Standard deep reinforcement learning (DRL) aims to maximize expected reward, considering collected experiences equally in formulating a policy. This differs from human decision-making, where gains and losses are valued differently and outlying outcomes are given increased consideration. It also fails to capitalize on opportunities to improve safety and/or performance through the incorporation of distributional context. Several approaches to distributional DRL have been investigated, with one popular strategy being to evaluate the projected distribution of returns for possible actions. We propose a more direct approach whereby risk-sensitive objectives, specified in terms of the cumulative distribution function (CDF) of the distribution of full-episode rewards, are optimized. This approach allows for outcomes to be weighed based on relative quality, can be used for both continuous and discrete action spaces, and may naturally be applied in both constrained and unconstrained settings. We show how to compute an asymptotically consistent estimate of the policy gradient for a broad class of risk-sensitive objectives via sampling, subsequently incorporating variance reduction and regularization measures to facilitate effective on-policy learning. We then demonstrate that the use of moderately "pessimistic" risk profiles, which emphasize scenarios where the agent performs poorly, leads to enhanced exploration and a continual focus on addressing deficiencies. We test the approach using different risk profiles in six OpenAI Safety Gym environments, comparing to state of the art on-policy methods. Without cost constraints, we find that pessimistic risk profiles can be used to reduce cost while improving total reward accumulation. With cost constraints, they are seen to provide higher positive rewards than risk-neutral approaches at the prescribed allowable cost.

This paper considers online classification problems where each object to be classified consists of a sequence of measurements, termed here a track. We present an approach that combines ideas from sequential hypothesis testing with those from conformal prediction to address track level outliers - entire measurement sequences that are novel relative to the statistical model. We show with analysis and empirical results that this approach preserves the optimal performance of the underlying sequential hypothesis testing when outliers are absent and provides an error rate guarantee in the presence of contamination by novel tracks.

This paper considers using online binary classification for target detection where the goal is to identify signals of interest within a sequence of received signals generated by a shifting background. In this setting, we assume there is significant class imbalance (100:1 or greater), the sequence of examples is arbitrarily long and the distribution of the majority (negative) class is slowly time-varying. This setting is typical in detection and classification problems in which time-varying effects are caused by some combination of shifting channel characteristics and interferers that enter and exit the scene. We show empirically that the addition of caching and minority class oversampling to online learners improves the g-means performance under these conditions by compensating for class imbalance.