UNIGE Data Science Day
Call for communications
Deadline: 6th of July 2022
The 2022 UNIGE Data Science Day will focus on the theme "Promises of Artificial Intelligence : an Interdisciplinary Revolution", and will take place at Uni Mail on 15 September 2022 (8h30 - 18h00).
After a morning plenary session, numerous panels/round tables will be held in the afternoon, organised by researchers from various disciplines.
Five panels/round tables are still open for new speakers!
Do not hesitate to send a communication proposal (title and abstract) to their organisers before the 6th of July 2022.
Digital Methods and the Liberation of Methods in the Humanities and Social Sciences
Béatrice Joyeux-Prunel, Tommaso Venturini et Simon Gabay
For too long, the humanities and social sciences have been bogged down in binary oppositions such as close VS distant reading or qualitative VS quantitative research, which, despite their historical raison d'être, have ended up anesthetizing our methodological imagination. Time has come to be even more creative and to seize the renewal opportunities offered by the advent of digital technologies. The rise of digital records and computation tools has brought about a real revolution in the humanities and social sciences. But not the one we too often imagine.
This event is dedicated to the methodological creativity spurred by digital methods. It offers a safe space to discuss playfully, but also very seriously, about new and different research protocols to study human and social phenomena. If you have a research method that feels unorthodox, a quirky dataset, and offbeat technique of analysis or visualization, this event is for you. Come with 2 slides and play with us!
The rise of generative modelling and anomaly detection in Science
Tobias Golling & Steven Schramm
Automated prediction is one of the major drivers of modern machine learning, be it weather prediction, protein folding or the prediction of readout signals in a measurement device. Generative modelling techniques have become the gold standard for a wide range of applications, with the most popular models including variational auto encoders (VAEs), generative adversarial networks (GANs), normalising flows or diffusion models. The automated detection of anomalies builds on this precise modelling of high-dimensional data distributions through generative modelling. Applications range from the search for the largest objects in space to the tiniest particles in the collisions of the Large Hadron Collider at CERN. In this panel, we intend to focus on early career scientists, inviting them to talk about their challenges and the proposed solutions, as well as how the approaches could be generalised.
AI for Good
This panel/round table discusses the ways in which AI has been / can be used for pro-social purposes. AI has demonstrated its unique potential for aiding businesses, research institutions, and other organizations. A growing movement, lead mainly by the AI for Good initiative at ITU, is pushing for the deployment of AI also to promote social good, in particular aiding initiatives aimed at achieving the SDGs. The goal of this panel is to discuss where we stand with respect to the adoption of AI for promoting social good, with a particular focus on discussing what are the most promising ways in which AI can be used by organizations working to promote public good, such as philanthropic organizations.
Data Science for the animal’s welfare and experimentation
When dealing with animal experimentation, one of the first aims is to work with the right number of animals, which for ethical reasons should be the smallest possible while keeping the statistical significance and power:
- Q1) How can the data science help us REDUCE the number of animals? How to calculate the number of animals needed for an experiment when not all information is available, for example in case of exploratory study? How to reduce not only the number of animals used in experiments, but also the number of animals crossed and born in animal facilities?
Pre-clinical studies with animals are sometime seen as the cause of the failure of clinical trials and part of the general “reproducibility crisis” in research. Exclusion of a sex in experiments with animals may give false hint when translating results to men and women; however, exclusion may be justified in some experimental models.
- Q2) How can the data science lead to REPRODUCIBLE RESULTS of existing studies with a more balanced layout? How data science can help researchers including as often as possible both sexes in order to obtain reproducible and translatable results, while keeping low the number of animals used low?
Last, but not least, a primary instrument when conducting research with animals is the monitoring of animal welfare by completion of dedicated clinical score sheets. A clinical score sheet or “welfare assessment protocol” was established by Morton and Griffiths in 1985 (Morton and Griffiths, 1985) as a tool to grade the suffering of animals and to determine humane endpoints. Retrospective analysis of score sheets is not commonly used and practiced in the animal experimentation domain.
- Q3) How can data science be used to REFINE further animal experimentation by using information from the “score sheet”?
The three questions above will be addressed and discussed by scientists and experimenters during the above-mentioned round table.
QoL Vadis? An Interdisciplinary Panel
The current behaviour and Quality of Life (QoL) assessment methods rely on lengthly self-assessments over a given period (e.g., a week), conducted via Patient-Reported Outcomes (PROs). These assessments are typically sparse and suffer from biases affecting reporting, ceiling and floor effects, and a lack of sensitivity to change at their scale’s extremes.
Conversely, personal smartphones and wearables are becoming increasingly accurate in measuring long-term behavioural Technology-Reported Outcomes (TechROs). The extent to which TechROs provide complementary information that is useful for domain experts and meaningful for individuals is unknown. To this end, this panel will critically discuss the scientific progress needed in their respective domains that may enable and benefit from the rise of these technology-enabled, data-driven methods and tools for objective, quantitative, longitudinal assessment of individual behaviours and QoL. The panel will also highlight the required factors for adopting and scaling the TechROs and discuss how these technologies can be leveraged for behaviour change, disease prevention, health management and long-term QoL enhancement in populations at large.