Software & Machine Learning Patents: Patentability Possibilities

Strona główna / Blog / Nauczanie maszynowe / Software & Machine Learning Patents: Patentability Possibilities

The skill of automating a task by creating rules for a computer to follow is known as software engineering. A further advancement made by machine learning is the automation of the rule-writing process. 

Spis treści

Software Engineering vs Machine Learning   

Machine learning and software engineering have a lot in common from the outset. Both have the same goal of solving problems, and both begin by familiarizing themselves with the problem domain by speaking with others and examining available information and tools. Execution choices make a difference. 

Software & Machine Learning Patents: Patentability Possibilities

Software developers employ their creative thinking to develop a solution and turn it into a precise programme that a computer can follow. Data scientists, or those who use machine learning systems, don’t attempt to create programmes on their own. Instead, they gather input data (such as video from a car’s dashboard and other sensor inputs) and desired goal values (the throttle level and the angle of the steering wheel). In order to locate a programme that computes an output for each input value, they then tell a computer to search for it (a programme that drives a car given the sensor inputs). 

Even more iterative and exploratory than the software engineering process is the development of a machine learning application. Machine learning is used to solve issues that are too challenging for humans to solve. A data scientist must have an experimental mindset and be willing to explore several ideas before choosing a successful one. 

Both types of workers spend a lot of time crouched over laptops, which gives the work environments a very similar appearance from the outside. Like traditional programmers, data scientists spend a lot of time creating code in Python or another general-purpose programming language. Writing scripts for merging, cleaning up, and displaying data, as well as connecting the machine learning subsystem with the rest of the application, take up the majority of time in a machine learning project. The toolkits undoubtedly differ from one another. While typical programmers are well-versed in REST APIs and web frameworks, data scientists have extensive knowledge of linear regression and other statistical procedures. 

US Patent Law  

The following two criteria must be met for software patent applications to be eligible for patent protection under the United States’ existing system of patentability: 

(1) The “abstract idea” requirement  

If a piece of software “improves computer functionality,” it qualifies as patentable. 

– enabling computations that were previously impossible for computer equipment to accomplish,   accelerating operations that were already possible,  

– lowering the number of computing resources needed to complete a task. 

(2) The “transformation” requirement 

Software may nonetheless qualify for patent protection even if it doesn’t generally “increase computer operation” if one or more of the following conditions are satisfied: 

  • the issue is not one that is “necessarily based in computer technology”;  
  • the issue is resolved by using “unconventional” components or by arranging conventional components in a “unconventional” manner;  
  • the patent claims do not cover all possible implementations of the concept. 

Case Law: Alicja przeciwko CLS Bank ? 

Wyrok Sądu Najwyższego w Alicja przeciwko CLS Bank, 134 S. Ct. 2347 from 2014 has caused confusion regarding the applicability of software patents (2014). 

The Supreme Court defined a two-step process in this case for evaluating whether a certain piece of software is patentable. First, an “abstract notion” cannot be the subject of a computer-related patent application. But if it is, the patent application must make certain claims that “transform” the alleged invention into one that qualifies for patent protection. 

Can Software Be Patented In The USA?  

Can you effectively patent a piece of software that enhances the memory configuration of a database system by using a self-referential lookup table? How about a piece of software that permits cellular network archiving of digital images? 

The response is “it depends” (as you could anticipate in practically all legal settings). A usually clear-cut analysis has become significantly more ambiguous as a result of recent Supreme Court judgements and subsequent lower court opinions. In order to answer the questions above, we currently require a great deal more information, including details on the underlying technology and the structure of the patent application. 

Inventions based on software are still eligible for patent protection in the US. However, software patent applications must adhere to specific technical specifications and be carefully drafted in order to be eligible for patent protection. 

From a technical perspective, your software might be patentable (1) if it enhances computer functionality in some way (for example, it makes computations possible that weren’t possible before, speeds up procedures, or uses fewer resources), or (2) if it finds an unconventional solution to a computing problem. 

Additionally, only if it is written with a precise focus on the technical merits of your specific software solution may your software be patent eligible. By outlining the technological difficulties encountered in your area of invention and in great detail outlining and claiming the remedies you have developed to meet those difficulties, you can increase your chances of receiving a patent. You’ll have a very tough time becoming patentable if you try to list every possible solution to address a certain problem or if you concentrate on the advantages that your programme allows a user to experience. 

Software Patents: Claims and Specification 

The patentability of software inventions frequently depends on how the patent and the patent claims are worded, much to the dismay of developers and founders. By outlining the technical difficulties in your area of innovation and in detail outlining the engineering solutions you have developed to meet those difficulties, you can increase your chances of receiving a patent. Additionally, you must be very careful when claiming your idea. You shouldn’t profess to know how to treat every type of pain. Instead, you should precisely focus your claims to just address the identified pain area. 

The following are the five fundamental procedures for formulating claims and specifications in software patents: 
  1. Approach the invention as a concept for a problem-solution.
  2. Create a clean, labelled flow chart diagram that includes all the features and functions that the innovation has disclosed.
  3. Create system architecture or block diagram that shows the network-connected connections between the basic hardware components.
  4. Enable correct synchronization of the flow charts and block diagrams.
  5. Prepare patent claims (systems or devices) that incorporate all block diagrams and method claim components.

