The asserted patents’ claims are directed to performing feature ranking, selection, and reduction using an SVM itself to facilitate an RFE process on a large dataset. The SVM analysis acts as the classifier, producing weight values for each feature in the set, and those values are then used to generate each feature's ranking criterion. See id. at 29:12–58. The feature(s) with the smallest ranking criterion are eliminated. See id. The process then begins again until a certain number of features remain.
According to the asserted patents’ written description, this SVM-RFE method can, relative to prior art methods, "provide subsets of genes that are both smaller and more discriminant." Id. at 39:52–54. Discriminant identification is "beneficial in confirming recent discoveries in research or in suggesting avenues for research or treatment." Id. at 24:51–60. The written description repeatedly compares conventional gene selection methods with the claimed SVM-RFE method, stating that SVM-RFE "provides the best results down to 4 genes." Id. at 49:31–38. It discards "genes that are tissue composition-related and keeps genes that are relevant to the cancer
vs. normal separation." Id. ; see also id. at 48:66–11; 49:46–58. Use of the SVM-RFE can "make a quantitative difference ... with better classification accuracy and smaller gene subset, but [it] also makes a qualitative difference in that the gene set is free of" noise like "tissue composition related genes." Id. at 44:31–35. This "[u]se of RFE provides better feature selection than can be obtained by using the weights of a single classifier" and it "consistently outperforms naive ranking, particularly for small feature subsets." Id. at 30:8–10; 30:19–23.
Section 101 defines subject matter eligible for patenting as "any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof." 35 U.S.C. § 101.1 The Supreme Court has long read exceptions into § 101 : laws of nature, natural phenomena, and abstract ideas are not patentable. See, e.g. , Diamond v. Diehr , 450 U.S. 175, 185, 101 S.Ct. 1048, 67 L.Ed.2d 155 (1981). According to the Court, these are the fundamental tools of scientific endeavor and granting monopolies over them risks dousing the flame of innovation the U.S. patent regime is meant to fan. See Gottschalk v. Benson , 409 U.S. 63, 67, 93 S.Ct. 253, 34 L.Ed.2d 273 (1972) ; Mayo Collaborative Servs. v. Prometheus Lab'ys, Inc. , 566 U.S. 66, 71, 132 S.Ct. 1289, 182 L.Ed.2d 321 (2012) ; see also Le Roy v. Tatham , 55 U.S. (14 How.) 156, 175, 14 L.Ed. 367 (1852) ().
In recent years, divining the bounds of these judicial exceptions has proved increasingly challenging, thanks in large part to the Supreme Court's 2014 decision in Alice Corp. Pty. v. CLS Bank Int'l , 573 U.S. 208, 134 S.Ct. 2347, 189 L.Ed.2d 296 (2014). There the Court established a two-step framework for determining whether a patent claims an ineligible concept. First, determine whether the claims are "directed to" a judicial exception. Id. at 217, 134 S.Ct. 2347. If so, proceed to the second step and "consider the elements of each claim both individually...
To continue reading
Request your trial
-
Axis Surplus Ins. Co. v. Mercer
... ... 2505, 91 L.Ed.2d 202 (1986) ; Celotex Corp. v. Catrett , 477 U.S. 317, 322, 106 S.Ct. 2548, 91 L.Ed.2d ... 2004) ; Principal Health Care of La., Inc. v. Lewer Agency, Inc. , 38 F.3d 240, 242 ... ...