【セミナーシリーズ】継続学習エージェントセミナー（次回6/10 水）

[Keigo Nishida]

IBISMLの皆様

理化学研究所の西田圭吾と申します。

理研AIP Emtiyaz Khan チームディレクタの代理で投稿いたします。

CoLLAs (Conference on Lifelong Learning Agents)では、継続学習エージェントに関するセミナーシリーズを開催しております。

主なトピックは、継続学習、Lifelong学習、適応的機械学習です。

https://lifelong-ml.cc/seminar

このセミナーは、日本からも参加しやすい時間帯で開催されるとのことです。

ご興味のある方はぜひご参加ください。

セミナーは原則として毎月第1水曜日に開催されますが、

回によって米国時間または欧州時間での開催となるようですので、ご注意ください。

次回の予定は以下のとおりです。

日時については、参加される方ご自身でも公式ページをご確認ください。

日時：6月10日 10:00（中央ヨーロッパ時間）

日本時間：6月10日 18:00頃

Speaker: Rahaf Aljundi（Toyota Motor Europe）

Title: Towards Continuous Accumulation of Knowledge and Experience

https://lifelong-ml.cc/seminar/next

 以上、どうぞよろしくお願い申し上げます。

西田

---------- Forwarded message ---------

From: Sarath Chandar Anbil Parthipan <sarath....@mila.quebec>

Date: Thu, Jun 4, 2026 at 4:53 AM

Subject: [CoLLAs Seminar][June-10 10am CET] Rahaf Aljundi on "Continuous Accumulation of Knowledge and Experience"

To: Contact CoLLAs <con...@lifelong-ml.cc>

Dear all,

We are excited to bring you the second speaker of the monthly CoLLAs Seminar Series, which has an excellent line-up of speakers from around the world: https://lifelong-ml.cc/seminar

With this series, we aim to bring together researchers working on continual, lifelong, and adaptive machine learning to share new ideas, and foster community-wide dialogue!

Our second talk will be by Rahaf Aljundi on June 10 at 10:00 AM CET.

Title: Towards Continuous Accumulation of Knowledge and Experience

Abstract: Continual learning enables models to adapt to streaming data, targeting knowledge accumulation, and avoiding catastrophic forgetting. Continual learning research has focused on continual parameter updates; however, forgetting and learning instability always seem unavoidable, rendering continual learning a far‑fetched problem. On the other hand, in‑context learning offers a complementary path to frequent parameter updates. In this talk, I argue that efficient, continuous adaptation needs not occur solely in parameter space. I will show how memory‑based mechanisms enable rapid adaptation and present some of my previous and recent continual learning works, then discuss how combining fast, memory‑driven updates with slow model consolidation could shape the future of continual learning.

Speaker bio: Rahaf Aljundi is a research scientist at Toyota Motor Europe. She is interested in building models that can keep learning (given any source of supervision), can tell when a novel input is provided and further point at instances that require annotation. This includes the topics of continual learning (incremental learning, lifelong learning), novelty detection (out of distribution, anomaly detection), active learning and also domain adaptation. She has obtained her PhD degree from KU Leuven university (Belgium) where she extensively worked on the topic of continual learning.

Zoom: https://polymtl-ca.zoom.us/j/84066718835?pwd=2cpVMN2XjRXyZrngOFBJSO3Fwifywo.1

We look forward to seeing you there!

Best regards,

Sarath Chandar

On behalf of the CoLLAs Seminar Organizers

--

西田圭吾（Keigo NISHIDA）

理化学研究所　基礎科学特別研究員

革新知能統合研究センター（AIP）　マセマティカルインテリジェンスグループ　適応ベイズ知能チーム