BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:Asia/Muscat BEGIN:DAYLIGHT DTSTART:20380119T071407 TZOFFSETFROM:+0400 TZOFFSETTO:+0400 RRULE:FREQ=YEARLY;BYDAY=3TU;BYMONTH=1 TZNAME:+04 END:DAYLIGHT BEGIN:STANDARD DTSTART:19200101T001848 TZOFFSETFROM:+0341 TZOFFSETTO:+0400 RRULE:FREQ=YEARLY;BYDAY=1TH;BYMONTH=1 TZNAME:+04 END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20261021T035127Z UID:7CA90B3F-B411-464D-AB48-42B1EB650AC4 DTSTART;TZID=Asia/Muscat:20261021T110000 DTEND;TZID=Asia/Muscat:20261021T120000 DESCRIPTION:Deep learning has become a central tool in Industry 4.0\, where models trained on multivariate time series can detect complex patterns th at are difficult for human experts to perceive. However\, in operationally sensitive applications\, predictive accuracy alone is not sufficient: a p rediction that cannot be understood or justified can be difficult to trust and act upon.\n\nThis talk presents a post-hoc explainable AI framework t hat aims to bridge the gap between predictive performance and actionable u nderstanding. The central idea is to move beyond feature attribution and g enerate counterfactual explanations that answer a more practical question: what should change for the prediction to change? Rather than relying only on numerical proximity\, the proposed approach uses domain knowledge to e nsure that counterfactual explanations are plausible and semantically cons istent with the application context.\n\nThe framework combines predictive models for multivariate time series\, counterfactual explanation generatio n\, ontology-based validation\, and semantic verification using rules. It also integrates large language models to extract structured knowledge from domain documents and to transform validated explanation artefacts into ac cessible\, human-readable explanations. The language model therefore acts as a synthesis layer over validated knowledge rather than as the source of scientific correctness.\n\nThe approach is illustrated through two indust rial applications: lithium-ion battery state estimation and production-lin e quality monitoring. The predictive models span several architectures\, i ncluding CNNs\, LSTMs\, spiking neural networks\, and Transformer-based mo dels. The talk will highlight how combining machine learning with domain k nowledge can make explanations not only more understandable\, but also mor e relevant to real-world industrial decision-making.\n\n[Click to join](ht tps://events.vtools.ieee.org/tego_/event/From%20Prediction%20to%20Actionab le%20Understanding:%20Knowledge-Grounded%20Counterfactual%20Explanations%2 0for%20Multivariate%20Time%20Series)(event starts on Oct 21 at 11AM\, GMT+ 4)\n\nSpeaker(s): Slimane Arbaoui\n\n***CANCELED***\n\nVirtual: https://ev ents.vtools.ieee.org/m/584718 LOCATION:Virtual: https://events.vtools.ieee.org/m/584718 ORGANIZER:nour.aburaed@ieee.org SEQUENCE:16 SUMMARY:From Prediction to Actionable Understanding: Knowledge-Grounded Cou nterfactual Explanations for Multivariate Time Series URL;VALUE=URI:https://events.vtools.ieee.org/m/584718 X-ALT-DESC:Description: <br /><p>Deep learning has become a central tool in Industry 4.0\, where models trained on multivariate time series can detec t complex patterns that are difficult for human experts to perceive. Howev er\, in operationally sensitive applications\, predictive accuracy alone i s not sufficient: a prediction that cannot be understood or justified can be difficult to trust and act upon.</p>\n<p>This talk presents a post-hoc explainable AI framework that aims to bridge the gap between predictive pe rformance and actionable understanding. The central idea is to move beyond feature attribution and generate counterfactual explanations that answer a more practical question: <em>what should change for the prediction to ch ange?</em> Rather than relying only on numerical proximity\, the proposed approach uses domain knowledge to ensure that counterfactual explanations are plausible and semantically consistent with the application context.</p >\n<p>The framework combines predictive models for multivariate time serie s\, counterfactual explanation generation\, ontology-based validation\, an d semantic verification using rules. It also integrates large language mod els to extract structured knowledge from domain documents and to transform validated explanation artefacts into accessible\, human-readable explanat ions. The language model therefore acts as a synthesis layer over validate d knowledge rather than as the source of scientific correctness.</p>\n<p>T he approach is illustrated through two industrial applications: lithium-io n battery state estimation and production-line quality monitoring. The pre dictive models span several architectures\, including CNNs\, LSTMs\, spiki ng neural networks\, and Transformer-based models. The talk will highlight how combining machine learning with domain knowledge can make explanation s not only more understandable\, but also more relevant to real-world indu strial decision-making.</p>\n<p><span style="background-color: rgb(251\, 2 38\, 184)\;"><strong><a style="background-color: rgb(251\, 238\, 184)\;" h ref="https://events.vtools.ieee.org/tego_/event/From%20Prediction%20to%20A ctionable%20Understanding:%20Knowledge-Grounded%20Counterfactual%20Explana tions%20for%20Multivariate%20Time%20Series">Click to join </a>(event start s on Oct 21 at 11AM\, GMT+4)</strong></span></p>\n<p><img src="https://eve nts.vtools.ieee.org/vtools_ui/media/display/ad37526a-23c2-4e2d-985e-b3f98d e2c3a1" width="822" height="1461"></p>\n<p>&nbsp\;</p> END:VEVENT END:VCALENDAR