John M. Fossaceca

Derrik E. Asher, Anjon Basak, Rolando Fernandez et al.

Reinforcement learning (RL) approaches can illuminate emergent behaviors that facilitate coordination across teams of agents as part of a multi-agent system (MAS), which can provide windows of opportunity in various military tasks. Technologically advancing adversaries pose substantial risks to a friendly nation's interests and resources. Superior resources alone are not enough to defeat adversaries in modern complex environments because adversaries create standoff in multiple domains against predictable military doctrine-based maneuvers. Therefore, as part of a defense strategy, friendly forces must use strategic maneuvers and disruption to gain superiority in complex multi-faceted domains such as multi-domain operations (MDO). One promising avenue for implementing strategic maneuver and disruption to gain superiority over adversaries is through coordination of MAS in future military operations. In this paper, we present overviews of prominent works in the RL domain with their strengths and weaknesses for overcoming the challenges associated with performing autonomous strategic maneuver and disruption in military contexts.

Guilin Zhang, Chuanyi Sun, Shahryar Sarkani et al.

Cloud LLM fine-tuning platforms increasingly serve RLHF workloads, where a learned reward model is optimized as a proxy for human quality. As Gao et al. (2023) showed, this proxy diverges from world feedback (downstream eval metrics) under sustained optimization pressure, a phenomenon known as reward overoptimization. Existing platform schedulers ignore this divergence: non-clairvoyant schedulers optimize JCT without any quality signal, SLAQ-style quality-aware schedulers use training loss (a weaker proxy that drops monotonically through hacking), and classical per-job early stopping requires human monitoring and does not free shared GPUs. We propose EvalStop, a composable scheduling primitive that terminates jobs on k consecutive eval-score declines, releases GPUs, preserves the best checkpoint, and delegates to any base scheduler. We frame scheduler-level early stopping as a detection problem and evaluate it in a discrete-event simulator whose RLHF workload mixes reward-hacking and structurally healthy runs, with ground-truth labels hidden from schedulers. On RLHF-heavy workloads (80% RLHF, 64 GPUs), EvalStop achieves precision 98% / recall 99% / FPR 1.5% while improving JCT by 9% and cutting wasted compute by 22% over SRTF-Est (p<0.05). Trivial fixed-progress and loss-plateau competitors either incur 65% FPR on healthy RLHF or miss over half of true hacking cases. Gains compose across every base scheduler tested (9-25% JCT) and detection quality stays stable under eval noise (precision at least 91% at noise std <= 0.05) and hacking base rate (precision at least 89% across 20-80% hacking fractions).