Neural Information Processing Systems http://nips.cc/

reinforcement learning; delayed reward; reward redistribution; return decomposition; bias-variance; credit assignment; LSTM

Under/overestimation of state/action values are harmful for reinforcement learning agents.

We propose RUDDER, a novel reinforcement learning approach for delayed rewards in finite Markov decision processes (MDPs).

reinforcement learning; delayed reward; reward redistribution; return decomposition; bias-variance; credit assignment; LSTM

The objective is to maximize the total reward accumulated over all episodes (e.g., the number of games won), expressed as

Reward sparsity in long-horizon reinforcement learning (RL) tasks remains a significant challenge, while existing outcome-based reward shaping struggles to define meaningful immediate rewards without introducing bias or requiring explicit task decomposition. Alternatively, verification-based reward shaping uses stepwise critics, but misalignment between immediate rewards and long-term objectives can lead to reward hacking and suboptimal policies. In this work, we address this problem in the context of software engineering (SWE) tasks, where multi-turn reasoning and rule-based verification are critical. We introduce the SWE-oriented RL Framework, a unified system supporting multi-turn interaction, docker-based execution, and customizable reward functions. Additionally, we propose Gated Reward Accumulation (G-RA), a novel method that accumulates immediate rewards only when high-level (long-term) rewards meet a predefined threshold, ensuring stable RL optimization. Experiments on SWE-bench V eri-fied and kBench demonstrate that G-RA leads to an increase in completion rates (47.6% 93.8% and 22.0% 86.0%) and modification rates (19.6% 23.8% and 12.0% 42.0%), while avoiding policy degradation caused by reward misalignment. Our findings highlight the importance of balanced reward accumulation in long-horizon RL and provide a practical solution.

We study a sequential decision-making problem motivated by recent regulatory and technological shifts that limit access to individual user data in recommender systems (RSs), leaving only population-level preference information. This privacy-aware setting poses fundamental challenges in planning under uncertainty: Effective personalization requires exploration to infer user preferences, yet unsatisfactory recommendations risk immediate user churn. To address this, we introduce the Rec-APC model, in which an anonymous user is drawn from a known prior over latent user types (e.g., personas or clusters), and the decision-maker sequentially selects items to recommend. Feedback is binary -- positive responses refine the posterior via Bayesian updates, while negative responses result in the termination of the session. We prove that optimal policies converge to pure exploitation in finite time and propose a branch-and-bound algorithm to efficiently compute them. Experiments on synthetic and MovieLens data confirm rapid convergence and demonstrate that our method outperforms the POMDP solver SARSOP, particularly when the number of user types is large or comparable to the number of content categories. Our results highlight the applicability of this approach and inspire new ways to improve decision-making under the constraints imposed by aggregated preference data.

While recommender systems (RSs) traditionally rely on extensive individual user data, regulatory and technological shifts necessitate reliance on aggregated user information. This shift significantly impacts the recommendation process, requiring RSs to engage in intensive exploration to identify user preferences. However, this approach risks user churn due to potentially unsatisfactory recommendations. In this paper, we propose a model that addresses the dual challenges of leveraging aggregated user information and mitigating churn risk. Our model assumes that the RS operates with a probabilistic prior over user types and aggregated satisfaction levels for various content types. We demonstrate that optimal policies naturally transition from exploration to exploitation in finite time, develop a branch-and-bound algorithm for computing these policies, and empirically validate its effectiveness.