Patent Drafting for Machine Learning: Structural Claim...

Patent Drafting for Machine Learning:

Structural Claim Limitations,

Avoiding §101 or §112 Rejections

Today’s faculty features:

1pm Eastern | 12pm Central | 11am Mountain | 10am Pacific

The audio portion of the conference may be accessed via the telephone or by using your computer's

speakers. Please refer to the instructions emailed to registrants for additional information. If you

have any questions, please contact Customer Service at 1-800-926-7926 ext. 1.

TUESDAY, FEBRUARY 13, 2018

Presenting a live 90-minute webinar with interactive Q&A

Gregory Rabin, Patent Attorney, Schwegman Lundberg & Woessner, New York

Michael D. Stein, Partner, Baker & Hostetler, Seattle

Tips for Optimal Quality

Sound Quality

If you are listening via your computer speakers, please note that the quality

of your sound will vary depending on the speed and quality of your internet

connection.

If the sound quality is not satisfactory, you may listen via the phone: dial

1-866-819-0113 and enter your PIN when prompted. Otherwise, please

send us a chat or e-mail [email protected] immediately so we can address

the problem.

If you dialed in and have any difficulties during the call, press *0 for assistance.

Viewing Quality

To maximize your screen, press the F11 key on your keyboard. To exit full screen,

press the F11 key again.

FOR LIVE EVENT ONLY

Continuing Education Credits

In order for us to process your continuing education credit, you must confirm your

participation in this webinar by completing and submitting the Attendance

Affirmation/Evaluation after the webinar.

A link to the Attendance Affirmation/Evaluation will be in the thank you email

that you will receive immediately following the program.

For additional information about continuing education, call us at 1-800-926-7926

ext. 2.

FOR LIVE EVENT ONLY

Copyright 2017 Schwegman Lundberg & Woessner. P.A. All Rights Reserved.

Patent Drafting for Machine

Learning: Structural Claim

Limitations, Avoiding 101 or 112

Rejections

Greg Rabin

[email protected]

February 13, 2018


Introduction

Greg Rabin

[email protected]

www.linkedin.com/in/gregrabin

Patent attorney at Schwegman, Lundberg, and Woessner, PC, based in Ithaca, NY.

5


Subject of this Talk

Anticipating and Avoiding 101 Rejections for Machine Learning

6


Disclaimer

• The subject matter of this presentation is the speaker’s opinion only, and does not necessarily correspond to the opinions of the speaker’s employer or clients.

• The information in this presentation may not correspond to constantly changing information/ guidance provided by the courts or the USPTO.

7


What is Artificial Intelligence?

• Artificial intelligence (AI) is the ability of a computer program or a machine to think and learn. It is also a field of study which tries to make computers “smart.”

• Example use cases include: understanding human speech, playing strategic games (e.g., chess), self-driving cars, and interpreting complex data.

• An extreme goal of AI is to create computers that can learn, solve problems, and think logically.

• AI involves a combination of many different fields, such as computer science, mathematics, linguistics, psychology, neuroscience, and philosophy.

8


Long-Term Problems in AI

• Creating a “general artificial intelligence,” which can solve many different problems, instead of focusing on just one.

• Including making new discoveries or inventions???

• Developing a machine capable of “perceiving its environment” and making decisions or taking actions accordingly, like a human.

9


AI Sub-Field

Search and Optimization

• Intelligently searching through many possible solutions, taking advantage of machine’s high processing speed.

10


AI Sub-Field

Neural Networks

• Neural networks are modeled after the neurons in the human brain, where a trained algorithm determines an output response for input signals. Multiple layers process an input, with each layer applying more and more complex processing.

11


AI Sub-Field

Control Theory

• Control of continuously operating dynamical systems in engineered processes and machines. The objective is to develop a control model for controlling such systems using a control action in an optimum manner without delay or overshoot and ensuring control stability.

12


What is Machine Learning?

• Machine Learning gives “computers the ability to learn without being explicitly programmed.”

