Scientific achievements

1. Two conceptual models for data and metadata respectively, both general, able to represent various multi-modal healthcare data, allowing the usage of ontologies, and extensible to various healthcare scenarios
2. An ETL algorithm to fully automatically convert existing data and metadata to instances of our models, resulting in an interoperable database
3. A metadata and data catalogue for exploration and ease federated learning algorithm design.

Practical outcomes

1. Each hospital has an interoperable database, part of the global BETTER network
2. Federated learning algorithms are co-created and implemented by researchers and practitioners
3. The catalogue implementation for the 3 use-cases (genetic rare diseases)

Associated publications

TODO

Artifacts

• I-ETL (Python application, 6k LOC, 1.3 year)

Scientific achievements

1. A user-oriented methodology for querying a relational data lake by selecting data pieces of interest instead of formulating SQL queries

Practical outcomes

1. A software suite for the data exploration of the underlying data lake

Associated publications

TODO

Artifacts

• ConnectionStudio (Java and Javascript application, 25k LOC, 1 year)

Scientific achievements

1. Efficiently enumerate all paths connecting named entities appearing in multi-modal and heterogeneous data by relying on an intermediate summary graph and a view-based algorithm
2. The ranking of paths based on their interestingness, a quantitative measure based on entity confidence and information dilution

Practical outcomes

1. A novel ChatGPT-based module for entity extraction
2. The efficient evaluation of paths by using a dedicated multi-query optimization algorithm

Associated publications

TODO

Artifacts

• PathWays (Java application, 4k LOC, 1.9 year)

Scientific achievements

1. A novel model-agnostic data summarization method relying on data kinds instead of specific model features
2. The selection of most representative entities, their attributes and relationships from the data summary by relying on a graph representation and PageRank node scores

Practical outcomes

1. A end-to-end pipeline to summarize any (set of) heterogeneous datasets
2. An interface to visualize summaries as Entity-Relationship diagrams

Associated publications

TODO

Artifacts

• Abstra (Java application, 10k LOC, 3.2 years)

Scientific achievements

1. An algorithm to automatically select the most useful subset of indicators among the hundreds collected ones
2. An algorithm to automatically predict in a supervised manner the environment of any French neighborhood based on several subsets of indicators

Practical outcomes

1. Collect and process cartographic and urban data from public data such as INSEE
2. An interface for predicting the environment of any French neighborhood
3. A general interface for testing, evaluating and comparing AI algorithms

Associated publications

TODO

Artifacts

• Predihood (Python and JavaScript application, 3k LOC, < 1 year)

Scientific achievements

1. A customizable similarity formula for estimating the matching similarity between geographic entities
2. The computation of the estimated quality of the matching without ground truth

Practical outcomes

1. A map interface for find, matching and merging similar geographic entities

Associated publications

TODO

Artifacts

• GeoAlign (PHP and JavaScript application, 4k LOC, < 1 year)

Scientific achievements

Practical outcomes

Associated publications

TODO

Artifacts

• VizLIRIS (Python and JavaScript application, 1k LOC, < 1 year)