In this work, we study the simple yet universally applicable case of reward shaping in value-based Deep Reinforcement Learning (DRL). We show that reward shifting in the form of a linear transformation is equivalent to changing the initialization of the $Q$-function in function approximation. Based on such an equivalence, we bring the key insight that a positive reward shifting leads to conservative exploitation, while a negative reward shifting leads to curiosity-driven exploration. Accordingly, conservative exploitation improves offline RL value estimation, and optimistic value estimation improves exploration for online RL. We validate our insight on a range of RL tasks and show its improvement over baselines: (1) In offline RL, the conservative exploitation leads to improved performance based on off-the-shelf algorithms; (2) In online continuous control, multiple value functions with different shifting constants can be used to tackle the exploration-exploitation dilemma for better sample efficiency; (3) In discrete control tasks, a negative reward shifting yields an improvement over the curiosity-based exploration method.

We consider a regression setting where observations are collected in different environments modeled by different data distributions. The field of out-of-distribution (OOD) generalization aims to design methods that generalize better to test environments whose distributions differ from those observed during training. One line of such works has proposed to minimize the maximum risk across environments, a principle that we refer to as MaxRM (Maximum Risk Minimization). In this work, we introduce variants of random forests based on the principle of MaxRM. We provide computationally efficient algorithms and prove statistical consistency for our primary method. Our proposed method can be used with each of the following three risks: the mean squared error, the negative reward (which relates to the explained variance), and the regret (which quantifies the excess risk relative to the best predictor). For MaxRM with regret as the risk, we prove a novel out-of-sample guarantee over unseen test distributions. Finally, we evaluate the proposed methods on both simulated and real-world data.

We prove a fundamental impossibility theorem: neural networks cannot simultaneously learn well-calibrated confidence estimates with meaningful diversity when trained using binary correct/incorrect supervision. Through rigorous mathematical analysis and comprehensive empirical evaluation spanning negative reward training, symmetric loss functions, and post-hoc calibration methods, we demonstrate this is an information-theoretic constraint, not a methodological failure. Our experiments reveal universal failure patterns: negative rewards produce extreme underconfidence (ECE greater than 0.8) while destroying confidence diversity (std less than 0.05), symmetric losses fail to escape binary signal averaging, and post-hoc methods achieve calibration (ECE less than 0.02) only by compressing the confidence distribution. We formalize this as an underspecified mapping problem where binary signals cannot distinguish between different confidence levels for correct predictions: a 60 percent confident correct answer receives identical supervision to a 90 percent confident one. Crucially, our real-world validation shows 100 percent failure rate for all training methods across MNIST, Fashion-MNIST, and CIFAR-10, while post-hoc calibration's 33 percent success rate paradoxically confirms our theorem by achieving calibration through transformation rather than learning. This impossibility directly explains neural network hallucinations and establishes why post-hoc calibration is mathematically necessary, not merely convenient. We propose novel supervision paradigms using ensemble disagreement and adaptive multi-agent learning that could overcome these fundamental limitations without requiring human confidence annotations.

We provide a visual overview of all eight environments considered in Figure 7. CartPole

This paper studies constrained optimal control, where the goal is to produce a policy that maximizes an objective function subject to a constraint. The authors provide great motivation for this setting, explaining why the constraint cannot simply be included as a large negative reward. They detail challenges in solving this problem, especially if the initial policy does not satisfy the constraint. They also note a clever extension of their method, where they use the constraint to define the objective, by setting the constraint to indicate whether the task is solved. Their algorithm builds upon CEM: at each iteration, if there are no feasible policies, they maximize the constraint function for the policies with the largest objective; otherwise, they maximize the objective function for feasible policies.

The Treacherous Turn refers to the scenario where an artificial intelligence (AI) agent subtly, and perhaps covertly, learns to perform a behavior that benefits itself but is deemed undesirable and potentially harmful to a human supervisor. During training, the agent learns to behave as expected by the human supervisor, but when deployed to perform its task, it performs an alternate behavior without the supervisor there to prevent it. Initial experiments applying DRL to an implementation of the A Link to the Past example do not produce the treacherous turn effect naturally, despite various modifications to the environment intended to produce it. However, in this work, we find the treacherous behavior to be reproducible in a DRL agent when using other trojan injection strategies. This approach deviates from the prototypical treacherous turn behavior since the behavior is explicitly trained into the agent, rather than occurring as an emergent consequence of environmental complexity or poor objective specification. Nonetheless, these experiments provide new insights into the challenges of producing agents capable of true treacherous turn behavior.

The increasing popularity of large language models has not only led to widespread use but has also brought various risks, including the potential for systematically spreading fake news. Consequently, the development of classification systems such as DetectGPT has become vital. These detectors are vulnerable to evasion techniques, as demonstrated in an experimental series: Systematic changes of the generative models' temperature proofed shallow learning-detectors to be the least reliable. Fine-tuning the generative model via reinforcement learning circumvented BERT-based-detectors. Finally, rephrasing led to a >90\% evasion of zero-shot-detectors like DetectGPT, although texts stayed highly similar to the original. A comparison with existing work highlights the better performance of the presented methods. Possible implications for society and further research are discussed.

In this work, we study the simple yet universally applicable case of reward shaping in value-based Deep Reinforcement Learning (DRL). We show that reward shifting in the form of a linear transformation is equivalent to changing the initialization of the Q -function in function approximation. Based on such an equivalence, we bring the key insight that a positive reward shifting leads to conservative exploitation, while a negative reward shifting leads to curiosity-driven exploration. Accordingly, conservative exploitation improves offline RL value estimation, and optimistic value estimation improves exploration for online RL. We validate our insight on a range of RL tasks and show its improvement over baselines: (1) In offline RL, the conservative exploitation leads to improved performance based on off-the-shelf algorithms; (2) In online continuous control, multiple value functions with different shifting constants can be used to tackle the exploration-exploitation dilemma for better sample efficiency; (3) In discrete control tasks, a negative reward shifting yields an improvement over the curiosity-based exploration method.

We introduce the ``soft Deep MaxPain'' (softDMP) algorithm, which integrates the optimization of long-term policy entropy into reward-punishment reinforcement learning objectives. Our motivation is to facilitate a smoother variation of operators utilized in the updating of action values beyond traditional ``max'' and ``min'' operators, where the goal is enhancing sample efficiency and robustness. We also address two unresolved issues from the previous Deep MaxPain method. Firstly, we investigate how the negated (``flipped'') pain-seeking sub-policy, derived from the punishment action value, collaborates with the ``min'' operator to effectively learn the punishment module and how softDMP's smooth learning operator provides insights into the ``flipping'' trick. Secondly, we tackle the challenge of data collection for learning the punishment module to mitigate inconsistencies arising from the involvement of the ``flipped'' sub-policy (pain-avoidance sub-policy) in the unified behavior policy. We empirically explore the first issue in two discrete Markov Decision Process (MDP) environments, elucidating the crucial advancements of the DMP approach and the necessity for soft treatments on the hard operators. For the second issue, we propose a probabilistic classifier based on the ratio of the pain-seeking sub-policy to the sum of the pain-seeking and goal-reaching sub-policies. This classifier assigns roll-outs to separate replay buffers for updating reward and punishment action-value functions, respectively. Our framework demonstrates superior performance in Turtlebot 3's maze navigation tasks under the ROS Gazebo simulation.