Guy Tennenholtz أُعيد تغريده
Our paper, "Spectral Bellman Method: Unifying Representation and Exploration in RL," has been accepted to ICLR 2026! 🚀
Paper: arxiv.org/abs/2507.13181
with @guytenn , Bo Dai and Shie Mannor
🧵 1/n
English
Guy Tennenholtz
129 posts
Guy Tennenholtz أُعيد تغريده
Ever had a perfect image in mind that a text-to-image model just couldn't capture? Our new reinforcement learning agent, PASTA, turns image generation into a collaborative conversation, learning your style to bring your vision to life. Learn how it works: goo.gle/4gRVEqA
English
We train an LLM to be "aware" of how it will be used during inference. We show we can do this efficiently in SFT and RL under a Best-of-N inference strategy. Our model explores more efficiently and accounts for errors in our scoring model. Check it out: arxiv.org/pdf/2412.15287
English
We're releasing a new dataset with sequential interactions for text to image generation with human feedback. More data will follow very soon.
Paper: arxiv.org/pdf/2412.10419
Dataset: kaggle.com/datasets/googl…
Ofir Nabati@ofirnabati
We're excited to share our new paper: "Personalized and Sequential Text-to-Image Generation"! Check out the paper and our new sequential human rater dataset! 👇 Paper: arxiv.org/pdf/2412.10419 Dataset: kaggle.com/datasets/googl… Details below.. 1/N 🧵
English
Guy Tennenholtz أُعيد تغريده
How can we address hidden confounding biases in #OfflineRL?
Presenting our latest work with fantastic co-authors @hugoyeche @bschoelkopf @gxr @guytenn at #ICML2023 workshops: LCDS (today, Fri), CMM (Sat) and oral at SCIS (Sat)!
📝: arxiv.org/pdf/2306.01157… (1/3)
English
We released a multi-agent RL framework for network congestion control with the first public realistic network simulator! github.com/NVlabs/RLCC. Based on the amazing work of @BenjaminFuhrer and @ChenTessler
English
Major kudos to @AlizeePace for leading this innovative work and to all collaborators (Hugo Yèche, @BSchoelkopf, and Gunnar Rätsch) for their valuable contributions. Your thoughts on our work are most welcome! [5/5] arxiv.org/pdf/2306.01157…
English
We construct a practical approach to estimate these three types of uncertainties, contributing to a more effective offline RL algorithm that performs well on synthetic and real-world medical data, improving sota offline RL performance. [4/5]
English
Very happy to share our new paper: "Delphic Offline Reinforcement Learning under Nonidentifiable Hidden Confounding" led by the brilliant @AlizeePace. Dive into a short research thread below! 🧵👇 [1/5] arxiv.org/pdf/2306.01157…
English
Our recent exploration into representation-driven #ReinforcementLearning provides some interesting findings. Definitely worth a read! arxiv.org/abs/2305.19922. Accepted to #ICML2023
Ofir Nabati@ofirnabati
🎉 Excited to share our latest work accepted at #ICML2023: "Representation-Driven Reinforcement Learning" 🚀. In collaboration with @guytenn and @MannorShie, we've developed a representation-driven framework for reinforcement learning. arxiv.org/abs/2305.19922. 🧵[Thread] [1/n]
English
Our approach emphasizes the vital role of exploration in mitigating the adverse effects of popularity bias. Rather than aiming to eradicate popularity bias entirely, we focus on alleviating its negative repercussions, paving the way for improved user welfare. 🧵[Thread] [4/5]
English
Excited to share our latest work on #RecommenderSystems 🚀. Joint work with @hastyboomalert, @NadavMerlis , and @CraigBoutilier, we tackle the 'rich get richer' phenomenon within recommender systems. arxiv.org/pdf/2305.18333… 🧵[Thread] [1/5]
English