Researcher @Alibaba_Qwen, PhD @rlai_lab. Ideas are my own.
Joined December 2017
- Tweets76
- Following1K
- Followers706
- Likes278
Qingfeng Lan retweeted
Meta learning and recursive self-improvement are old ideas. Foundation models breathe new life into them. Our new survey, “Self-Improvements in Modern Agentic Systems,” reviews how the concepts are continuing to evolve.
Paper: arxiv.org/abs/2607.13104
Project: selfimproving-agent.github.i…
Github: github.com/selfimproving-age…
Qingfeng Lan retweeted
Excited to share our new survey! As Juergen noted, foundation models have breathed new life into the long-standing idea of self-improvement. But how has this happened, and what makes this new wave different?
Our survey seeks to answer these questions by developing a foundational understanding of self-improvement, tracing how the field has evolved and intersected with related research areas, and providing a comprehensive overview of current technologies and research.
We warmly welcome your suggestions, feedback, and discussion!
For long horizon tasks in LLM, PPO and GRPO are no longer suitable. Scaling off-policy algorithms will the next bet.
There are three wailing walls for intelligent systems: the exploration-exploitation dilemma, credit assignment, and continual learning.
# Made a small contribution
Putting in some extra hours on Chinese New Year's Eve – finally ready to enjoy the holidays!
🚀 Qwen3.5-397B-A17B is here: The first open-weight model in the Qwen3.5 series.
🖼️Native multimodal. Trained for real-world agents.
✨Powered by hybrid linear attention + sparse MoE and large-scale RL environment scaling.
⚡8.6x–19.0x decoding throughput vs Qwen3-Max
🌍201 languages & dialects
📜Apache2.0 licensed
🔗Dive in:
GitHub: github.com/QwenLM/Qwen3.5
Chat: chat.qwen.ai
API：modelstudio.console.alibabac…
Qwen Code: github.com/QwenLM/qwen-code
Hugging Face: huggingface.co/collections/Q…
ModelScope: modelscope.cn/collections/Qw…
blog: qwen.ai/blog?id=qwen3.5
Qingfeng Lan retweeted
🚀 Introducing Qwen3-Max-Thinking, our most capable reasoning model yet. Trained with massive scale and advanced RL, it delivers strong performance across reasoning, knowledge, tool use, and agent capabilities.
✨ Key innovations:
✅ Adaptive tool-use: intelligently leverages Search, Memory & Code Interpreter without manual selection
✅ Test-time scaling: multi-round self-reflection beats Gemini 3 Pro on reasoning
✅ From complex math (98.0 on HMMT Feb) to agentic search (49.8 on HLE)—it just thinks better.
🧠 Think deeper. Solve harder.
Try the adaptive reasoning experience now: chat.qwen.ai/
Completions API: modelstudio.console.alibabac…
Responses API: alibabacloud.com/help/en/mod…
blog: qwen.ai/blog?id=qwen3-max-th…
Qingfeng Lan retweeted
📢 Just 8 weeks until #CoLLAs2025!
Modern ML thrives in benchmarks but struggles in the wild. CoLLAs is where we tackle non-stationarity head-on: catastrophic forgetting, distribution shift, continual RL, online adaptation, and lifelong learning for ML.
📍 Aug 11–14 @Penn
🧠 Keynotes, Workshops, tutorials, posters & community
🔗 lifelong-ml.cc/Conferences/2…
#ContinualLearning #LifelongLearning #AI #MachineLearning
🚀RL algorithms are shaping the post-training of LLMs, but how do their objectives connect? In this blog, I explore their relationships and provide a unified perspective through the Policy Gradient Theorem—the backbone of policy gradient methods.