• Machine learning explores the study and construction of algorithms that can learn from and make predictions on data – such algorithms overcome following strictly static program instructions by making data-driven predictions or decisions, through building a model from sample inputs.

• Example applications include email filtering, detection of network intruders, optical character recognition (OCR), and computer vision.

• Source: Wikipedia: Machine Learning [https://en.wikipedia.org/wiki/Machine_learning]

13


Types of Machine Learning

Supervised Learning: The computer is presented with example inputs and their desired outputs, given by a “teacher,” and the goal is to learn a general rule that maps inputs to outputs.

– E.g. The animals in the picture are labeled as “cat” or “dog”, and the computer trains itself to identify other animals, in other pictures, as either “cat” or “dog”.

14



Unsupervised Learning: No labels are given to the learning algorithm, leaving it on its own to find structure in its input. Unsupervised learning can be a goal in itself (discovering hidden patterns in data) or a means towards an end (feature learning).

– E.g. Classify the animals below into different types.

15



Reinforcement Learning: A computer program interacts with a dynamic environment in which it must perform a certain goal (such as driving a vehicle or playing a game against an opponent). The program is provided feedback in terms of rewards and punishments as it navigates its problem space.

– E.g. Learn to drive…

16


Example ML Claim Ver. A

1. A method comprising: accessing, at one or more computing devices, an image; determining whether the accessed image includes a cat; and providing an output indicating whether the accessed image includes the cat.

17



This claim is directed to cat-recognition technology. It is likely invalid under 35 U.S.C. 101 because it preempts all other cat-recognition technologies. Every algorithm that identifies cats will access an image, determine whether the image includes a cat, and provide a corresponding output.

18



This claim might become patentable by…

• Adding claim limitations related to training the machine learning (Ver. B).

• Adding claim limitations related to the feature vector used in the machine learning (Ver. C).

• Adding claim limitations related to the output of the machine learning (Ver. D).

19


Example ML Claim Ver. B

1. A method comprising: accessing, at one or more computing devices, an image; determining, using an image classification engine stored in one or more memories of the one or more computing devices, whether the accessed image includes a cat, the image classification engine being trained, using unsupervised learning, to classify images into a plurality of classes and being notified, via a graphical user interface (GUI), which classes correspond to cats, the GUI displaying at least one image from each class of the plurality of classes; and providing an output indicating whether the accessed image includes the cat.

20



The method claim in Version B does not preempt all use of “cat-recognition technology.”

Rather, the method claim details a specific technique by which the one or more computing devices are trained to recognize cats.

21



Other inventors could possibly invent other “cat-recognition” techniques that include different operations/ components from those listed in Version B of claim 1, for example, supervised learning or reinforcement learning could be used.

22



It should be noted that the Version B claim might still be invalid and some Examiners may insist on more implementation details being added to the claim before finding it patentable under 35 U.S.C. 101.

23


Example ML Claim Ver. C

1. A method comprising: accessing, at one or more computing devices, an image; determining, based on a feature vector comprising a plurality of features that are extracted from the accessed image, whether the accessed image includes a cat, the plurality of features comprising at least an identification of a portion of the accessed image as representing a paw, a symmetry of the paw, and a shape of the paw; and providing an output indicating whether the accessed image includes the cat.

24



The method claim in Version C does not preempt all use of “cat-recognition technology.”

Rather, the method claim details a specific technique by which the one or more computing devices recognize cats using explicitly specified features of a feature vector.

25



Other inventors could possibly invent other “cat-recognition” techniques that include different operations/ components from those listed in Version C of claim 1. For example, features other than the paw can be used to recognize the cat.

26



It should be noted that the Version C claim might still be invalid and some Examiners may insist on more implementation details being added to the claim before finding it patentable under 35 U.S.C. 101.

27


Example ML Claim Ver. D

1. A method comprising: accessing, at one or more computing devices, an image; determining whether the accessed image includes a cat; and providing an output indicating whether the accessed image includes the cat, the output comprising overlaying the accessed image with a first text if the accessed image includes the cat, and overlaying the accessed image with a second text, different from the first text, if the accessed image does not include the cat.

