CORTEXA
← Browse
arxivcs.LGstat.ML2026-06-29

Predictable GRPO: A Closed-Form Model of Training Dynamics

Rajat Ghosh, Datta Nimmaturi, Aryan Singhal, Vaishnavi Bhargava, Henry Wong, Johnu George, Debojyoti Dutta

We develop a first-principles reduced-order model of these dynamics. Under a single mean-field assumption that summarizes the policy by its expected reward, we reduce the GRPO update to a stochastically-forced damped oscillator whose mass, damping, and stiffness are fixed in closed form by the optimizer hyperparameters together with a single measured curvature scale -- momentum supplies the inertia, off-policy lag erodes the damping, and the group size enters, to leading order, as a noise temperature. The reduction has three consequences. First, it subsumes the empirical single-exponential saturation law as its overdamped limit, recasting the fitted plateau, timescale, and size exponent as the fixed point, inverse stiffness, and curvature-scaling exponent of the underlying potential, and adding, through the retained inertial term, the slow-start phase the single exponential cannot represent. Second, it yields predictions tied to independently measurable quantities rather than fitted ones: group-size invariance of the deterministic trajectory with a $1/G$ stationary fluctuation, a sharp stability threshold in the refresh interval, and an overdamped-to-oscillatory transition. Third, it furnishes diagnostics that separate failure modes a reward curve alone conflates -- reward hacking, advantage degeneracy, policy concentration, and dynamical instability. Across three models and two group sizes, the closed-form trajectory fits training reward to $R^2 \geq 0.91$ and the mean trajectory is group-size invariant to leading order -- on both the reward curve and out-of-distribution transfer to eight math benchmarks -- while the within-group reward spread retains a residual $G$-dependence that the leading-order temperature picture does not capture.

View free PDFSource page

Related papers

arxivcs.LGcs.AIstat.ML2026-07-13

Are we Merging the Right Models? Impact of Expert Training Duration on Model Merging for LLMs

Nikita Kozodoi, Zainab Afolabi, Jack Butler

Multi-task model merging combines separately trained expert models into a single model that handles all tasks without co-training. Standard practice merges experts at their optimal validation loss. We challenge this convention by systematically studying how training duration of d…

View free PDFSource page
arxivcs.LGcs.IRstat.ML2026-07-15

A Temporal Machine Learning-Based Time-to-Event Model for Predicting ALS Progression and Healthcare Utilization

Zongliang Yue, Qi Li, Terry Heiman-Patterson, Frank Bearoff, Zhaohui Qin, Huanmei Wu

Amyotrophic lateral sclerosis (ALS) is a progressive and heterogeneous neurodegenerative disease in which predicting clinically meaningful milestones, such as assistive device use, remains challenging. We developed a time-to-event, digital-twin-inspired framework that integrates…

View free PDFSource page
arxivstat.MLcs.AIcs.LG2026-06-26

Spectral Perturbation of the Empirical Fisher Information Matrix under Weight Quantization

Rahid Zahid Alekberli, Hikmat Karimov

We study the spectral perturbation of the empirical Fisher Information Matrix (FIM) of a parametric statistical model under two structured perturbations: departure of the input from a reference (in-distribution) ensemble, and finite-precision (quantized) perturbation of the model…

View free PDFSource page
arxivcs.LGcs.AIstat.ML2026-06-29

What Drives the Inlier-Memorization Effect? A Theory of Outlier Detection via Early Training Dynamics

Kunwoong Kim, Dongha Kim

Outlier detection (OD) aims to identify anomalous instances by learning the underlying structure of normal data (inliers), and is particularly challenging in fully unsupervised settings where no information about anomalies is available during training. Recent advances have levera…

View free PDFSource page