Improved accuracy of a statistical prediction is an enhancement to the abstract idea, not a technological improvement

2021-03-25Fed. Cir.

In re Board of Trustees of the Leland Stanford Junior University, 991 F.3d 1245, No. 2020-1288 (Fed. Cir. Mar. 25, 2021) (precedential) (Reyna, J., joined by Prost, C.J., and Lourie, J.), affirming the Patent Trial and Appeal Board's rejection of claims 1 and 22 through 43 of United States Patent Application No. 13/486,982

United States Patent Application No. 13/486,982, filed by Stanford on June 1, 2012, claims a computerized method for inferring haplotype phase in a collection of unrelated individuals using a statistical model the specification calls PHASE-EM. Claim 1 recites receiving genotype data, imputing an initial haplotype phase for each individual based on a statistical model, building a data structure describing a hidden Markov model containing the imputed phases and parameters comprising local recombination rates and mutation rates, repeatedly randomly modifying imputed phases to recompute the parameters, replacing a phase when the new parameters indicate the modified phase is more likely, extracting a final predicted phase, and storing it. The specification states that PHASE-EM performs optimization on haplotypes rather than Markov chain Monte Carlo sampling. The Federal Circuit, in a second opinion by Judge Reyna for the panel that decided the related appeal on March 11, 2021, affirmed the Board. At step one the claims are directed to mathematical calculations and statistical modeling and recite no application, concrete or otherwise, beyond storing the haplotype phase. Stanford forfeited its computational-efficiency argument by not raising it before the Board, and the improvement in computational accuracy alleged does not qualify as an improvement to a technological process but is merely an enhancement to the abstract mathematical calculation of haplotype phase itself. At step two the receiving, extracting and storing steps were conventional.

What it changesRead with the March 11 decision on the related application, this opinion closes the argument that a model is eligible because it is more accurate. Accuracy is a property of the calculation, and the court treats a better calculation as a new abstract idea. Two points are practical. First, the efficiency argument was forfeited, not rejected: an applicant who can show that the claimed method runs with less memory or fewer operations on the same hardware should raise that improvement before the examiner and the Board, in the specification and the claims, because it is the one technical improvement the court left open. Second, the claim drafted around a data structure and a parameter-update loop was still a calculation, so reciting the training loop is not by itself a path to eligibility. Machine learning applications should be written with a claimed downstream use of the inference, or with a claimed change to the computing process, and the specification should state the computational cost comparison in numbers.

More on Eligibility of software and AI claims

← All recent developments · Search all developments

Burbage Law is the practice of Jon Burbage, a registered patent attorney. 342 United States patents have issued from applications he wrote; the full list is on the Patents page. More about Jon Burbage.

This is general information about a published decision, agency guidance or piece of legislation. It is not legal advice and does not create an attorney-client relationship. Items concerning offices outside the United States are reported for information only and are not counsel on the law of those countries.