Dive in: lancelqf.github.io/note/llm_…
Qingfeng Lan retweeted
BREAKING: Amii Chief Scientific Advisor, Richard S. Sutton, has been awarded the A.M. Turing Award, the highest honour in computer science, alongside Andrew Barto! Read the official @TheOfficialACM announcement: hubs.la/Q039nZXM0
#TuringAward #AI #ReinforcementLearning
Qingfeng Lan retweeted
🚨 Call for Reviewers! 🚨
Want to contribute to advancing lifelong learning research? #CoLLAs2025 is looking for expert reviewers! 📝
Help shape the field by reviewing cutting-edge research in continual and lifelong learning.
🔗 Apply here: forms.gle/z1VFR4FRCbHjDvSa6
#MachineLearning #LifelongLearning #ContinualLearning #AI #AcademicReview
Qingfeng Lan retweeted
Would you believe that deep RL can work without replay buffers, target networks, or batch updates? Our recent work gets deep RL agents to learn from a continuous stream of data one sample at a time without storing any sample. Joint work with @Gautham529 and @rupammahmood.
Qingfeng Lan retweeted
Our NeurIPS paper is now on arXiv:
We introduce Action Value Gradient (AVG), a novel incremental deep RL method that learns in real-time, one sample at a time — no batch updates, target networks or a replay buffer!
Co-authors @mhmd_elsaye @bellingerc @white_martha @rupammahmood
Qingfeng Lan retweeted
RLC will be held at the Univ. of Alberta, Edmonton, in 2025. I'm happy to say that we now have the conference's website out: rl-conference.cc/index.html
We'll continue to update it, and the CFP will be out soon, but the relevant dates are already there.
@RL_Conference @UAlberta
Qingfeng Lan retweeted
📢 Exciting News! The Fourth Conference on Lifelong Learning Agents (CoLLAs 2025) will be held at the University of Pennsylvania (@Penn) in Philadelphia, USA 🇺🇸
🗓️ Important Dates:
Abstract Deadline: Feb 21, 2025
Submission Deadline: Feb 26, 2025
Conference Dates: Aug 11 - Aug 14, 2025
We invite submissions that present new theories, methodologies, applications, or insights into algorithms and benchmarks designed for non-i.i.d. and non-stationary settings. Accepted papers will be published in the Proceedings of Machine Learning Research (PMLR). 📚
Full CFP: lifelong-ml.cc/Conferences/2…
#CoLLAs2025 #AI #MachineLearning #ContinualLearning #LifelongLearning #ResearchConference #CallForPapers #NonStationaryLearning
Qingfeng Lan retweeted
A year later and our work on Loss of Plasticity is finally published, in Nature no less! The Nature version is totally rewritten and has many new results:
nature.com/articles/s41586-0…
Congratulations to the authors:
@s_dohare
@JFernandoHG
@LanceLan3
@rahman_parash
@rupammahmood
We finally have a version of our paper on loss of plasticity and continual backprop that is polished and submitted to a journal. Good work led by my PhD student Shibhansh Dohare.
arxiv.org/abs/2306.13812v2
Qingfeng Lan retweeted
It’s a deep learning problem that was ‘hidden in plain sight:’ A new Nature paper by Amii researchers explores why continual learning models can all of a sudden stop working, and what to do about it: hubs.la/Q02M4h8G0 #AI #ContinualLearning #MachineLearning
Qingfeng Lan retweeted
I couldn't be prouder of my colleagues at the @UAlberta! The work led by @s_dohare, in collaboration w/ J. F. H.-Garcia, @LanceLan3, @rahman_parash, @rupammahmood, & @RichardSSutton on continual learning and loss of plasticity is now published at @Nature!
nature.com/articles/s41586-0…
2. *Weight Clipping for Deep Continual and Reinforcement Learning*
In this paper, we introduce weight clipping, which clips neural network weights to limit them to a specific range.
This simple technique can be easily added to existing optimizers.
It helps improve generalization, maintain plasticity, stabilize learning, and facilitate learning with a large replay ratio. Joint work with Mohamed Elsayed @mhmd_elsaye, Clare Lyle @clarelyle, and Rupam Mahmood.
- Paper: arxiv.org/abs/2407.01704
- Code: github.com/mohmdelsayed/weig…
