🚨Can we steer a generative model towards a gold reward we can't evaluate at inference time? Introducing 🍸 MARTINI — Steered Not Stirred🍸 We combine multiple imperfect reward surrogates from offline values OR preferences to steer without querying the oracle. Huge shout out to our dream team, brilliantly led by @alec_sargood28, with stellar work from @olivgallup and Noam Ghenassia, plus key input from Georg Maierhofer and @pduckw. A true honour to co-supervise with @bose_joey. 📄 Paper: arxiv.org/abs/2610.25502 🌐 Blog: alecsargood.github.io/martin… 🧵1/10

Thrilled to see our NeurIPS 2024 paper, Sequential Monte Carlo Policy Optimisation (arxiv.org/abs/2402.07963), featured in Kevin's Reinforcement Learning: A Comprehensive Overview, which additionally recognises SMC as a competitive, scalable online planner. A fantastic modern resource on RL! @EdanToledo @DonalByrne2 @pduckw @AlexLaterre

Excited to be sharing our work on scalable model-based search at NeurIPS tomorrow. Come chat to us! Work led by @MattVMacfarlane, @EdanToledo, @DonalByrne2 and @AlexLaterre

Read latest #openaccess paper: Planning under uncertainty for #saferobotexploration using #Gaussianprocess prediction by @astephens_ @m_budd @pduckw @MauriceFallon @hawesie 🔓bit.ly/3MAWdXv @oxfordrobots @dynamicrobots

Fantastic achievement by @ClementBonnet16 and the whole Jumanji team!

It’s official! 🎉Today InstaDeep becomes a proud member of the BioNTech family. Want to apply your AI skills to make a difference? Join us, we’re hiring!

RT @ClementBonnet16: We released a new version of Jumanji (v0.3) with now 18 environments, all with networks and trained checkpoints 🚀 Chec…

I'm excited to announce that towards the end of 2022, I started a new role @instadeepai as a Research Scientist. Since then, we've been acquired by @BioNTech_Group. Exciting times ahead!

If you're at @corl_conf #CoRL2022 and interested in planning under uncertainty for robot missions, make sure to visit @m_budd's poster on Saturday morning! His work uses GPs as part of a Bayes-optimal framework for planning single missions in partially-known environments.

Friday was my final day at @oxfordrobots and @UniofOxford. I've had a blast. Many thanks to @hawesie and @maosbot for being sensational humans and PIs. On to the next chapter.

You can find me at sigmoid.social/@pduckworth also

How do you collect valuable data from sensors far under the ocean surface? Our #IROS2022 paper's planner reasons over uncertain localisation 📍, navigation 🧭 and communication 📶 for underwater autonomous vehicles. 1/9 bit.ly/3C3ufQi @GOALS_oxford @hawesie @pduckw

Without planning it, we were all in the same place at the same time today. Here's a picture of GOALS in 2022!

Pleasantly surprised that our paper received runner-up for the best paper award at @ICAPSConference! The paper is about mitigating risk in MDPs without sacrificing expected performance: arxiv.org/abs/2110.12746 Work with @pduckw, Bruno Lacerda, and @hawesie

The main deadline for DPhil/PhD applications to @oxengsci is in a month (3 Dec 2021). Make sure to have your application in for this date to be eligible for scholarships including for @oxfordrobots. ox.ac.uk/admissions/graduate…

On Solving a Stochastic Shortest-Path Markov Decision Process as Probabilistic Inference deepai.org/publication/on-so… by Mohamed Baioumy et al. including @pduckw #ArtificialIntelligence #MarkovDecisionProcess

Really excited to be part of the Astronaut Health Team at @FDL_AI 2021!! Stay tuned for exciting work on causal inference and out-of-distribution/domain transfer learning

Amazing team to be part of.

Our work on adaptive control and state-estimation recently got accepted at #ICRA2021! Paper: arxiv.org/abs/2005.05894 Repo and videos: github.com/MoBaioumy/active_… With @pduckw, Bruno Lacerda and @hawesie. This work was supported by @ORCA_Hub, @RAIN_hub and @oxfordrobots.