28



The method claim in Version D does not preempt all use of “cat-recognition technology.”

Rather, the method claim details a specific technique by which the one or more computing devices output information representing that a cat was recognized in an image.

29



Other inventors could possibly invent other “cat-recognition” techniques that include different operations/ components from those listed in Version D of claim 1, for example, the output could be presented in a different format.

30



It should be noted that the Version D claim might still be invalid and some Examiners may insist on more implementation details being added to the claim before finding it patentable under 35 U.S.C. 101.

31


A Real-Life Example from Public PAIR

ADDITIONAL DISCLAIMER

The real-life example described here is not, in any way, representative of a 101 trend. The speaker makes no guarantees that this claim, or any similar claim, will be upheld in litigation, in review by the Patent Office, or in another patent application.

The real-life example is only one example of a published patent. The slides contain excepts from the file wrapper history and do not include the entire history.

32



The real-life example corresponds to a US patent application, filed in February 2016, which lead to a patent that issued in August 2017.

The real-life example relates to uses of machine learning in a question answering system.

33



Note: All data used in this presentation related to the above-referenced patent application is copied from Public PAIR, patft.uspto.gov, patents.google.com, and other publicly accessible sources. The speaker and the speaker’s employer were not involved in the drafting or prosecution of the above-noted patent application.

34


Claim Rejected Under 35 U.S.C. 101

1. A method, in a data processing system, for question answering using multi-instance learning, the method comprising: training an answer ranking multi-instance learned model using a ground truth question and answer-key pair set; receiving an input question from a user; generating one or more candidate answers to the input question, wherein each of the one or more candidate answers has an associated set of supporting passages; determining a confidence value for each of the one or more candidate answers using the answer ranking multi-instance learned model based on the set of supporting passages; ranking the one or more candidate answers by confidence value to form a ranked set of answers; and presenting a final answer from the ranked set of answers, the confidence value for the final answer, and supporting evidence for the final answer to the user.

35


Amendment Leading to Patent-Eligibility Under 101

1. A method, in a data processing system comprising at least one processor and at least one memory, the at least one memory comprising instructions executed by the at least one processor to implement a question answering system, for question answering using multi-instance learning, the method comprising: training an answer ranking multi-instance learned model using a ground truth question and answer-key pair set, wherein the data processing system executes in accordance with the answer ranking multi-instance learned model to implement the question answering system, wherein the answer-ranking multi-instance learned model is trained based on whether a passage set returned for each answer in the question and answer-key pairs set collectively does or does not support a correct answer; receiving, by the question answering system, an input question from a user;

36


Amendment Leading to Patent-Eligibility Under 101

generating, by a hypothesis generation phase of a question answering pipeline of the question answering system, one or more candidate answers to the input question, wherein each of the one or more candidate answers has an associated set of supporting passages; determining, by a hypothesis and evidence scoring phase of the question answering pipeline of the question answering system, a confidence value for each of the one or more candidate answers using the answer ranking multi-instance learned model based on the set of supporting passages; ranking, by an answer ranking phase of the question answering pipeline of the question answering system, the one or more candidate answers by confidence value to form a ranked set of answers; and presenting, by the question answering system, a final answer from the ranked set of answers, the confidence value for the final answer, and supporting evidence for the final answer to the user.

37


Result of this Amendment

• 35 U.S.C. 101 (patent-eligibility) rejection was withdrawn.

• 35 U.S.C. 102 (novelty) rejection was withdrawn.

• A new rejection under 35 U.S.C. 103 (non-obviousness) was entered.

• The Applicants eventually overcame the 35 U.S.C. 103 rejection, and a US Patent was issued in August 2017.

38


Discussion of Claim Amendments

The response states: “Applicants… amend claim 1… to more specifically recite that the data processing system executes in accordance with an answer ranking multi-instance learned model to implement the question answering system and to recite a hypothesis generation phase, a hypothesis and evidence scoring phase, and an answer ranking phase of a question answering pipeline of the question answering system… Therefore, independent claim 1… recites a specific configuration of, and improvement to, a data processing system.”

