reinforcement learning

Reduced model-based control in gambling disorder despite seemingly intact neural value and task structure representations

Abstract Background Disordered gambling has been linked to impairments in goal-directed (model-based) control and reinforcement learning. Methods Here we investigated the potential neural basis of this impairment using a sequential reinforcement learning task (modified two-step-task), computational modeling, and functi...

Developing Strategic Communication Insights for Social Change through Dialogic Learning

Scholars, researchers, and practitioners are aware of the urgency to develop boundary-spanning concepts in strategic communication to create dialogues for social change. While literature is abundant about conducting research, collecting data, creating strategies, and conceptualising big ideas, the complex non-linear pr...

从控制到动态平衡:超级智能对齐的博弈论与热力学重构

Current research on AI alignment mainly follows paths such as reinforcement learning from human feedback, scalable oversight, and interpretability. Its implicit premise is that there exists a centralized designer capable of defining clear objectives and exercising effective control over the system. However, when AI sys...

RLVacDiffSim: Reinforcement learning-driven simulation of vacancy diffusion

Code used to train mean-first-passage-time/time-estimator models and run deep-reinforcement-learning-accelerated atomistic simulations of vacancy diffusion.

A trust-aware edge-assisted reinforcement learning framework for secure and low-latency vehicular communication

Abstract Vehicular Ad Hoc Networks enable real-time intelligent transportation services but remain affected by dynamic mobility, unstable routing, congestion, malicious vehicles, and delayed emergency communication. This paper proposes a Trust-Aware Edge-Assisted Reinforcement Learning framework, termed TEARL-VANET, fo...

Comment on egusphere-2026-4021

Abstract. The Arctic Weather Satellite (AWS), launched by the European Space Agency (ESA) in August 2024, has enabled the first-ever global ice cloud remote sensing using 325 GHz terahertz channels. Owing to its short wavelength, the 325 GHz frequency band is highly sensitive to scattering by the three frozen hydromete...

Co-Designing TWT Scheduling and RF Energy Harvesting for Sustainable Heterogeneous Wi-Fi: A Deep Reinforcement Learning Approach

IEEE 802.11ax Target Wake Time (TWT) coordinates station sleep and wake schedules for power savings, while RF energy harvesting promises battery-free operation for IoT sensors. Though studied in isolation, the two are coupled; the same access point (AP) that schedules TWT service periods, controls the RF harvest of bat...

MODEL PEMBELAJARAN HOMESCHOOLING BLENDED LEARNING BERBASIS DIFERENSIASI KONTEN DAN PROSES

Homeschooling requires learning that is flexible and responsive to differences in learners’ readiness, interests, and needs. This study aimed to describe the implementation of content- and process-differentiated blended learning at Homeschooling Candrawinata Bandung, including its planning, implementation, evaluation,...

Review of: "Solving Sokoban using Hierarchical Reinforcement Learning with Landmarks"

NeatRL: Readable Single-File Reinforcement Learning Implementations in PyTorch

NeatRL is a reinforcement learning library built around single-file, self-contained implementations of classic and modern deep RL algorithms: DQN, A2C, PPO, DDPG, TD3, SAC and more, with Gymnasium and Atari support and Weights & Biases experiment tracking. NeatRL source code is licensed under the MIT License.

Calibrated Decisions Are Not Calibrated Probabilities: An Exact-Target Audit of Jev and Three Open Decision Models

Decision models answer typed questions with probabilities instead of text, and their main selling point is that those probabilities can be trusted. TypeSafe says its Jev model, trained with an unpublished method called Reinforcement Learning for Calibrated Decisions (RLCD), returns "epistemically honest" probabilities....

Enhancing large language model navigation via two-stage fine-tuning and sparse-reward reinforcement learning