正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

Does off-policy value-based RL scale? In LLMs, larger scale predictably improves performance. Value-based RL learns from arbitrary data and is sample-efficient, but folk wisdom says it doesn't scale 🧵⬇️We show predictability for scaling value-based RL!

Oleg Rybkin

1,333 subscribers

23,968 次观看 • 1 年前 •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 条评论

暂无评论

原始帖子的评论将显示在这里

相关视频

🚨Current scalable RL algos train a policy w/o value func, which is limiting with learning in open-ended, non-stationary, dynamic environments. But, how to scale value-based RL with more data/compute is unclear... Not anymore: presenting scaling laws for value-based RL 🧵⬇️

🚨Current scalable RL algos train a policy w/o value func, which is limiting with learning in open-ended, non-stationary, dynamic environments. But, how to scale value-based RL with more data/compute is unclear... Not anymore: presenting scaling laws for value-based RL 🧵⬇️

Aviral Kumar

37,286 次观看 • 1 年前

Introducing CQN: Coarse-to-fine Q-Network, a value-based RL algorithm for continuous control🦾Initialized with 20~50 demonstrations, it learns to solve real-world robotic tasks within 10 mins of training, without any pre-training and shaped rewards! (1/4)

Introducing CQN: Coarse-to-fine Q-Network, a value-based RL algorithm for continuous control🦾Initialized with 20~50 demonstrations, it learns to solve real-world robotic tasks within 10 mins of training, without any pre-training and shaped rewards! (1/4)

Younggyo Seo

16,413 次观看 • 1 年前

New research from Databricks: LLMs Can Learn to Reason via Off-Policy RL Optimal Advantage-based Policy Optimization with Lagged Inference policy (OAPL) shows you don’t need strict on-policy training to improve reasoning. It matches or beats Group Relative Policy Optimization (GRPO), stays stable with large policy lag, and uses ~3× fewer training generations. For Databricks customers, it’s a simpler, practical, and equally powerful approach to RL that Databricks is pioneering internally — and bringing directly to Databricks customers, so enterprises can improve agents using the same methods we use for our in-house agents, without complex infrastructure changes.

New research from Databricks: LLMs Can Learn to Reason via Off-Policy RL Optimal Advantage-based Policy Optimization with Lagged Inference policy (OAPL) shows you don’t need strict on-policy training to improve reasoning. It matches or beats Group Relative Policy Optimization (GRPO), stays stable with large policy lag, and uses ~3× fewer training generations. For Databricks customers, it’s a simpler, practical, and equally powerful approach to RL that Databricks is pioneering internally — and bringing directly to Databricks customers, so enterprises can improve agents using the same methods we use for our in-house agents, without complex infrastructure changes.

Databricks AI Research

12,439 次观看 • 3 个月前

US-based K-Scale Labs launched pre-orders for its open-source humanoid, priced at $9K The K-Bot stands 4′7″, weighs 77lbs, and integrates with K-Scale's open-source stack - covering RL/VLA models, Rust firmware, sim2real pipeline, and hardware design

US-based K-Scale Labs launched pre-orders for its open-source humanoid, priced at $9K The K-Bot stands 4′7″, weighs 77lbs, and integrates with K-Scale's open-source stack - covering RL/VLA models, Rust firmware, sim2real pipeline, and hardware design

Brett Adcock

10,994 次观看 • 11 个月前

1/ While most RL methods use shallow MLPs (~2–5 layers), we show that scaling up to 1000-layers for contrastive RL (CRL) can significantly boost performance, ranging from doubling performance to 50x on a diverse suite of robotic tasks. Webpage+Paper+Code:

1/ While most RL methods use shallow MLPs (~2–5 layers), we show that scaling up to 1000-layers for contrastive RL (CRL) can significantly boost performance, ranging from doubling performance to 50x on a diverse suite of robotic tasks. Webpage+Paper+Code:

