Imad Aouali

In summary: in large-action offline RL, trainability > perfect value estimation. Check out the full paper for our theoretical proofs, deep dives into the optimization landscapes, and extensive ablations. See you at #ICML2026! 🚀

We backed these claims up on two fronts: 📐 Theoretically: We prove the favorable optimization properties of PWLL. 📊 Empirically: We tested across large recommendation datasets (scaling up to 1M actions, like GoodReads-1M).

🎉 Excited to share our (with @OtmaneSakhi) #ICML2026 paper! "Off-Policy Learning in Large Action Spaces: Optimization Matters More Than Estimation" We highlight that in offline RL with large action spaces, the objective's trainability > its estimation accuracy. 🧵👇

My solo paper, Diffusion Models Meet Contextual Bandits, will appear at NeurIPS 2025. #NeurIPS2025 📄 imadaouali.github.io/papers/2025_di…

If you’re at #AISTATS2025, come check out my poster! 📝 Bayesian Off-Policy Evaluation and Learning for Large Action Spaces 📍 Hall A-E 58 ⏰ Poster Session 1 Joint work with Victor-Emmanuel Brunel, David Rohde, @Korba_Anna @aistats_conf

Mark your calendars ! 🏔️Machine Learning Autumn School at Aussois (France). 🏔️ 16-21 nov 2025. Submission deadline to present your work via talk or poster: 28th of April. mlataussois.sciencesconf.org Great line-up of speakers ! ⬇️

Last year, at @NeurIPSConf our team reached a new milestone: 15 papers accepted (13 papers - one spotlight, 1 dataset paper and 1 challenge paper) 🎉 Check all our contributions in this exhaustive recap from #NeurIPS2024 👇 techblog.criteo.com/all-our-contri…

I'm happy to share that "Bringing NeRFs to the Latent Space: Inverse Graphics Autoencoder" was accepted at #ICLR2025! 🎉 Another great paper from the @CriteoAILab - pushing 3D representations forward! 🚀 Huge congrats to Antoine Schnepf & Karim Kassab!

2 papers accepted at AISTATS #AISTATS2025 🥳 📜 Bayesian Off-Policy Evaluation and Learning for Large Action Spaces (@Korba_Anna, David & Victor) 📜 Prior-Dependent Allocations for Bayesian Fixed-Budget Best-Arm Identification in Structured Bandits (Nicolas, @vernadec & András)

Come visit our poster at #NeurIPS2024 to dive into policy learning, whether you're into theory or practice, we've got you covered 😉. Can't wait to chat!

@OtmaneSakhi Paper: neurips.cc/virtual/2024/p… Talk: youtube.com/watch?v=GF2EoO… or see @OtmaneSakhi and @imad_aouali in person at #NeurIPS2024

@imad_aouali youtube.com/watch?v=8Od8i_…

@imad_aouali - Scaleable Off-Policy Learning - Impossible without side information? #NeurIPS2024

@OtmaneSakhi and @imad_aouali - Is Principled Offline A/B Testing Achievable? youtube.com/watch?v=GF2EoO… #NeurIPS2024

Thursday @OtmaneSakhi and @imad_aouali Logarithmic Smoothing for Pessimistic Off-Policy Evaluation, Selection and Learning #neurips2024. Register here: criteo.zoom.us/webinar/regist…

@CONSEQUENCES_ws @ACMRecSys Monday! “Welfare-Optimized Recommender Systems” - SURE (@BenjaminHe12201 @BayesianIn @flavianv) "Logarithmic Smoothing for Pessimistic Off-Policy Evaluation, Selection and Learning" - Consequences Workshop @OtmaneSakhi @imad_aouali

The paper can be found in arxiv.org/pdf/2405.14335 and will be updated soon, I'll make a thread explaining the main ideas, stay tuned!

Indeed, our paper "Logarithmic Smoothing for Pessimistic Off-Policy Evaluation, Selection and Learning" has been accepted to @NeurIPSConf 2024 as a spotlight! Always a pleasure to collaborate with @imad_aouali , @PierreAlquier and Nicolas Chopin! Congrats to my co-authors 🎉