The aim of WORK PACKAGE 2 is to develop innovative explainable computational intelligence algorithms and integrate developed technologies including:
(1) Development of the ExplainMe software module for data collection and processing (M12-24)
Algorithms for collecting and aggregating voice data will be developed using libraries such as opensmile , bipolar and others. As part of this task, the processing and modeling algorithms will be significantly expanded data in the bipolar package [https://github.com/ITPsychiatry/bipolar], which allows you to include data uncertainty from psychiatric surveys for learning and reasoning. ExplainMe significantly expands the scope of bipolar because it provides explanations in natural language and creates a new, broad class of systems for explaining how speaking in the context of health information.
(2) Explainable classification methods for data streams (M12-36, collaboration with Dr. G. Casalino from the University of Bari)
The task will develop explainability-by-design classification methods for explaining health data in in the context of evolving data streams representing acoustic features extracted from the voice. They will be implemented works including (i) unsupervised learning inspired by fuzzy cluster analysis (ii) explainable hidden Markov models. The task significantly extends the previous work of the main contractor, e.g. [ K. Kaczmarek-Majer, et al. PLENARY: Explaining black-box models in natural language through fuzzy linguistic summaries, Information Sciences 614 (2022) 374 – 399, DOI: 10.1016/j.ins.2022.10.010] because the developed explainable methods will be take into account the temporal structure of data.
(3) Explainable linguistic summarization methods for data streams (M12-48, collaboration with prof. M. Dankova from the University of Ostrava )
The task will involve developing innovative algorithms for constructing linguistic summaries for the streams. data and complex medical data. The task will be expanded by, among others, the prototype developed by Kaczmarek-Majer et al. (2022) Explaining smartphone-based acoustic data in bipolar disorder: Semi-supervised fuzzy clustering and relative Linguistic summaries, Information Sciences 588 (2022) DOI: 10.1016/j.ins.2021.12.049.
(4) Interactive module (dashboard) for visualizing results in the form of linguistic summaries (M12-36)
The module will enable easy-to-interpret presentation of human-readable explanations. In this task, the following will be achieved following results: easy access to a web application enabling visualization of knowledge structures; interactive elements, managing a team of users, including sharing charts between users.
(5) Integration and technical validation of the ExplainMe system on benchmark data (M24-48)
The aim of this task is to integrate the modules (Task 2.1-2.4). A technical validation will be carried out, including including workshops for 100+ users evaluating the system’s performance (accuracy, increasing reliability and compliance) with the law; security, privacy, transparency, etc.). The result of the task will be D.2.3. Comparative methods will include deep learning networks and domain adaptive Few-Shot continuous learning. In parallel, Validation will be performed on actual data in Stage 3 for implementation data.