39


Discussion of Claim Amendments

The amended claim specifically recites the training mechanism used, and the phases for generating the answer. (Similar to Version B of the cat-recognition claim).

Note that the amended claim does not preempt all “question answering” solutions because other solutions could use a different training mechanism and, possibly, different phases for generating the answer.

40


Take Away Points

Claiming features related to the machine training mechanism, feature extraction, or output formats may be sufficient to overcome a 35 U.S.C. 101 rejection for a machine learning-related patent application.

However, nothing is guaranteed, and it is necessary to consider the specific features of any given invention before deciding which claim amendments or arguments would best advance prosecution in view of a 35 U.S.C. 101 rejection.

41


101 Rejections by Art Unit

101/Alice rejections are more commonly issued in some technology centers/ art units than in others. One strategy may be to draft a patent application to encourage or avoid its entry into one of the technology centers/ art units.

Source: http://www.ipwatchdog.com/2015/12/14/the-most-likely-art-units-for-alice-rejections/

42


Conclusion

Hopefully, in the near future, the USPTO will issue clear guidelines regarding which software (and machine learning) inventions are patentable under 35 U.S.C. 101 in view of Alice and the related cases.

43

Thank you for your participation.

Questions?

Greg Rabin

[email protected]

Protecting Software Inventions

Including AI and ML Algorithms

Michael D. Stein,

BakerHostetler LLP

Presented for Strafford Patent Drafting for Machine Learning: Structural Claim Limitations, Avoiding 101 or 112 Rejections Feb. 13, 2018

Outline

1. Patenting AI/ML Inventions

a. Software patents and functional claim

limitations

b. Structural claim limitations for software

c. Inventorship questions

2. Trade Secret Protection

3. Strategic Use of Patent and Trade Secret

Protection

46

I. Problem Overview

47

Litigation USPTO

Legal Solutions: Use § 101 to invalidate overly

broad ( abstract ) claims (Alice/Mayo)

Use § 112(b) to invalidate unclear claims (Aristocrat)

Alternative Solution: Include structural

limitations in claims for software-related inventions.

Applicants and USPTO ensure patent file history reflects examiner s consideration of § 101 and § 112 issues.

Provide context for courts.

Problem: Courts lack guidance/procedure

to effectively apply the law. Courts lack context for

evaluating claims from perspective of POSITA.

Too many patents are being invalidated on incorrect grounds!

Background

Critics argue:

• Patents do less good and cause more harm in the software industry than in other industries.

• The aggregation of overbroad and unclear claims can result in innovation-stifling patent thickets.

• Software patents are not necessary to spur innovation because software innovation is less costly than innovation in the life sciences.

48

USPTO-led Executive Action for Clarity in

Patent Claims

49

Why functional claim limitations

are used in software patents

Why functional claiming is so common in software patents:

1. Software structure and function can be separated. A software developer can write new software without knowing details of the hardware.

2. The software industry lacks a commonly accepted “vocabulary” for defining software elements. In software, a broad claim requires defining the invention at a higher level of abstraction. Software developers coin new terms to define the functional elements of the software, and the meaning and scope of such new terms is often not explicitly defined.

50

Functional claim limitations

and 35 USC §112(f)

51

35 U.S.C. 112 Specification.

. . .

(f) ELEMENT IN CLAIM FOR A COMBINATION.—

An element in a claim for a combination may be

expressed as a means or step for performing a

specified function without the recital of structure,

material, or acts in support thereof, and such claim

shall be construed to cover the corresponding

structure, material, or acts described in the

specification and equivalents thereof.

Aristocrat Techs. Australia Pty Ltd, v. Int’l Game

Tech., 521 F.3d 1328 (Fed. Cir. 2008)

52

Claim Disclosed Structure

53

Data structures Program structures

Structural

claim

limitations

for software

Describing computer software in

terms of “structure”

54

Essential feature(not shown): Relationship

between the executable software and

hardware processor(s).

