on Thursday 23 February 2023, from 14.00 to 15.00 hrs
The EPO's first ever public CodeFest tackles one of today's key sustainability challenges: ridding the planet of plastic waste. Global success depends on access to the right kind of know-how and that's what this major new competition is all about: making the know-how contained in patents concerning green plastics more readily available to innovators everywhere.
The ultimate aim is to inspire future innovation that supports healthy ecosystems and drives the circular economy for plastics. In alignment with the United Nations Sustainable Development Goals (SDGs), we are taking our commitment to sustainability and monitoring green technologies to the next level with initiatives such as the CodeFest on Green Plastics, which is of special relevance to SDG 12.
The EPO's unparalleled data resources are relied upon by a range of governmental and international organisations, such as IP Offices, the OECD, International Energy Agency (IEA), International Renewable Energy Association (IRENA) and European Commission, as well as researchers and patent information specialists. The CodeFest on Green Plastics is a majoThr opportunity to help improve access to these data resources and contribute to the development of an automated classification scheme for sustainable technologies. The competition will also help prepare the way towards creating a platform for exchange on AI-based approaches to organising the EPO's wealth of patent information.
Some of the groundwork has already been done in a recent EPO study on patents for tomorrow's plastics. This shows how patent data reflects the rapid rise in innovation in plastics that are easier to recycle, as well as in recycling technologies. The study also suggests that there is great potential to turn pioneering research in the field into inventions that can then be brought to market. "Patents for tomorrow's plastics" provides an excellent introduction to the topic and an important starting point for anyone wishing to take on the current code challenge, which is:
To develop creative and reliable artificial intelligence (AI) models for automating the identification of patents related to green plastics.
We are calling on talented minds aged 18 and over to register for the event by 13 November 2022, either in an individual capacity or as a team of up to five people. Anyone resident in one of the EPO member states is eligible to enter. Teams of EPO colleagues, external entrants or a mixture of both are all welcome!
In order to participate you must meet all eligibility criteria included in the rules of the competition.
Each of the prizes will be accompanied by a trophy.
Submissions will be evaluated by senior specialists from across the EPO working in sustainability, IT, data science and AI, as well as patent information and analysis.
Thursday 15 September 2022, 12.00 hrs CEST
Entrants must register for the challenge until 13 November 2022, 23.59 hrs CET. Registration includes entrants submitting an outline of their approach for solving the code challenge. Proposals must include an explanation of the intended approach and indicate the intended data selection, the possible model implementation and model evaluation.
14 November 2022
The jury evaluates the submissions.
15 November 2022, 12.00 hrs CET
The code challenge is on!
The jury communicates the outcome of the previous evaluation to participants. All successful candidates will be given access to our EP full-text dataset and to Open Patent Services (OPS). The code challenge runs for just over eight weeks and solutions must be submitted until 15 January 2023 at 23.59 hrs CET.
The jury evaluates the submitted solutions and selects the finalists of the code challenge.
Announcement of the finalists!
Finalists are invited to the EPO CodeFest on Green Plastics ceremony.
EPO CodeFest on Green Plastics ceremony
During the online ceremony, the winners are announced and prizes awarded. Details of the full programme will follow soon.
Sunday 13 November 2022.
Over 60 individuals took up the code challenge to develop an AI model that improves access to patent information on green plastics. After fierce competition, six finalists convinced the jury with their creative code, demonstrating how their models could help innovators learn from patent know-how and rid the planet of plastic waste.
Follow the online award ceremony on Thursday 23 February 2023, from 14.00 to 15.00 hrs, where all six finalists will present their ingenious solutions before we reveal the three winners.
Don’t miss out!
The registration is handled via XING SE but there is no need to have a XING profile.
CodeFest was open to residents of EPO member states, including EPO staff. The challenge drew a broad range of individuals and teams from all over Europe, with each developing creative solutions for identifying green plastics. These are the finalists.
