Our next AI Developers\u2019 Work Group will take place on 28 October 2024. \ud83c\udf89<\/p>\n
Luca Becker from the Institute of Communication Acoustics (IKA) at Ruhr University Bochum will present three key areas of his research in a talk titled \u2018Advances in Privacy-Preserving AI: Feature Extraction, Federated Learning in Acoustics, and Deep Learning in Medical Applications\u2019:<\/p>\n
Privacy-Preserving Feature Extraction for Deep Learning \ud83d\udd10<\/strong><\/strong><\/p>\n In today\u2019s AI landscape, privacy protection is of critical importance, particularly as machine learning models are capable of processing large volumes of sensitive data. Luca Becker presents the IKA\u2019s approaches to privacy-friendly feature extraction, which minimise the risk of data loss whilst ensuring high performance. This includes the development of innovative architectures that prioritise data protection whilst maintaining high performance. <\/p>\n Federated Learning and Heterogeneous Data in Challenging Acoustic Scenarios \ud83d\udd0a<\/strong><\/strong><\/p>\n Federated learning offers a decentralised approach to training machine learning models, but it also presents significant challenges, particularly when dealing with heterogeneous data from diverse and challenging acoustic environments. In his talk, Luca Becker will discuss the techniques developed by the IKA to address these issues. <\/p>\n Conditional Label Smoothing for Data Augmentation in Medical Text Classification <\/strong>\ud83c\udfe5<\/strong><\/p>\n As part of the HUMAINE project, a new method for data augmentation using conditional label smoothing is being introduced in collaboration with BO-I-T, specifically for the classification of medical texts. This approach improves model generalisation, particularly in the context of limited datasets, whilst also addressing the sensitivity of medical data. <\/p>\n Look forward to fascinating insights into the IKA\u2019s interdisciplinary approaches, which combine theoretical advances and practical applications with a focus on human-centred design to push the boundaries of AI.<\/p>\n <\/p>\n