INTELLECTPERITUS KNOWLEDGE SPHERE....We Share, We Grow!

Articles

DEEP LEARNING & PATENTS - CHALLENGES FOR RESEARCH & ANALYSIS

IKS Category: Artificial Intelligence and Patenting
IKS Article No: IKS_Article_06 January_22_2018
Compilation by: Chintan Gorasiya; Pritesh Gohel; Chintan Modi; Tejas Patel
Why This Article? We learned that Deep Learning / Machine Learning is an area in which searching prior art is a challenge. We accepted this challenge and we learned what the related areas are which pose challenges and what can be the remedies around those challenges.
Our Disclaimer: We undergo rigorous study of various materials available in books, journals, conference proceedings, trade magazines, catalogs/manuals, blogs and general web search. We have provided references as much as possible. Feel free to contact us if you believe that we have used any copyrighted material without providing references. We apologize.

CONTENTS:

  1. Abstract
  2. Introduction to Deep Learning:
  3. Hurdles for Patentability of Deep Learning:
    • Region - Europe
    • Region - USA

 

  1. Few solutions to overcome the Hurdles
    • Highlighting technical advantage
    • Describing implementation of mathematical method:
    • Functional claiming strategies:
    • Trade Secret protection:
    • Copyright protection:
  2. References:

  1. Abstract
  2. Deep Learning is recited among top three technologies of year 2017, which is expected to welcome high demand in next few years. It is predicted that the deep learning market is expected to be worth USD 1772.9 Million by 2022, growing at a CAGR of 65.3% between 2016 and 2022. This article mainly focuses on what is Deep Learning, overview of patentability hurdles for Deep Learning in Europe and USA region and few solutions to overcome the hurdles faced.

  3. Introduction to Deep Learning:
  4. Scientific American, in collaboration with the World Economic Forum, published a special report on The Top Ten Emerging Technologies of 2017, which recites Deep Learning among top three technologies which is expected to welcome high demand in next few years, based on selection by Global Panel Experts. [1]

    Although deep learning is not a new technique, having its roots in 1943, when Warren McCulloch and Walter Pitts create computational model for neural networks based on mathematics and algorithms called threshold logic, it caught major attention and resurgence of interest only after Google DeepMind's algorithm, AlphaGo masters art of the complex board game Go, and beats the professional go player, Lee Sedol at a highly publicized tournament in Seoul, in 2016. [2]

    Deep learning aka deep structured learning or hierarchical learning is branch of machine learning based on learning data representations rather than task-specific algorithms. Learning can be supervised, partially supervised or unsupervised. Deep learning or Artificial intelligence was once part of thriller and Science Fictions, who can forgot the Arnold Schwarzenegger's Terminator, but now it is a reality and is currently being used in various areas like healthcare, education, and finance in a very impressive way. Scientists had used deep learning algorithms with multiple processing layers (hence "deep") to make better models from large quantities of unlabeled data (such as photos with no description, voice recordings or videos on YouTube).

    Deep learning is a topic that is making big waves at the moment. Google's search engine, voice recognition system and self-driving cars all rely heavily on deep learning. Google has also created one program that picks out an attractive still from a YouTube video to use as a thumbnail.

    According to a market research report on deep learning, this market is expected to be worth USD 1772.9 Million by 2022, growing at a CAGR of 65.3% between 2016 and 2022. Report also indicates the advertisement, finance, and automotive as the major drivers for the growth of the market [3]. Many companies has invested their time and research to take advantage of speed and quality of Deep learning algorithm for expanding their business to a newer heights or in dynamic areas. Hence it is necessary, to protect their intellectual property.

    IP issues mainly have two business objectives, first one is maintaining freedom to operate without violating third-party rights, and protecting own investments in AI research and development. As deep learning is basically a computer algorithm to collect, recognize, analyze and/or categorize the audio or visual, mathematical data, it is non-patentable generally in majority of countries, leaving it as "open to use" invention although it demands tremendous development efforts.