This team's model uses state-of-the-art AI pipelines and large language models from OpenAI for zero-shot, few-shot and other approaches to arrive at a custom MLP neural network for binary and multi-label classification.
As there is currently no classification scheme or labelled data available in this field, Green Hands proposed a new classification scheme, and developed a strategy to automatically assign labels to patents in order to create a labelled training dataset.
This team created a deep learning architecture to classify patent documents by fusing features from figures and text, thus exploiting the multimodal nature of patents.
Nikolaos developed a machine learning model that incorporates both semantic and lexical features, and that was trained on a dataset of patents and scientific publications.
Nikolaos Gialitsis has a proven track record in data science, software, and machine learning engineering roles in large-scale EU projects as well as in commercial projects. He has published research findings at top conferences, showcasing his expertise in natural language processing and machine learning. Nikolaos has also contributed to the full-stack development of open-source tools and virtual marketplaces, web frameworks and and theoretical knowledge of object-oriented principles and design patterns.. In his free time, he organises volunteering activities related to interdisciplinarity and science communication.
The team converted the problem into a sequence-to-sequence challenge, asking the user to define green plastics and then testing any patent claim against that hypothesis.
Using a gradient boosting machine, Thomas focussed on high sample efficiency, unbiased validation metrics and maximising specificity.
“Green plastics” is a generic term and does not come with an exact scientific definition.
Plastics referred to as “green” are generally plastics with a reduced or minimised environmental impact when they are produced, used, recycled, upcycled, disposed of, when they decompose, etc. They exhibit one or more of the following properties: source renewability; biodegradability/compostability after end of life; possible to process in an environmentally friendly way.
Equally, a process that makes plastics greener would also fall under “green plastics”, for example a form of chemical recycling that makes PET greener by enabling it to be recycled more efficiently and minimising waste.
As “green plastic” is not a specific material but a way to look at the whole lifecycle of a material, we ask participants to include an explanation of their intended approach to solving the code challenge (see rules of competition “Proposals for solving the challenge” and “Submissions of solutions to the code challenge”). This explanation should clearly indicate what aspects of green plastics are included in the proposed solution to the code challenge. For further background information on the kinds of plastics, materials and processes that may be considered for inclusion, please consult the publication Patents for tomorrow’s plastics.
The expected output is an AI model capable of identifying patent and non-patent documents related to “green plastics”. Any published information, including patents, can be used to train the AI. However, you could come up with a more creative idea on how to train the AI. We welcome your creativity.
Participants are free to draw upon all forms of established classification, including IPC/CPC/Y02, etc., that are considered useful to developing creative and reliable AI models for automating the identification of patent and non-patent literature related to “green plastics”.
We state in the “Rules of Competition”, in the section “Intellectual property”, that participants own the rights to the Submission they create and, therefore, the rights to publish and upload the software. However, as also stated in the same section, if you use data and/or APIs or other copyrightable material owned by third parties, you must abide by the terms and restrictions they might have specified.
If you do publish your Submission, we would appreciate it if you acknowledge that the code and any results were created while participating in the EPO’s CodeFest on Green Plastics.
There is a priori no limitation on the languages to be used in the CodeFest: all patent documents can be considered, regardless of language.
However, we request that the results are presented using any of the EPO official languages: English, French, German.
Registration for the CodeFest can only be completed once we receive your proposal. If you have started the registration process, remember to submit your proposal by 13 November.
There are no restrictions. Feel free to use the format you consider best suited to describing the concept, process and data object of your proposal.
The successful candidates selected from all those who have completed the registration process will be given free access to two EPO products: our EP full-text dataset and Open Patent Services (OPS).
Access to the EP full-text dataset provides the successful candidates with the full text of the entire collection of EP publications (A and B). Further information on the EP full-text dataset.
OPS data is extracted from the EPO's bibliographic, worldwide legal event, full-text and image databases and is therefore from the same sources as used for Espacenet and European Patent Register data. For further information on OPS.