BEGIN:VCALENDAR VERSION:2.0 PRODID:IEEE vTools.Events//EN CALSCALE:GREGORIAN BEGIN:VTIMEZONE TZID:America/New_York BEGIN:DAYLIGHT DTSTART:20260308T030000 TZOFFSETFROM:-0500 TZOFFSETTO:-0400 RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3 TZNAME:EDT END:DAYLIGHT BEGIN:STANDARD DTSTART:20261101T010000 TZOFFSETFROM:-0400 TZOFFSETTO:-0500 RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11 TZNAME:EST END:STANDARD END:VTIMEZONE BEGIN:VEVENT DTSTAMP:20260520T172551Z UID:21772C88-DABF-4941-891C-6CC46EF5D52C DTSTART;TZID=America/New_York:20260520T120000 DTEND;TZID=America/New_York:20260520T130000 DESCRIPTION:Whole slide images (WSIs) provide rich spatial and contextual i nformation for cancer diagnosis\,\nyet their gigapixel scale and susceptib ility to artifacts\, adversarial noise\, and distribution shifts\ncomplica te reliable deep learning integration. In this talk\, a unified graph-base d framework for\nrobust and uncertainty-aware WSI analysis is presented. T he proposed approach represents WSIs\nas graphs to capture spatial depende ncies and leverages graph neural networks (GNNs)\,\nincluding Graph Convol utional Networks (GCNs)\, Graph Attention Networks (GATs)\, and\nGraphSAGE \, for feature learning. To enhance robustness\, a denoising module\, enab les input-\nspecific optimization that mitigates both acquisition-related artifacts (e.g.\, motion\, staining\nvariability) and adversarial perturba tions at image and graph levels. A transformer-based module\nfurther refin es global contextual modeling for accurate cancer classification across mu ltiple\ndatasets. To address reliability under distribution shifts\, we de ploy a multi-head GNN framework\nfor predictive uncertainty estimation\, w here head-wise divergence provides a principled measure\nof confidence and supports out-of-distribution detection. Extensive evaluations demonstrate that\nthe proposed framework significantly improves classification accura cy\, robustness to noise and\nattacks\, and uncertainty quantification com pared to conventional and non-robust methods.\n\n[Download Event Flyer](ht tps://events.vtools.ieee.org/event_media/download/92120)\n\n[]\n\nCo-spons ored by: IEEE CIS Schenectady Chapter\n\nSpeaker(s): Nasim\n\nVirtual: htt ps://events.vtools.ieee.org/m/557649 LOCATION:Virtual: https://events.vtools.ieee.org/m/557649 ORGANIZER:lkmestha@ieee.org, john.pole@ieee.org, murali.flute@gmail.com SEQUENCE:218 SUMMARY:IEEE CIS & CS Schenectady Chapter Technical Lecture on "Robust Grap h-based Learning in Digital Pathology" URL;VALUE=URI:https://events.vtools.ieee.org/m/557649 X-ALT-DESC:Description: <br /><p>Whole slide images (WSIs) provide rich spa tial and contextual information for cancer diagnosis\,<br>yet their gigapi xel scale and susceptibility to artifacts\, adversarial noise\, and distri bution shifts<br>complicate reliable deep learning integration. In this ta lk\, a unified graph-based framework for<br>robust and uncertainty-aware W SI analysis is presented. The proposed approach represents WSIs<br>as grap hs to capture spatial dependencies and leverages graph neural networks (GN Ns)\,<br>including Graph Convolutional Networks (GCNs)\, Graph Attention N etworks (GATs)\, and<br>GraphSAGE\, for feature learning. To enhance robus tness\, a denoising module\, enables input-<br>specific optimization that mitigates both acquisition-related artifacts (e.g.\, motion\, staining<br> variability) and adversarial perturbations at image and graph levels. A tr ansformer-based module<br>further refines global contextual modeling for a ccurate cancer classification across multiple<br>datasets. To address reli ability under distribution shifts\, we deploy a multi-head GNN framework<b r>for predictive uncertainty estimation\, where head-wise divergence provi des a principled measure<br>of confidence and supports out-of-distribution detection. Extensive evaluations demonstrate that<br>the proposed framewo rk significantly improves classification accuracy\, robustness to noise an d<br>attacks\, and uncertainty quantification compared to conventional and non-robust methods.</p>\n<p><a href="https://events.vtools.ieee.org/event _media/download/92120" target="_blank" rel="noopener">Download Event Flyer </a></p>\n<p><img src="https://events.vtools.ieee.org/vtools_ui/media/disp lay/87d171c8-cc53-4466-88e8-66fd51441302" alt="" width="1024" height="1357 "></p> END:VEVENT END:VCALENDAR