Luisa Zintgraf
@luisa_zintgraf
Followers
5K
Following
4K
Media
76
Statuses
576
Research Scientist @GoogleDeepMind. PhD from @UniofOxford.
London
Joined January 2014
Proud of this work and the incredible team at @GoogleDeepMind ✨ Huge shout-out to my co-first authors @dancalian, @greg_far, & @iurii_kemaev. And to our amazing collaborators: @matteohessel, @shar_jeremy, @junh_oh, András György, Tom Schaul, @JeffDean, @hado, and David Silver.
0
1
10
We believe that the DataRater is a promising step towards more automated and principled dataset curation. This could be especially important for filtering and making the best use of massive synthetic datasets in the future. For a deeper dive, check out
1
3
12
So what does the DataRater learn? It automatically identifies and down-weights data that aligns with human intuitions of low quality, such as incorrect text encodings, OCR errors, and irrelevant content.
2
1
11
The result? The DataRater is highly effective at filtering data, leading to significant compute efficiency improvements. In our experiments, we observed up to a 46.6% net compute gain while often improving final model performance.
3
2
11
We introduce the DataRater, a meta-learning method that learns to rate the value of each data point for training. Instead of manually specifying filtering rules, we train the DataRater to optimize for a simple goal: improving the training efficiency on a held-out dataset.
1
1
11
Foundation models are trained on large datasets, but not all data is created equal. Dataset curation often relies on manual, coarse-grained filtering and hand-crafted rules. This is becoming a major challenge, especially with the rise of synthetic data.
1
1
14
Excited to share our new paper, "DataRater: Meta-Learned Dataset Curation"! We explore a fundamental question: How can we *automatically* learn which data is most valuable for training foundation models? Paper: https://t.co/N2ozU2RXWb to appear @NeurIPSConf Thread 👇
10
51
326
📘 Journal: https://t.co/oCDkuEY2o6 📝 ArXiv: https://t.co/QBTxKUbvlQ 🎙️ Podcast: https://t.co/wjA8JGLBD2 🎥 Talk:
0
0
9
🎉 Our Meta-RL survey is now published in Foundations and Trends in Machine Learning! A deep dive into how agents can learn to learn 🤖🧠 Huge kudos to @jakeABeck & @ristovuorio for leading the charge, and to co-authors Evan Liu, Zheng Xiong, @chelseabfinn & @shimon8282!
2
13
58
Big news—our survey paper “A Tutorial on Meta-Reinforcement Learning” is officially published! Meta-RL = learning how to adapt through interaction. It embraces The Bitter Lesson: don’t hardcode agents—train them to adapt on their own https://t.co/R3qHbNTGnW 🧵⬇️
arxiv.org
While deep reinforcement learning (RL) has fueled multiple high-profile successes in machine learning, it is held back from more widespread adoption by its often poor data efficiency and the...
2
80
339
Wanna work on Gemini? DeepMind is hiring! 🚀
Interested in helping us make Gemini Pro even better? The Gemini pre-training team is looking for a Research Scientist in London to push the boundaries of LLM scaling: understanding, predicting, and improving. ♊️🚀 Apply here:
1
2
22
Interested in helping us make Gemini Pro even better? The Gemini pre-training team is looking for a Research Scientist in London to push the boundaries of LLM scaling: understanding, predicting, and improving. ♊️🚀 Apply here:
job-boards.greenhouse.io
2.0 Pro Experimental is our best model yet for coding and complex prompts, refined with your feedback. 🤝 It has a better understanding of world-knowledge and comes with our largest context window yet of 2 million tokens - meaning it can analyze large amounts of information.
0
21
63
It's that time of year again! We've just announced our game intelligence research internship - join us to learn, work with a fantastic team, and tackle hard problems. "Internship Opportunity: Research Intern – Multimodal Generative Models for Video Games"
1
23
116
Very excited to talk about Leveraging AlphaZero to Improve our Understanding & Creativity in Chess ♟️🤯 with @_beenkim at the @StanfordHAI Fall Conference! In this work, we dig into finding chess concepts that are beyond the current collective human knowledge 🧵1/3
3
22
102
RL agents are notoriously slow to learn 🐢 However, meta-RL can make RL agents that learn fast!🔥 Check out this talk introducing the field of Meta-RL just given by our lab members @jakeABeck and @ristovuorio in Berlin at AutoML 2023! 📺 Link:
2
17
52
🔥 Podcast episode on Meta-RL 🔥
Episode 39 @jakeABeck and @ristovuorio of @whi_rl at @UniofOxford on their recent Survey of Meta-Reinforcement Learning. https://t.co/d5TCHfNdTb
0
1
24
Heyyo! Was just interviewed by the TalkRL Podcast!! 🎙️🔥 @ristovuorio and I explain meta-RL. Give it a listen!
Episode 39 @jakeABeck and @ristovuorio of @whi_rl at @UniofOxford on their recent Survey of Meta-Reinforcement Learning. https://t.co/d5TCHfNdTb
0
6
17
Check out a new @TalkRLPodcast episode with @jakeABeck and me where we talk about our recent meta-RL survey with Evan Liu, Zheng Xiong, @luisa_zintgraf, @chelseabfinn, and @shimon8282. This should hopefully be an accessible discussion to anyone in ml curious about meta-RL!
Episode 39 @jakeABeck and @ristovuorio of @whi_rl at @UniofOxford on their recent Survey of Meta-Reinforcement Learning. https://t.co/d5TCHfNdTb
0
4
14
Episode 39 @jakeABeck and @ristovuorio of @whi_rl at @UniofOxford on their recent Survey of Meta-Reinforcement Learning. https://t.co/d5TCHfNdTb
podcasts.apple.com
Podcast Episode · TalkRL: The Reinforcement Learning Podcast · 03/07/2023 · 1h 7m
1
3
20