Kevin Wang

154,559 次观看 • 1 年前

Your value doesn't decrease based on someone's inability to see your worth 💌

Your value doesn't decrease based on someone's inability to see your worth 💌

Persephanii Aka Thick Yonce

380,674 次观看 • 2 年前

Why am I working on RL for LLMs, when I can go work on RL for Olaf... #humanoidssummit

Why am I working on RL for LLMs, when I can go work on RL for Olaf... #humanoidssummit

Shane Gu

12,440 次观看 • 6 个月前

Frontier research just crossed a new threshold. Mind Lab is now public. And it begins with trillion-scale RL trained at only 10% of the usual compute.

Frontier research just crossed a new threshold. Mind Lab is now public. And it begins with trillion-scale RL trained at only 10% of the usual compute.

Chidanand Tripathi

89,813 次观看 • 6 个月前

Thanks AK! Finally, robot can do continuous, agile, autonomous, adaptive jumping over stair and stepping stone Key idea: combine the pros of model-free RL and model-based control. RL (for CoM refs) + QP (for GRF) + WBC (for torque) Open-sourced:

Thanks AK! Finally, robot can do continuous, agile, autonomous, adaptive jumping over stair and stepping stone Key idea: combine the pros of model-free RL and model-based control. RL (for CoM refs) + QP (for GRF) + WBC (for torque) Open-sourced:

Guanya Shi

32,142 次观看 • 1 年前

Does LLM RL post-training need to be on-policy?

Does LLM RL post-training need to be on-policy?

Kianté Brantley

113,263 次观看 • 3 个月前

RL is back! But is it always the best choice? In a new paper, we investigate under what circumstances neuroevolution outperforms RL in transfer learning tasks. See more details in the thread below 🧵 While NE performs best in simpler domains, it will be interesting to see if the lessons learned here can also be applied to more complex systems/tasks (LLMs?).

RL is back! But is it always the best choice? In a new paper, we investigate under what circumstances neuroevolution outperforms RL in transfer learning tasks. See more details in the thread below 🧵 While NE performs best in simpler domains, it will be interesting to see if the lessons learned here can also be applied to more complex systems/tasks (LLMs?).

Sebastian Risi

11,130 次观看 • 10 个月前

X_Acc_Flags - for ios The value of ‘Location (accurate)’ is coming from X’s servers, so I have no idea what it was based on Source-Code :

X_Acc_Flags - for ios The value of ‘Location (accurate)’ is coming from X’s servers, so I have no idea what it was based on Source-Code :

 CrazyMind

19,392 次观看 • 6 个月前

"I think just based on vehicle autonomy, we can 10x the value of the company." 一 Elon Musk

"I think just based on vehicle autonomy, we can 10x the value of the company." 一 Elon Musk

DogeDesigner

37,893 次观看 • 11 个月前

Crypto can’t scale without solving value transfer. Not just between blockchains but across fiat, stablecoins, and bank networks. We’re building it in Kima! From CBDCs to DEXs, from TradFi to DeFi this is how money moves in the bridgeless Kima era. Join us. 💥

Crypto can’t scale without solving value transfer. Not just between blockchains but across fiat, stablecoins, and bank networks. We’re building it in Kima! From CBDCs to DEXs, from TradFi to DeFi this is how money moves in the bridgeless Kima era. Join us. 💥

Kima Network

25,026 次观看 • 11 个月前

This figure from HIL-SERL is one of the clearest visualisations of how RL learns differently from imitation learning. The difference comes down to this: imitation learning treats each (state, action) pair as independent. A correction at timestep 20 teaches nothing about timestep 19 or 21. RL propagates reward backward through time. One successful insertion updates the value estimate of every state along the trajectory. So RL builds a full map of "which states lead to success"; imitation learning just memorizes individual snapshots. Setup: a robot inserting a RAM stick into a motherboard slot. Each dot is an end-effector position (Y = lateral, Z = height). Starting position is randomized. Left to right = training progressing. Top row (RL): the policy builds a funnel. Broad at the top, narrowing into the target. It systematically fills in the state space, learning which paths lead to success from many different starting positions. Bottom row (imitation learning / HG-DAgger, same human data): sparse, diffuse, no funnel. The policy only learns near states the human demonstrated. Both have access to the same data, including human corrections, but a completely different structure emerges.

