User feedback in machine learning

USER FEEDBACK IN MACHINE LEARNING / USER-IN-THE-LOOP IN MACHINE LEARNING: Feasible or not ?

CNES Experts Communities (COMET) OPS (Operations and Exploitations of Space Missions), SIL (Computer System Architecture and Software Engineering) and ORB (Orbital mechanics) organize a workshop on Machine Learning / deep learning techniques and more specifically around the problem of user feedback in these algorithms.

By user feedback it is meant the ability for the user, based on his / her technical knowledge, to criticize / invalidate the results of the algorithms (misclassification, false alarm, etc.) and that this information is taken into account in upstream learning models. In the end, the goal is for the algorithms to gradually gain in performance by integrating this information, while offering the user better control over them.

CNES is working on these questions from various angles (classification of soils in imagery, detection of anomalies in satellite telemetry, etc.) and wishes through this COMET workshop to allow sharing on this theme between actors in the field.

The event was a great success and showed the tranversality of such issue through the wide variety of use-cases presented during the day (from and outside the space sector). The solutions presented to implement user feedback ranged from computer engineering (GUI, user-tagging, assistant,etc.), mathematics (active learning solutions, TéSA Online-Addict algorithm, etc.) to how to adapt your company organisation to these issues (Learning Services for Space Operations from ADS).

Many recommendations were taken from this event to be applied into CNES applications, as well as partnerships, and we hope that other companies will do the same.

If you want to know more about this past event, please contact : Pierre-Baptiste.Lambert@cnes.fr

The program of the event is avalaible below.