Technically speaking, if you can explain how your innovation enhances computer functioning and how it differs from existing solutions in your specific field of invention, the likelihood that you will be granted a software patent will increase. 

Can Machine Learning (ML) Be Patented in ton USA? 

The machine learning (ML) industry has taken up, assisting businesses with anything from better breast cancer detection to raising ad conversion rates. The market for machine learning is expected to grow to $8.8 billion by 2022. 

Companies are interested to understand the criteria and restrictions of AI and ML-related software patents, from huge tech enterprises to lean start-ups. However, there is frequently misunderstanding about what is patentable and it has become a contentious issue in recent years. 

As a broad perspective, the majority of current technology functions via inputs and outputs. In this instance, a human provides the input data, and a machine or piece of software computes the output. Keep in mind that a person must still be involved in the process in this case. In contrast, machine learning is exactly what its name implies: future computations and behaviours that are taught by a computer independently of human intervention. In this instance, the input and output are provided by the machine. 

This brings up the crucial issue of patentability for machine learning algorithms. 

In actuality, it depends on what an algorithm is to you. An algorithm cannot be directly patented under U.S. patent law. You can, however, patent the sequence of operations in your method. This is due to the fact that an algorithm is viewed as a set of mathematical operations and steps under US patent law. 

Examples of Machine Learning-Related Patents 

Google, Samsung, and Amazon are the key participants in this market. Here are three instances of machine learning-related patents from these companies: 

  1. Samsung’s drone that can be operated with hand gestures and face recognition: The most recent drone patent from Samsung can recognize a person’s face, pupils, and hand movements. The camera is described in the patent as a system that sends data to the primary control unit. As a result, it provides the inputs. 
  2. Amazon files a patent that can record and save conversational details: Smart speakers are constantly observant of their surroundings. With Amazon’s most recent patent, your personal hobbies as well as a trigger word may be used to activate the Alexa smart speaker. 
    The patent claims that Alexa will register words that have deep meanings. For instance, “I enjoy Italian cuisine.” The smart speaker will assess this information after hearing a phrase including the word “love” and utilize it to tailor adverts. You’ll probably soon start to see advertisements for Italian food. 
    This method can also be applied to block keywords. For instance, if you declare, “I detest sushi,” this will be noted as a distaste of yours, enabling Amazon advertisers to avoid promoting to you. 
  3. Google wants to provide you with quick, precise responses: Google was pretty straightforward in its early years. It would display web pages with those terms if you typed in specific keywords. 

Their goal has always been to respond to search requests with better results. They soon discovered, though, that their previous approach had been more focused on web pages than it had been on providing insightful solutions. 

Machine Learning: Claims and Specification  

The claim’s writing must have a goal of producing successful outcomes. The following is a list of recommendations for writing strong claim language and requirements for AI/ML patents: 

  1. Pay attention to the claim’s ML model’s structure 
  2. Claim the training process.  
  3. Determine if the invention is in the training phase, the execution phase, or both. 
  4. Highlight the preparation of the input data 
  5. Address the input to model mapping. 
  6. Claim the post-processing and explain the results of the data. 
  7. Create distinct claim sets for the execution phase and the training phase. 
  8. Disregard the assertion that a model should be routinely applied to current data. 


The biggest American and Japanese IT businesses are responsible for the majority of AI and machine learning patents, which is not surprising. Chinese businesses have expanded their patent portfolios in recent years. However, their caliber is the issue. According to the World Intellectual Property Organization, the number of patents for artificial intelligence has significantly increased during the previous five years (WIPO) 

 WIPO reports that from 2013 to 2017, the number of patent applications relating to AI increased dramatically, at 193%. The increase in patent applications, according to the WIPO director general, “means we may expect a very considerable number of innovative AI-based goods, applications, and approaches that will affect our daily lives and also impact future human interaction with the robots we developed.” WIPO discovered that 434 companies have been bought since 1998 in the AI industry, with more than half of those transactions occurring after 2016. 

 AI and machine learning are currently the driving forces propelling innovation in all sectors of the economy, and they will continue to provide enormous value to businesses that are savvy enough to safeguard their intellectual property. 


Nieustannie dostrzegamy wartość nowych technologii wdrażanych przez naszą wykwalifikowaną załogę kierowniczą z doświadczeniem przypominającym naszych profesjonalistów. Podobnie jak profesjonaliści z zakresu własności intelektualnej, których wspieramy, nasz głód rozwoju nigdy się nie kończy. IMPROWIZUJEMY, ADAPTUJEMY i WDRAŻAMY w sposób strategiczny.

Konsultanci TT oferuje szereg wydajnych, wysokiej jakości rozwiązań do zarządzania własnością intelektualną, począwszy od

i wiele więcej. Dostarczamy rozwiązania pod klucz zarówno kancelariom prawnym, jak i korporacjom wielu branż.

Skontaktuj się z nami
Udostępnij artykuł



Poproś o oddzwonienie!

Dziękujemy za zainteresowanie konsultantami TT. Wypełnij formularz, a my wkrótce się z Tobą skontaktujemy



    Twojego Pomysły

    Podnieś swoją wiedzę patentową
    Ekskluzywne spostrzeżenia czekają na Ciebie w naszym biuletynie

      Poproś o oddzwonienie!

      Dziękujemy za zainteresowanie konsultantami TT. Wypełnij formularz, a my wkrótce się z Tobą skontaktujemy