This figure from HIL-SERL is one of the clearest visualisations of how RL learns differently from imitation learning. The difference comes down to this: imitation learning treats each (state, action) pair as independent. A correction at timestep 20 teaches nothing about timestep 19 or 21. RL propagates reward backward through time. One successful insertion updates the value estimate of every state along the trajectory. So RL builds a full map of "which states lead to success"; imitation learning just memorizes individual snapshots. Setup: a robot inserting a RAM stick into a motherboard slot. Each dot is an end-effector position (Y = lateral, Z = height). Starting position is randomized. Left to right = training progressing. Top row (RL): the policy builds a funnel. Broad at the top, narrowing into the target. It systematically fills in the state space, learning which paths lead to success from many different starting positions. Bottom row (imitation learning / HG-DAgger, same human data): sparse, diffuse, no funnel. The policy only learns near states the human demonstrated. Both have access to the same data, including human corrections, but a completely different structure emerges.

Dominique Paul

24,399 次观看 • 3 个月前

Polygon 2.0 is a concrete vision to build the Value Layer of the Internet, and expand Ethereum to Internet-scale. It is a series of proposals for unlimited scalability & unified liquidity.

Polygon 2.0 is a concrete vision to build the Value Layer of the Internet, and expand Ethereum to Internet-scale. It is a series of proposals for unlimited scalability & unified liquidity.

Polygon | POL

72,469 次观看 • 3 年前

Some personal news: I recently joined Cursor. Cursor is a small, ambitious team, and they’ve created my favorite AI systems. We’re now building frontier RL models at scale in real-world coding environments. Excited for how good coding is going to be.

Some personal news: I recently joined Cursor. Cursor is a small, ambitious team, and they’ve created my favorite AI systems. We’re now building frontier RL models at scale in real-world coding environments. Excited for how good coding is going to be.

Sasha Rush

335,873 次观看 • 1 年前

Pay for what you create, not for seats. FLORA's introducing credit-based pricing: make your ROI measurable at scale.

Pay for what you create, not for seats. FLORA's introducing credit-based pricing: make your ROI measurable at scale.

FLORA ©

22,998 次观看 • 4 个月前

Video diffusion models are just overqualified depth estimators! Deterministic single-pass depth estimation based on WanV2.1. - SOTA 5.5 AbsRel on ScanNet - data-efficient than baselines; - no temporal flicker + infinite-length estimation w/ zero scale drift.

Video diffusion models are just overqualified depth estimators! Deterministic single-pass depth estimation based on WanV2.1. - SOTA 5.5 AbsRel on ScanNet - data-efficient than baselines; - no temporal flicker + infinite-length estimation w/ zero scale drift.

Wildminder

49,235 次观看 • 3 个月前

How does high-fidelity tactile simulation help robots nail the last millimeter? We’re releasing VT-Refine, accepted to CoRL: a real-to-sim-to-real visuo-tactile policy using a GPU-parallel tactile sim for our piezoresistive skin FlexiTac. Then fine-tuning a diffusion policy with large-scale RL in simulation. Website: #CoRL2025 #RobotLearning #Sim2Real

How does high-fidelity tactile simulation help robots nail the last millimeter? We’re releasing VT-Refine, accepted to CoRL: a real-to-sim-to-real visuo-tactile policy using a GPU-parallel tactile sim for our piezoresistive skin FlexiTac. Then fine-tuning a diffusion policy with large-scale RL in simulation. Website: #CoRL2025 #RobotLearning #Sim2Real

Binghao Huang

46,967 次观看 • 7 个月前