  5. Hurdles for Patentability of Deep Learning:
    • Region - Europe
      • Exclusions from Patentability:
        • Under Article 52(2) EPC, following individual features of the claims fall within the exclusions, which reads as
          (2) The following in particular shall not be regarded as inventions within the meaning of paragraph 1:
          (a) Discoveries, scientific theories and mathematical methods;
          (b) Aesthetic creations;
          (c) Schemes, rules and methods for performing mental acts, playing games or doing business, and programs for computers;
          (d) Presentations of information
      • Prior Art:
        • Many of the underlying techniques are within the public knowledge in form of publications and repositories of electronic pre-prints such as arXiv
      • Domain of Invention:
        • Approaches to fields in engineering are more considered more positively
        • Approaches to fields in business or enterprise are more likely to be excluded on being non-technical
      • Mathematical Methods:
        • Field is closely linked to the field of statistics/ statistical methods
        • Mathematical representation are considered "non-technical" by EPO
      • Schemes, Rules and Methods for Performing Mental Acts:
        • A claim feature is likely to be considered part of schemes, rules and methods for performing mental acts when the scope of the feature is too broad or abstract. For example, if a claimed method step also covers a human being performing the step manually, it is likely that the scope is too broad
      • Schemes, Rules and Methods for Doing Business:
        • When the information processing relates to a business aim or goal and information processing is dependent on the content of data being processed, and that content does not relate to a low-level recording or capture of a physical phenomenon
          For example, processing of a digital sound recording to clean the recording of noise would be considered "technical"; processing row entries in a database of information technology assets to remove duplicates for licensing purposes would likely be considered "non-technical"
      • Presentation of Information:
        • When the innovation relates to user experience (UX) or user interface (UI) features,
          For example, machine learning algorithm that adaptively arranges icons on a smart phone according to use may receive objections on the grounds that features relate to mathematical methods (the algorithm) and presentation of information (the arrangement of icons on the graphical user interface)
        • As per Guideline G-II, 3.7.1, grant is unlikely if information is simply displayed to a user and any improvement occurs in the mind of the user [4]

    • Region - USA
      • Defining Inventorship:
        • Section 100(f) of the Patent Act, 35 U.S.C.A. defines "inventor", and also indicates that Congress intended statutory subject matter to "include anything under the sun that is made by man". [5] Accordingly, policy makers need to rethink current patent law with respect to AI systems and replace it with tools more applicable to the new era of advanced automated and autonomous AI systems, like AI algorithm-without any human intervention-develops a new drug, a method of recognizing diseases in medical images, or a new blade shape for a turbine? [6]
      • Limits of Disclosure:
        • Section 112 of the Patent Act, 35 U.S.C.A, requires inventor to disclose enough information needed to perform deep learning by person ordinary skilled in the art, which is challenging. For deep learning and artificial neural networks related inventions, claims directed to a broader scope of application may not be enabled by the rules developed, and disclosure of specific rules only, may not satisfy the disclosure obligations of Section 112 of the Patent Act, 35 U.S.C.A. Inventor have to decide what to protect under patent, i.e. processes by which the system is created, trained and validated or to protect final product deployed after these processes have run their course, as insufficient disclosure defining broader claim scope may introduce risk. [7]
      • Patent-Eligible Subject Matter:
        • Under Patent Act, 35 U.S.C.A. § 101, subject matter of a patent claim must be directed to a "process, machine, manufacture or composition of matter". Claims directed to nothing more than an abstract idea, such as a mathematical algorithm, or to natural phenomena or a law of nature are not eligible for patent protection. [8] As technology underlying AI is generally based on computer programming or hardware implementing mathematical models, deep learning algorithms or a neural network, AI applications are more likely to fall within this exception to patent-eligible subject matter, unless strong technical feature is associated with it
      • Determining Infringement by Deep learning algorithm:
        • Artificial intelligence indeed has the capacity to increase the pace and scope of innovation to meteoric or exponential levels. Yet, without explicit guidelines and/or sui-generis legislations to establish culpability for actions performed in conjunction with AI, AI based expert systems may continue to repeatedly infringe the existing patents and still roam scot-free. As per the current laws, a machine or algorithm can't be charged for committing patent infringement or other crimes and hauled into court
        • The current laws continue to hinge on the premise that 'culpability' has to be established with the party responsible for the use of the AI. Such responsibility is often shared among several entities in a complex and disputable manner. Owing to the aforesaid uncertainty in establishing the patent-rights ownership and infringement, the establishment of culpability remains arduous and prone to invite controversy
        • Accordingly, the need of the hour is a greater oversight and enactment of regulation as well as protection for the AI based technology. If legislators don't act quickly, unprotected and unregulated AI based expert systems could lead to an unforeseen catastrophic erosion of existing IP across the world
      • Disclosure to competitors:
        • All patent applications did not result in a "granted patent", but patent publication is must except provisions of national security and defense are applied. Publication of patent application lead to disclosure of inventive proprietary details to the public - including competitor - contained in the application. However, only claimed part is protected, other part of invention becomes free to use by others, irrespective of whether patent is granted or rejected
        • Patent process last for years, during which anybody can further develop or modify invention and get patent with expeditious mode [9]

