Computer Vision News - December 2022

45 Mara Graziani and Vincent Andrearczyk user tests is on the to-do list for everyone developing interpretable models. “ The point of this work was not only the definition but also restructuring the way we develop systems and adding social scientists, who study human understanding and how we process information, in the loop, ” Mara points out. “ I’ve already seen other papers published after ours do this, so I think it’s a shared view, and it’s one we should push forward! ” a unique database from the University of Lausanne with MRIs at different time points for evaluating the segmentation of brain metastases, theevolutionof themetastases, the appearance of new metastases, and patient outcomes. I’m working on the technical side, and we have experts at the hospital helping us. ” In terms of next steps, Mara and Vincent plan further development to integrate the human in the loop and say that running Element of transparency. Should be understood by the average audience Providing already a sufficient level of understandability by inherent or built-in interpretability Interpretable AI terminology Main terms and domains TERM In DOMAIN 2. EXPLAINA BLE/INTER PRETABLE 5. ACCOUNT ABLE meaning Usedwith 7. AUDITABL E 3. TRANSPARE NT EUROPEAN UNIONLAW PHILOSOPHY SOCIOLOGY LABOUR COGNITIVE SCIENCES TECHNOLOGY/ SCIENCES Provide meaningful information about the underlying logic, the significance and envisaged consequences of the processing. Fair. Accountability before and after development and deployment Guiding the reasoning and assigning causal responsibilities Novelty was not always directly explainable in history Building a social relationship of trust between humans and machines Component of technical robustness, ensuring correct functioning with a range of inputs and situations Means to improve worker’s satisfaction and safety at work The enablement of the assessment of algorithms, data and design processes Providing justifications for predictions 4. INTELLIGI BLE Understanding the mechanism, ensuring model simulabilit and decomposability Explain the technical process of the AI system and its application Transparency of the data, the system and business models 6. RELIABLE Element of transparency. Should be understood by the average audience Interpretable AI terminology Main terms and domains TERM In DOMAIN 2. EXPLAIN BLE/INT PRETAB meaning Used with 7. AUDITABL E EUROPEAN UNION LAW Provide meaningful information about the underlying logic, the significance and envisag d consequences of the processing. Novelty was not always directly explainable in 4. INTELL BLE Explain the technical process of the AI system and its application Transparency of the data, the system and business models