Delighted to share our #neurips2023 paper w @grockious @hmd_palangi et al
Evaluating Cognitive Maps & Planning in LLMs with CogEval
We test planning in 8 LLMs.
Failures like hallucinating invalid paths/falling in loops don't support emergent planning.
1/n
https://t.co/x4AdQyzekw
Happy to share that "LCRL: Logically-Constrained Reinforcement Learning" by M Hasanbeig (@grockious), D Kroening, and A Abate has been accepted to #CONFEST#QEST2022.
Congrats to @HjalmarWijk for getting his paper on "Shielding Atari Games with Bounded Prescience" accepted at @aamas2021! Work done with @mircogiacobbe and @DiffKroening, and the preprint will be on arXiv soon!
Happy to share that our paper "DeepSynth: Automata Synthesis for Automatic Task Segmentation in Deep Reinforcement Learning" has been accepted to #AAAI2021
Congrats to Hosein Hasanbeig (@grockious), Natasha Jeppu, Alessandro Abate, Tom Melham, and Daniel Kroening (@DiffKroening)
It is 2020 and I still have to argue with reviewers that gridworlds are perfectly suitable for testing the limits of current RL SotA. Some photorealistic 3D environment might be many orders of magnitude easier to solve than complex procgen environments like MiniGrid or NetHack.
Happy to announce that "Cautious Reinforcement Learning with Logical Constraints" by Mohammadhosein Hasanbeig (@grockious), Alessandro Abate, and Daniel Kroening (@DiffKroening) has been accepted to #AAMAS2020.
https://t.co/ZbcHQJeRlh