Formal Verification of Probabilistic Deep Reinforcement Learning Policies with(…)

on YouTube ↗33m

" summary

Describes an abstract-training method for formally verifying probabilistic deep reinforcement learning policies by discretizing the continuous state space into finite decision units, training a DNN on those units, and treating the trained policy as a Markov decision process (MDP). Uses black-box computation of probabilistic decisions for sets of states, reuses abstract states to reduce state explosion, and applies probabilistic model checking to compute upper bounds on unsafe-event probabilities, claiming tighter bounds and better scalability than prior work.

AI-written summary · about

" related talks

• 55mTowards Verified Deep LearningSanjit A. Seshia · Emerging Challenges in Deep Learning (Simons Institute workshop, 2019)
• 01h 01mA Tutorial on Reinforcement Learning IEmma Brunskill · Foundations of Machine Learning Boot Camp 2017
• 42mReinforcement Learning using Generative Models for Continuous State and Action Space SystemsRahul Jain · Reinforcement Learning from Batch Data and Simulation
• 17mInput-Relational Verification of Deep Neural NetworksGagandeep Singh, Debangshu Banerjee, Changming Xu · PLDI 2024
• 01h 17mTraining intelligent game agents using deep reinforcement learningImran Rashid · PyData London 2019
• 21mUsing Deep Reinforcement Learning for Dialogue SystemsHarm van Seijen · MLconf SF 2016
• 01h 05mFoundations of RLRayadurgam Srikant · Data-Driven Decision Processes Boot Camp 2022
• 32mStructured Deep Learning with Probabilistic Neural ProgJayant Krishnamurthy · Scale By The Bay 2018