← Back to the schedule

Beyond accuracy and time: Preservation of Privacy, Explainability or Reduced-Precision approaches in Machine Learning

Calendar icon

Wednesday 14th

Time icon

11:00 | 11:40

Location icon

Theatre 20



Machine learning is one of the areas of Artificial Intelligence that has an increasing important role in Big data systems, as it is able to efficiently discover new information and knowledge from data. Most approaches of Machine Learning algorithms for Big Data era exploit parallel or distributed strategies for speeding up learning while maintaining or improving accuracy. Although both are undoubtedly of utmost importance, trusted Artificial Intelligence asks for more challenging aspects to be approached, such as privacy-preservation (needed for example when several organizations do not want to share their data, or have different privacy policies, while at the same time a joint data processing is a must for learn or identify new patterns), explainability (for example, not only homogeneity is needed for customer segmentation, but besides the number of variables included in each cluster could be a clue for a human understanding of the results and corresponding action programming). Finally, another side of the Big Data scenario is the ratio between data volume and device size. With the growing importance of portable embedded systems, there is also an increased interest in implementing machine-learning algorithms with a limited number of bits, which allow obtaining information from the data that can be stored in our smart phones, health wearables, fitness trackers, domestic energy recorders, etc.