  6. Few solutions to overcome the Hurdles:
    • Highlighting technical advantage
      • Chances of grant can be increased by highlighting practical application of the algorithm to a specific field or low-level technical area. Application drafted after in-depth discussion with inventors and framing patent application as a "technical" or engineering innovation, i.e. a technical solution to a technical problem, increase the chances of approval

    • Describing implementation of mathematical method:
      • Specification should frame in view of clearly defining how attributes of the physical world are represented within the computer. Use of pseudo-code is beneficial rather than mathematical formulae as they might seems deemed "technical" according to standard definition of the term, they are often not deemed "technical" according to definition applied by patent offices [4]

    • Functional claiming strategies:
      • 35 U.S.C. Section 112(f), allows invention to be claimed not merely by structural form, but by its function or purpose. Chances of patent acceptance can be increased by incorporation of functional representation of important structure or means, in sufficient detail that enable person skilled in the art to surmise the what structures the "means" or "step" language encompasses. Be cautious towards use of "Means plus function" as it renders patent prone to opposition by broadening the scope of Claims. Keep in mind that Functional claim strategy do not entitle inventor with all structure for performing the functions claimed, rather than entitles only for structure defined or disclosed on specification. In absence of sufficient disclosure, application can be invalid [10]

    • Trade Secret protection:
      • Protecting AI inventions as trade secrets can be viable option than patent as we all knew that, proprietary technology remains a trade secret as long as it is not publicly disclosed. Trade secret does not need any disclosure to public; it does not involve any application, examination or any consequent prosecution fees and can last longer than 20 year term, until not disclosed to the public. Of course, it having major disadvantage that you cannot take any action against competitor, who independently develop the technology or had reverse engineering it from products in the public domain. Still, trade secret protection are particularly well-suited for rapidly developing and changing artificial intelligence and related inventions [11]

    • Copyright protection:
      • Copyrights can be used as another form of protecting AI. Although it is associated with conflicts, due to lack of harmonization for IP law for software internationally, some offices are allowing the copyright to AI created work, which involves significant human contribution/intelligence, for example, implementation of creative judgment to output and depend upon work need to protect [12]

  7. References:
    1. WEF releases list of top 10 emerging technologies (June 28, 2017). [Source]
    2. A Short History Of Deep Learning -- Everyone Should Read (March 22, 2016). [Source]
    3. Deep Learning Market by Application (November, 2016). [Source]
    4. Can you protect Artificial Intelligence inventions at the European Patent Office? (2 August, 2017). [Source]
    5. Diamond v. Chakrabarty, 447 U.S. 303 (1980). [Source]
    6. When Artificial Intelligence Systems Produce Inventions: The 3A Era and an Alternative Model for Patent Law (11 May, 2017). [Source]
    7. The challenges of patenting artificial intelligence (November 27, 2017). [Source]
    8. Alice Corp. v. CLS Bank Int'l, 573 U.S. (2014). [Source]
    9. Regulatory Framework for Artificial-Intelligence - A Need of the Hour? (July 3, 2017). [Source]
    10. Federal Circuit Cases Clarify What Makes a Valid Software Patent (April 4, 2017). [Source]
    11. The top 4 advantages of trade secret protection (September 18, 2014). [Source]
    12. Can an AI Machine Hold Copyright Protection Over Its Work? [Source]