Update: Deadline extended to July 15th, 2019.
ELITE-S provides up to 16 postdoctoral fellowships to work for 2 years in the ADAPT Research Centre at Trinity College Dublin, University College Dublin, or other Irish universities as well as with an industry partner. ELITE-S is a EU Marie Skłodowska-Curie COFUND Action.
You can apply with any research proposal in the field of ICT standardisation that fits into one of the following four themes.
- Regulations and Compliance
- Content Analytics
- Content Preservation Data and Process Management
If you have a proposal relating to recommender systems, (automated) machine learning, or APIs, we would gladly act as a supervisor for you. In that case, please contact us.
If you do not have your own idea, you may pick one of the following topics. Contact us to discuss any of the ideas in more detail.
Recommender-Systems Description Language (RSDL)
Publications in the field of recommender-systems vary a lot in terms of the information they contain about data, algorithms, and evaluations. Consequently, it is difficult to replicate research results, or re-implement recommender-systems algorithms.
A recommender-systems description language could help in making research more reproducible and publications more structured.
A formal description language for recommender systems would also make publications more machine-readable, and ease the process of finding recommender-systems publications that e.g. used a certain evaluation metric for a certain algorithm in a certain domain. Similarly, if algorithms were described in a standardised language, a search engine for recommender-system algorithms (or source code) could be implemented.
Also, recommender-system datasets such as MovieLens or RARD II all have their own definitions, variables etc. Consequently, when using a dataset with a recommender-system library (e.g. LensKit, Surprise, Mahout, …), a lot of manual work is required. A standardized language, data format, or algorithm description, could help to use datasets easier in recommendation frameworks.
If you are interested in developing a Recommender-System Description Language, you may develop and test this language in our recommender-system as-a-service that serves multiple partners with various data formats and several algorithms.
A Description Language and Standard for Meta-Learning / Automated Machine Learning
Meta-Learning and Automated Machine Learning (AutoML) — including Neural Network Architecture Search, Hyper-Parameter Optimization, Automated Algorithm Configuration, and Learning to Learn — is a strongly growing field of research, with very high relevance to industry. There have been massive advances in recent years, and many tools were published to ease the process of automating machine learning. The list of tools is long, and includes AutoWeka, Google’s AutoML and Auto sklearn. However, these tools are not interoperable and, similar to the problems in the recommender-systems field (see above), publications etc. would benefit from a standardized description language.
Open Recommender Systems API
Many disciplines have open standards for APIs. Prominent examples include the OpenAPI Initiative and OpenSearch. The recommender-systems community could benefit from a similar initiative, or applying and adjusting existing solutions to the special features of recommender systems. First attempts have been made (Garcia & Bellogin, 2018) but they were not comprehensive. You could extend the existing work, and potentially include and test your new standard in our recommender-system APIs Darwin & Goliath, Mr. DLib or the APIs of some recommendation frameworks and industry partners.
Garcia & Bellogin, Iván Garcia and Alejandro Bellogin, “Towards an open, collaborative REST API for recommender systems,” in Proceedings of the 12th ACM Conference on Recommender Systems (ACM, 2018), 504–505.