• The executable code is not an

unconstrained, abstract set of commands

written down by the programmer.

• The executable code must be committed to

memory using “machine codes” selected

from the specific machine language

instruction set, or “native instructions,”

designed into the hardware processor.

• The native instruction set is known to, and

essentially built into, the hardware

processor(s).

• Each native instruction is a discrete code

recognized by the processing architecture

and can specify particular registers for

arithmetic, addressing, or control functions.

• Complex operations are built up by

combining the simple native instructions.

Describing computer software in

terms of “structure”

55

Key ideas:

• Relationship between executable

software instructions and

hardware processor constrains

the claimed system.

• Problems related to abstractness,

indefiniteness, and overbreadth

can be addressed by describing

inventive computer software

elements in the specification at

this level of detail, and including

appropriate limitations in the

claims.

• I do not advocate inclusion of

specific machine code or even

source code listings in patent

applications.

Example: Artificial Neural Network

56

Example claim to computer-implemented system

comprising an ANN

1. A classification system, comprising:

a processor configured to execute instructions programmed using a predefined set of machine

codes; and

an artificial neural network (ANN) comprising:

first, second, and third input nodes, wherein each input node includes a memory location for

storing an input value;

first, second, third, and fourth hidden nodes, wherein each hidden node is connected to each

input node and includes computational instructions, implemented in machine codes of the

processor, for computing first, second, third, and fourth output numbers, respectively; and

first and second output nodes, wherein the first output node includes a memory location for

storing a first output signal indicative of a first classification, and the second output node

includes a memory location for storing a second output signal indicative of a second

classification.

57

Example: Linked List Data Structure and

Linear Search Algorithm

58

Boolean logic expressions and operators: Logical operands and logical operators are combined together. Operands

are statements (that can be proven True or False) and the operators are logical AND, OR and NOT. Boolean expressions

involve comparison operators that can be evaluated to determine if they are True or False. Comparison operators

include: = < > ≥ ≤ ≠.

Expressions combining Boolean operators and comparison operators are written and evaluated in computer

programming languages using control statements that control which sections of code in a program are executed.

Decision or Conditional statements are a type of Control statement and are often referred to as IF..THEN..ELSE

statements.

Iterative constructs, or Loops, allow a section of code to be repeated. There are several variations to iterative or

looping constructs but for the most part they fall into two categories: FOR loops and WHILE/DO loops.

Pseudocode for a linear search algorithm implemented with Boolean logic and FOR and IF..THEN loops:

procedure linear_search (list, value)

for each item in the list

if match item == value

return the item's location

end if

end for

end procedure

Example claim to computer-implemented search

system comprising a linked list data structure

59

1. A search system, comprising:

a processor configured to execute instructions programmed using a predefined set of machine codes;

memory coupled to the processor;

a linked-list data structure implemented in said memory, comprising a first node, a plurality of intermediate nodes, and a terminator;

wherein the first node is connected to a first intermediate node;

wherein each of intermediate nodes except for a last intermediate node is connected to a subsequent one of the intermediate nodes; and

wherein the last intermediate node is connected to the terminator; and

a linear search algorithm implemented in machine codes for the processor.

2. The system of claim 1, wherein the linear search algorithm comprises an iterative construct and a Boolean logic expression.

PTAB Example: Ex parte Boucher (Dec. 8, 2017)

1. A method for a computer for assisting the maintenance of a system of an aircraft comprising … a diagnostic device that performs computer-readable instructions using a processor for executing the steps of the method comprising:

providing an onboard aircraft database in the aircraft, …

detecting, using the detector, a fault of the system of the aircraft;

identifying… said each item of the list of said at least one item being liable to be the cause of said detected fault…

determining … said probability of the corresponding cause of the identified fault …

identifying … said at least one item of said equipment of the aircraft actually being the cause of said detected fault;

updating… an onboard temporary database on a basis of said fault identifier …

sending … a content of said onboard temporary database to said remote ground database …

erasing … said onboard temporary database upon completion of the update …

determining … said probability of the corresponding cause of the identified fault …; and

storing both of the onboard aircraft and temporary databases in the system of said aircraft.

