摘要：终身学习智能体会议 (Conference on Lifelong Learning Agents，CoLLAs) 是AI、机器学习等相关领域研究人员的年度学术会议，旨在交流如何推进机器学习，使AI系统能够在其使用期间不断学习，知识积累渐进上升、推断与识别能力

1 会议简介

终身学习智能体会议 (Conference on Lifelong Learning Agents，CoLLAs) 是AI、机器学习等相关领域研究人员的年度学术会议，旨在交流如何推进机器学习，使AI系统能够在其使用期间不断学习，知识积累渐进上升、推断与识别能力逐渐加强，并在复杂的非平稳环境中表现良好。CoLLAs由Mila， Google DeepMind，Google Brain，蒙特利尔大学，宾大，阿尔伯塔大学等机构研究人员组织发起。首届会议CoLLAs 2022于2022年8月22日至24日在加拿大蒙特利尔的麦吉尔大学举行。

CoLLAs 的Advisory Board由多位来自世界各国的著名AI、机器学习专家学者组成：

Abhinav Gupta - Carnegie Mellon University

Chelsea Finn - Stanford University

Claudia Clopath - Imperial College London

Hugo Larochelle - Google Brain

Joelle Pineau - McGill University and FAIR

Jürgen Schmidhuber - KAUST, Swiss AI Lab IDSIA, USI, AIRI, and NNAISENSE

Konrad Kording - University of Pennsylvania

Kyunghyun Cho - New York University

Leslie Kaelbling - MIT

Marc Aurelio Ranzato - DeepMind

Michael Littman - Brown University

Peter Stone - University of Texas at Austin

Pieter Abbeel - University of California Berkeley

Raia Hadsell - DeepMind

Sepp Hochreiter - Johannes Kepler University Linz

Yee Whye Teh - University of Oxford and DeepMind

Yoshua Bengio - University of Montreal

2 会议组织

CoLLAs 2025是第四届会议，计划于2025年在美国费城的宾夕法尼亚大学举行。会议主席 General Chairs由Mila的Sarath Chandar 和Google DeepMind的Razvan Pascanu组成。Program Chairs 由Eric Eaton (University of Pennsylvania), Bing Liu (University of Illinois Chicago), Rupam Mahmood (University of Alberta), 以及Amal Rannen-Triki (Google DeepMind)组成。

此外，还有以下来自多位学术界和工业界的Associate Program Chairs：

3 会议主题以及征稿范围

CoLLAs会议涵盖了多种与持续学习相关的主题，包括但不限于：

终身学习/持续学习的理论研究

持续学习范式（如类别增量、任务增量、领域增量、课程学习、主动学习、联邦学习、在线学习、元学习、小样本学习等）

非平稳学习中的挑战（如遗忘、分布漂移、开放环境学习等）

持续强化学习（如选项、技能发现、分层强化学习、内在动机学习、多智能体强化学习）

大规模语言模型的持续学习（如上下文学习、预训练、模型编辑、微调、适应）

知识迁移（如迁移学习、多任务学习、领域适应、模拟到现实、元学习）

非平稳优化

流式学习、设备端实时学习

开放世界学习、开放式学习

神经科学启发的持续学习

应用（如控制、机器人、医疗保健等）

数据集、基准测试、评估、软件库

会议就以上主题征稿。会议还特别鼓励探索持续学习与其他领域（包括神经科学、机器人、教育等）之间的联系。

4 重要日期

Abstract deadline: Feb 21, 2025

Submission deadline: Feb 26, 2025

Review released: Apr 7, 2025

Author rebuttals due: Apr 15, 2025

Notifications of decision: May 12, 2025

Resubmissions deadline: Jun 12, 2025

Decision of resubmissions: Jun 23, 2025

以下为英文征稿：

The Conference on Lifelong Learning Agents (CoLLAs) is an annual gathering of researchers to exchange ideas on advancing machine learning toward systems that can continually learn throughout their lifetime, become increasingly knowledgeable and capable, and perform well in complex non-stationary environments.

Traditional machine learning has largely relied on the assumption that data samples are i.i.d., which emphasizes optimizing test performance within a single task. While this approach has been highly influential, it limits further progress. One of the defining traits of human intelligence is the ability to learn continually and adaptively. In recent years, attention has shifted to alternative paradigms that embrace continual learning under non-i.i.d. and non-stationary settings and move away from a single fixed objective.

We invite submissions to the 4th edition of CoLLAs that describe new theories, methodologies, applications, or insights into existing algorithms and benchmarks for learning in non-i.i.d. and non-stationary settings. Accepted papers will be published in the Proceedings of Machine Learning Research (PMLR). Topics of submission may include but are not limited to:

Theory for continual/lifelong learning

Continual learning paradigms(class-incremental, task incremental, domain incremental, curriculum learning, active learning, federated learning, online learning, meta-learning, few-shot learning, and other non-stationary learning paradigms)

Challenges with non-stationary learning(loss of plasticity, catastrophic forgetting, policy collapse, unlearning, OOD generalization, distribution shift, etc.)

Continual reinforcement learning(options, skill discovery, hierarchical RL, intrinsically motivated learning, multi-agent RL)

Continual learning in LLMs(in-context learning, pre-training, model editing, fine-tuning, adaptation)

Knowledge transfer(transfer learning, multi-task learning, domain adaptation, sim2real, meta-learning)

Non-stationary optimization

Streaming learning, on-device, real-time learning

Open-world learning, open-ended learning

Neuroscience-inspiredcontinual/lifelong learning

Applications(control, robotics, healthcare, etc.)

Datasets, benchmarks, evaluation, software libraries

We are particularly interested in submissions that explore connections between continual learning and other areas, including neuroscience, robotics, and education.

Submitted papers will be evaluated based on their novelty, technical quality, and potential impact. Experimental methods and results are expected to be reproducible, and authors are strongly encouraged to make code and data available. We also encourage submissions of proof-of-concept research that puts forward novel ideas and demonstrates potential, as well as in-depth analysis of existing methods and concepts.

CoLLAs 2025 will be held at the University of Pennsylvania in Philadelphia, USA.

For any questions about the conference, you can contact us at contact@lifelong-ml.cc.

Dates and Deadlines (Anywhere on Earth)

来源：AI科技评论