Challenges d'analyse de données : une formation par la pratique transversale et multidisciplinaire en science des données

Jean-Baptiste DurandHAL 2017

TL;DR

This paper introduces a novel pedagogical approach at Université Grenoble Alpes (UGA) centered around data analysis challenges. The goal is to bridge the gap between theoretical knowledge and practical application in data science, fostering interdisciplinary collaboration and preparing students for real-world challenges in academia and industry.

What Problem Does This Paper Address?

The paper addresses the growing demand for data science skills across various academic disciplines and the subsequent need for effective training. Traditional university curricula often struggle to provide students with the hands-on experience and interdisciplinary exposure necessary to tackle complex, real-world data analysis problems. This gap is particularly evident in bridging the understanding between mathematical/algorithmic methods and their semantic interpretation in specific application contexts. The authors aim to create a more engaging and practical learning environment that mirrors industry and research settings, making data science more accessible and relevant to a broader student population.

Methods

The core of the proposed solution lies in organizing "data challenges." These are essentially competitions where students, working in multidisciplinary groups, apply data mining and machine learning techniques to real-world, often complex datasets. The methodology emphasizes a "learning by doing" approach, integrating theoretical learning with practical problem-solving. Key aspects include:

  • Multidisciplinary Teams: Students from diverse backgrounds (statistics, computer science, physics, biology, etc.) collaborate, bringing complementary skills.
  • Real-World Data: Challenges are based on authentic data from various domains, exposing students to the messiness and complexity of actual data.
  • Professional Skills Development: Beyond technical skills, the challenges aim to cultivate crucial professional competencies like communication, teamwork, and problem comprehension in unfamiliar domains.
  • Platform and Infrastructure: The project involves developing a dedicated platform for hosting challenges and a multimodal classroom to facilitate collaborative work and learning.

Datasets

The paper does not specify particular datasets used for training or evaluation within the described challenges. It mentions that challenges will be based on "real-world, often complex datasets" from various domains such as health, biology, meteorology, imaging, and social sciences. The authors also plan to build a database of challenges, including datasets, their characteristics, and associated results, which will be made accessible to external users.

Key Insights and Ideas

  • Bridging Theory and Practice: Data challenges offer a powerful mechanism to bridge the gap between theoretical data science concepts and their practical application, making learning more engaging and effective.
  • Interdisciplinary Synergy: Bringing together students from different fields is crucial for tackling complex data problems, as it fosters diverse perspectives and complementary skill sets.
  • Preparing for the Job Market: The challenges are designed to equip students with the skills and experiences highly sought after by employers in data-intensive industries.
  • Fostering Autonomy and Self-Evaluation: Beyond competitive rankings, the project aims to encourage students to develop self-assessment skills regarding their data analysis process.
  • Leveraging Existing Infrastructure: The project builds upon and integrates with existing university initiatives and platforms, promoting resource sharing and collaboration.

Key Results

The paper describes a pedagogical project that is set to be implemented in the 2017-2018 academic year. Therefore, concrete quantitative results from the challenges themselves are not yet available. The expected outcomes are primarily qualitative and related to the successful implementation and adoption of the pedagogical approach. These include:

  • Increased student engagement and practical skill development in data analysis.
  • Enhanced interdisciplinary collaboration among students and faculty.
  • Improved preparation of students for careers in data science and related fields.
  • The establishment of a sustainable platform and infrastructure for data challenges at UGA.

Limitations and Future Work

Limitations:

  • Early Stage of Implementation: The project is presented as being in its initial phase, with implementation planned for the upcoming academic year. Therefore, empirical evidence of its effectiveness is not yet available.
  • Marginal Use in Grenoble: The paper notes that the use of data challenges, while popular internationally, is still relatively marginal at the Bachelor's and Master's levels within the Grenoble community.

Future Work:

  • Continuous Improvement: The project team will regularly analyze feedback from various student groups to enrich and improve the diversity of challenges in terms of applications and methodologies.
  • Platform Enhancement: Ongoing efforts will focus on improving the platform's quality and quantity of features, including forums, code sharing, and cluster computing resources.
  • Externalization: The long-term vision includes opening the platform to external actors, increasing its visibility and impact.
  • Essaimage (Spreading): The pedagogical approach will be disseminated to other programs and institutions, potentially through multidisciplinary projects involving socio-economic partners.
  • Integration with Teacher Training: A specific project involves ESPE (École supérieure du professorat et de l'éducation) students to develop pedagogical experiments and analyze data, with support from statistics students.

Summary updated

Abstract

International audience
BibTeX
@inproceedings{durand2017challenges,
  author = {Jean-Baptiste Durand},
  title = {{Challenges d'analyse de données : une formation par la pratique transversale et multidisciplinaire en science des données}},
  booktitle = {HAL},
  year = {2017},
}
Figure