[Eoas-seminar] [Seminar-announce] Scientific Computing Colloquium with Adrian Barbu

[eoas-seminar at lists.fsu.edu]

"Scalable Learning with Probabilistic PCA"

Adrian Barbu
Department of Statistics, Florida State University

Please feel free to forward/share this invitation with other groups/disciplines that might be interested in this talk/topic. All are welcome to attend.

NOTE: In-person attendance is requested in our 499 Dirac Science Library (DSL) Seminar Room. Zoom access is intended for external (non-departmental) participants only.

https://fsu.zoom.us/j/94273595552
Meeting # 942 7359 5552

🎦 Colloquium recordings will be made available here, https://www.sc.fsu.edu/colloquium<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.sc.fsu.edu%2Fcolloquium&data=05%7C02%7Csc-seminar-announce%40lists.fsu.edu%7Cee7743f828d5455dbf6b08dd4219d171%7Ca36450ebdb0642a78d1b026719f701e3%7C0%7C0%7C638739400828235727%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=ALejqfr6e52DSXvOpvaU2ZTrBfBqzc3R0y4fT%2BZRqHg%3D&reserved=0>

Wednesday, Feb 5, 2025, Schedule:

* 3:00 to 3:30 PM Eastern Time (US and Canada)
☕ Nespresso & Teatime - 417 DSL Commons

* 3:30 to 4:30 PM Eastern Time (US and Canada)
🕟 Colloquium - 499 DSL Seminar Room

Abstract:
Deep neural networks have drawn much attention due to their success in various vision tasks, including classification. Class Incremental Leaning is a paradigm where instances from new object classes are added sequentially. Traditional methods are faced with catastrophic forgetting, where the updated model forgets the old classes and focuses only on the new classes. In this work, we introduce a framework called incremental PPCA for class incremental learning. It uses a self-supervised pre-trained feature extractor to obtain meaningful features and trains Probabilistic PCA models on the extracted features for each class separately. The Mahalanobis distance is used to obtain the classification result, and an equivalent equation is derived to make the approach computationally affordable. Experiments on standard and large datasets show that the proposed approach outperforms existing state-of-the-art incremental learning methods by a large margin. The fact that the model is trained on each class separately makes it applicable to training on very large datasets such as the whole ImageNet with more than 10,000 classes. To better handle so many classes, we take inspiration from our understanding of the human hierarchical cognition models and propose a framework called Hierarchical PPCA for image classification. The framework uses probabilistic PCA models as basic classification units and groups the image classes into a smaller number of super-classes. During classification, Hierarchical PPCA assigns a sample to a small number of most likely super-classes, and restricts the image classification to the image classes corresponding to these super-classes. Experiments on ImageNet indicate the hierarchical classifier can achieve a 4-16 times speedup compared to a standard classifier without any loss in accuracy.

Additional colloquium details can be found here,
https://www.sc.fsu.edu/news-and-events/colloquium/1856-colloquium-with-adrian-barbu-2025-02-5<https://nam04.safelinks.protection.outlook.com/?url=https%3A%2F%2Fwww.sc.fsu.edu%2Fnews-and-events%2Fcolloquium%2F1856-colloquium-with-adrian-barbu-2025-02-5&data=05%7C02%7Csc-seminar-announce%40lists.fsu.edu%7Cee7743f828d5455dbf6b08dd4219d171%7Ca36450ebdb0642a78d1b026719f701e3%7C0%7C0%7C638739400828250929%7CUnknown%7CTWFpbGZsb3d8eyJFbXB0eU1hcGkiOnRydWUsIlYiOiIwLjAuMDAwMCIsIlAiOiJXaW4zMiIsIkFOIjoiTWFpbCIsIldUIjoyfQ%3D%3D%7C0%7C%7C%7C&sdata=tlbwAdWOnLV7V4xAFp46%2Fry0oPtj%2BUstueyZFqglsmM%3D&reserved=0>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.fsu.edu/pipermail/eoas-seminar/attachments/20250131/4e5e3a2c/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: text/calendar
Size: 5793 bytes
Desc: not available
URL: <http://lists.fsu.edu/pipermail/eoas-seminar/attachments/20250131/4e5e3a2c/attachment.ics>
-------------- next part --------------
_______________________________________________
SC-Seminar-announce mailing list
SC-Seminar-announce at lists.fsu.edu
https://lists.fsu.edu/mailman/listinfo/sc-seminar-announce