Reward-Aware Population Scaling of Evolutionary Strategies in LLM Fine-Tuning

• URL: http://arxiv.org/abs/2607.19408v1
• Date: Wed, 08 Jul 2026 00:19:03 GMT
• Title: Reward-Aware Population Scaling of Evolutionary Strategies in LLM Fine-Tuning
• Abstract summary: Evolutionary Strategies for fine-tuning large language models is attractive because it is memory-efficient, parallel, and compatible with black-box or discrete rewards.<n>Yet its population-size conclusions conflict sharply: fine-tuning with cross-entropy reward succeeds with $N=1$, while binary-reward training often needs $N approx 30$.<n>We show this gap is largely about reward design and normalization, not population size.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Using Evolutionary Strategies (ES) for fine-tuning large language models is attractive because it is memory-efficient, parallel, and compatible with black-box or discrete rewards. Yet its population-size conclusions conflict sharply: fine-tuning with cross-entropy (CE) reward succeeds with $N=1$, while binary-reward training often needs $N \approx 30$. We show this gap is largely about reward design and normalization, not population size. In the capable-model regime we study, z-score advantage normalization can cause $N=2$ to fail. Disabling normalization lets binary-reward ES with $N=2$ improve on GSM8K and TREC across capable models spanning 0.5B-7B, where the normalized variant collapses or degrades. This small-$N$ risk is set by reward granularity: binary accuracy reward induces a zero-advantage probability $q$ that depends in closed form on base accuracy, batch size, and intra-pair correctness correlation; a zero-training probe on Qwen2.5-Instruct/GSM8K matches the formula with mean absolute error 0.020 across 12 configurations and finds the availability threshold $N_{\mathrm{avail}}$ to be small in this capable-model regime. The implication is not that $N=2$ is universally sufficient, but that small-population failure in capable-model binary ES can be an implementation artifact rather than an intrinsic population limit.

Related papers

• Does 1/2-Tsallis-INF Also Work Well for Best-Arm Identification? [11.223878908981725]
  $1/2$-Tsallis-INF is a canonical best-of-both-worlds FTRL algorithm.<n>It achieves logarithmic pseudo-regret in bandits while retaining minimax-optimal regret in adversarial bandits.<n>This raises a natural question: can the same algorithm, without additional exploration, also identify the best arm reliably?
  arXiv  Detail & Related papers  (2026-08-15T18:38:36Z)
• Representation Without Reward: A JEPA Audit for LLM Fine-Tuning [1.2691047660244335]
  Joint-embedding predictive architectures (JEPAs) propose that a model should learn more useful abstractions when trained to predict latent representations rather than observed outputs.<n>For autoregressive language-model fine-tuning the principle entails a stricter requirement: the induced hidden-state geometry must reach the language-model head emphand improve the decoded task metric.<n>We test that requirement under a fixed Llama-3.2-1B-Instruct LoRA harness on natural-language-to-regex generation.
  arXiv  Detail & Related papers  (2026-05-14T20:27:32Z)
• Binary Rewards and Reinforcement Learning: Fundamental Challenges [7.106986689736826]
  Reinforcement learning with rewards RLVR has become a standard approach for improving reasoning in language models.<n>We provide a structural account of this phenomenon grounded in the properties of binary rewards.
  arXiv  Detail & Related papers  (2026-05-04T09:17:26Z)
• Bayesian Modeling of Collatz Stopping Times: A Probabilistic Machine Learning Perspective [0.0]
  Empirically, $(n)$ is a skewed and heavily overdispersed count with pronounced arithmetic heterogeneity.<n>We develop two complementary models.<n>A mechanisticgenerative approximation based on the odd-block decomposition yields a calibrated approximation via a Dirichlet-multinomial update.<n>On held-out data, the NB2-GLM achieves substantially higher predictive likelihood than the odd-block generators.
  arXiv  Detail & Related papers  (2026-03-04T17:16:49Z)
• Unregularized Linear Convergence in Zero-Sum Game from Preference Feedback [50.89125374999765]
  We provide the first convergence guarantee for Optimistic Multiplicative Weights Update ($mathtOMWU$) in NLHF.<n>Our analysis identifies a novel marginal convergence behavior, where the probability of rarely played actions grows exponentially from exponentially small values.
  arXiv  Detail & Related papers  (2025-12-31T12:08:29Z)
• Evolution Strategies at the Hyperscale [57.75314521465674]
  We introduce EGGROLL, an evolution strategies (ES) algorithm designed to scale backprop-free optimization to large population sizes.<n>ES is a set of powerful blackbox optimisation methods that can handle non-differentiable or noisy objectives.<n>EGGROLL overcomes these bottlenecks by generating random matrices $Ain mathbbRmtimes r, Bin mathbbRntimes r$ with $rll min(m,n)$ to form a low-rank matrix perturbation $A Btop$
  arXiv  Detail & Related papers  (2025-11-20T18:56:05Z)
• Computational-Statistical Tradeoffs at the Next-Token Prediction Barrier: Autoregressive and Imitation Learning under Misspecification [50.717692060500696]
  Next-token prediction with the logarithmic loss is a cornerstone of autoregressive sequence modeling.<n>Next-token prediction can be made robust so as to achieve $C=tilde O(H)$, representing moderate error amplification.<n>No computationally efficient algorithm can achieve sub-polynomial approximation factor $C=e(log H)1-Omega(1)$.
  arXiv  Detail & Related papers  (2025-02-18T02:52:00Z)
• Stable Minima Cannot Overfit in Univariate ReLU Networks: Generalization by Large Step Sizes [29.466981306355066]
  We show that gradient descent with a fixed learning rate $eta$ can only find local minima that represent smooth functions. We also prove a nearly-optimal MSE bound of $widetildeO(n-4/5)$ within the strict interior of the support of the $n$ data points.
  arXiv  Detail & Related papers  (2024-06-10T22:57:27Z)
• Learning with Norm Constrained, Over-parameterized, Two-layer Neural Networks [54.177130905659155]
  Recent studies show that a reproducing kernel Hilbert space (RKHS) is not a suitable space to model functions by neural networks. In this paper, we study a suitable function space for over- parameterized two-layer neural networks with bounded norms.
  arXiv  Detail & Related papers  (2024-04-29T15:04:07Z)
• Feature Learning in $L_{2}$-regularized DNNs: Attraction/Repulsion and Sparsity [9.077741848403791]
  We show that the loss in terms of the parameters can be reformulated into a loss in terms of the layerwise activations $Z_ell$ of the training set. This reformulation reveals the dynamics behind feature learning.
  arXiv  Detail & Related papers  (2022-05-31T14:10:15Z)
• Minimax Optimal Quantization of Linear Models: Information-Theoretic Limits and Efficient Algorithms [59.724977092582535]
  We consider the problem of quantizing a linear model learned from measurements. We derive an information-theoretic lower bound for the minimax risk under this setting. We show that our method and upper-bounds can be extended for two-layer ReLU neural networks.
  arXiv  Detail & Related papers  (2022-02-23T02:39:04Z)
• Curse of Dimensionality on Randomized Smoothing for Certifiable Robustness [151.67113334248464]
  We show that extending the smoothing technique to defend against other attack models can be challenging. We present experimental results on CIFAR to validate our theory.
  arXiv  Detail & Related papers  (2020-02-08T22:02:14Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.