61

PTAB Example: Ex parte Blackwell (Nov. 2, 2017)

1. A method comprising:

providing search results in a first dedicated screen space of a user interface and associated with a first collection

based on at least one search term, and which excludes search results associated with any of one or more second

collections;

providing a separate customizable preview of search results … in a separate pane located in a second dedicated

screen space of the user interface …; and

providing options to a user to customize the second dedicated space and the separate pane …in the user interface

that is configured to provide customization controls which:

allow the user to remove the separate pane in the second dedicated space;

order the separate pane with at least one other customizable preview pane in the second dedicated space; and

allow the user to add another pane in the second dedicated space …,

wherein the first collection is a first category of information for the at least one search term and the at least one

second collection includes a second category of information for the at least one search term which is different from

the first category of information.

62

Inventorship Questions

Who is the inventor?

• 35 U.S.C. 100 (definitions) – (f) The term "inventor" means the individual or, if a joint

invention, the individuals collectively who invented or discovered the subject matter of the invention.

– (g) The terms "joint inventor" and "coinventor" mean any 1 of the individuals who invented or discovered the subject matter of a joint invention

• 35 U.S.C. 115 (Inventor’s oath or declaration) – (a) NAMING THE INVENTOR; … each individual who is

the inventor or a joint inventor of a claimed invention in an application for patent shall execute an oath or declaration …

63

Inventorship Questions

Improper Naming of Inventors

• MPEP 2157 Improper Naming of Inventors – Although the AIA eliminated pre-AIA 35 U.S.C. 102(f), the patent laws

still require the naming of the actual inventor … The Office presumes that the named inventor or joint inventors in the application are the actual inventor or joint inventors …

• 37 CFR 1.56 Duty to disclose information material to patentability – … the Office is aware of and evaluates the teachings of all information

material to patentability. Each individual associated with the filing and prosecution of a patent application has a duty of candor and good faith in dealing with the Office, which includes a duty to disclose to the Office all information known to that individual to be material to patentability as defined in this section.

Duty to Disclose “Information” on Inventorship

• MPEP 2001.04 Duty of Disclosure, Candor, and Good Faith In addition to prior art such as patents and publications, 37 CFR 1.56 includes, for example, information on enablement, possible prior public uses, sales, offers to sell, derived knowledge, prior invention by another, inventorship conflicts, and the like.

64

What Is a Trade Secret?

A trade secret is information ( formula, , pattern, compilation, program, device, method, technique, or process) that:

(1) derives economic value from not being generally known or readily ascertainable by proper means; and

(2) is the subject of "reasonable efforts" under the circumstances to maintain its secrecy.

Trade Secrets Can Be Used As Sword and Shield

Defend Trade Secrets Act (2016)

DTSA created a private, federal, civil right of action for alleged victims of trade secrets theft, as well as a uniform federal definition of “trade secret” applicable in federal criminal and civil cases.

The DTSA’s definition of “trade secret” appears substantively identical to the UTSA

35 U.S.C. §273 (prior commercial use defense)

In the presence of clear and convincing evidence that it had engaged in commercial use of the patented subject matter at least date, the “prior user” would be allowed free and clear historical and future use of that patented subject matter

65

Trade Secret Protection

66

Strategic Use of Patent & Trade Secret

Protection

Patent Protection Trade Secret Protection

ML Joke

A ML algorithm walks into bar…

• Bartender: What are you having?

• Algorithm: What is everyone else having?

67

Michael Stein

[email protected]

(206) 332-1384

68

Thank you!

Questions?

mailto:[email protected]

Patent Drafting for Machine Learning: Structural Claim...

Documents

Transcript of Patent Drafting for Machine Learning: Structural Claim...