February 2026

On-policy distillation (OPD), which aligns the student with the teacher’s logit distribution on student-generated trajectories, has demonstrated strong empirical gains in improving student performance and often outperforms off-policy distillation and reinforcement learning (RL) paradigms. In this work, we first theoretically show that OPD is a special case of dense KL-constrained RL where the reward function and the KL regularization are always weighted equally and the reference model can by any model. Then, we propose the Generalized On-Policy Distillation […]

Ver mais

Demonstration selection is a practical bottleneck in in-context learning (ICL): under a tight prompt budget, accuracy can change substantially depending on which few-shot examples are included, yet selection must remain cheap enough to run per query over large candidate pools. We propose Meta-Sel, a lightweight supervised meta-learning approach for intent classification that learns a fast, interpretable scoring function for (candidate, query) pairs from labeled training data. Meta-Sel constructs a meta-dataset by sampling pairs from the training split and […]

Ver mais

Hello! I am looking for research that has successfully applied AlphaZero/MuZero-style learning to sequential, perfect information, non-zero sum board games, e.g. Terra Mystica where the winning player is decided by a numerical score (associated with each player) at the end of the game, rather than the zero sum outcomes of games such as Chess, Shogi, Go, etc. I figure there must exist an approach that works for multi-agent (> 2 player) games. Any suggestions? Thank you submitted by […]

Ver mais

We study offline reinforcement learning under $Q^star$-approximation and partial coverage, a setting that motivates practical algorithms such as Conservative $Q$-Learning (CQL; Kumar et al., 2020) but has received limited theoretical attention. Our work is inspired by the following open question: "Are $Q^star$-realizability and Bellman completeness sufficient for sample-efficient offline RL under partial coverage?" We answer in the negative by establishing an information-theoretic lower bound. Going substantially beyond this, we introduce a general framework that characterizes the intrinsic complexity […]

Ver mais

Estimating the state of an environment from high-dimensional, multimodal, and noisy observations is a fundamental challenge in reinforcement learning (RL). Traditional approaches rely on probabilistic models to account for the uncertainty, but often require explicit noise assumptions, in turn limiting generalization. In this work, we contribute a novel method to learn a structured latent representation, in which distances between states directly correlate with the minimum number of actions required to transition between them. The proposed metric space formulation […]

Ver mais

Large language models (LLMs) have demonstrated strong code generation capabilities, yet the runtime performance of generated code is not guaranteed, and there have been few attempts to train LLMs using runtime performance as a reward in the HPC domain. We propose an online reinforcement learning approach that executes LLM-generated code on a supercomputer and directly feeds back the measured runtime performance (GFLOPS) as a reward. We further introduce a Staged Quality-Diversity (SQD) algorithm that progressively varies the permitted […]

Ver mais

eBPF is increasingly used around databases, but prior systems and papers often conflate goals and therefore blur the trade space: instrumentation versus control versus in-kernel state. This impedes principled comparisons and hides the kernel constraints that determine feasibility. We present a unified analysis organized into three modes of database–kernel integration via eBPF: observability (deriving DB-relevant signals from kernel and user-space hooks), policy injection (installing workload-specific cache and networking policies at kernel choke points), and kernel-resident state (providing safe […]

Ver mais