We theoretically establish that our algorithms almost surely converge to locally optimal policies of our safe optimization framework.

Many practical settings call for the reconstruction of temporal signals from corrupted or missing data. Classic examples include decoding, tracking, signal enhancement and denoising.

We name this problem Safe-RL-SW . Our step-wise violation constraint differs from prior expected violation constraint (Wachi & Sui, 2020; Efroni et al., 2020b; Kalagarla et al., 2021) in two aspects: (i) Minimizing the step-wise violation enables the agent to learn an optimal policy that avoids unsafe regions deterministically,

We name this problem Safe-RL-SW . Our step-wise violation constraint differs from prior expected violation constraint (Wachi & Sui, 2020; Efroni et al., 2020b; Kalagarla et al., 2021) in two aspects: (i) Minimizing the step-wise violation enables the agent to learn an optimal policy that avoids unsafe regions deterministically,

In many applications, learning systems are required to process continuous non-stationary data streams.

The setting described is summarized in Box A.