RL-GRIT: Reinforcement Learning for Grammar Inference

Woods, Walt

doi:10.1109/spw53761.2021.00031

Cited by 3 publications

(1 citation statement)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Drawing inspiration from prior work on RL for grammatical inference (RL-GRIT), 26 we leveraged a pairwise policy update approach. This has a number of potential advantages: (1) as with appendix A.2, loss scales according to changes in expected reward, (2) it works the same for any kind of action space, and (3) as discussed in appendix A.6, it lends itself very well to an alternative of MCTS that is faster to compute while conferring similar benefits.…”

Section: A4 Complex Action Spaces Via a Pairwise Policy Updatementioning

confidence: 99%

Combining AI control systems and human decision support via robustness and criticality

Woods,

Grushin,

Khan

et al. 2024

Disruptive Technologies in Information Sciences VIII

Self Cite

View full text Add to dashboard Cite

AI-enabled capabilities are reaching the requisite level of maturity to be deployed in the real world. Yet, the ability of these systems to always make correct or safe decisions is a constant source of criticism and reluctance to use them. One way of addressing these concerns is to leverage AI control systems alongside and in support of human decisions, relying on the AI control system in safe situations while calling on a human co-decider for critical situations. Additionally, by leveraging an AI control system built specifically to assist in joint human/machine decisions, the opportunity naturally arises to then use human interactions to continuously improve the AI control system's accuracy and robustness.We extend a methodology for adversarial explanations (AE) to state-of-the-art reinforcement learning frameworks, including MuZero. Multiple improvements to the base agent architecture are proposed. We demonstrate how this technology has two applications: for intelligent decision tools and to enhance training / learning frameworks. In a decision support context, adversarial explanations help a user make the correct decision by highlighting those contextual factors that would need to change for a different AI-recommended decision. As another benefit of adversarial explanations, we show that the learned AI control system demonstrates robustness against adversarial tampering. Additionally, we supplement AE by introducing strategically similar autoencoders (SSAs) to help users identify and understand all salient factors being considered by the AI system. In a training / learning framework, this technology can improve both the AI's decisions and explanations through human interaction. Finally, to identify when AI decisions would most benefit from human oversight, we tie this combined system to our prior art on statistically verified analyses of the criticality of decisions at any point in time.

show abstract

Section: A4 Complex Action Spaces Via a Pairwise Policy Updatementioning

confidence: 99%