Optimal Resource Allocation for ML Model Training and Deployment under Concept Drift
Beytur, Vikalo, Chan, de Veciana · cs.LG,cs.NI · 2026-07-31 · 原文 · 证据态 未学
We study how to allocate resources for training and deployment of machine learning (ML) models under concept drift and limited budgets. We consider a setting in which a model provider distributes trained models to multiple clients whose devices support local inference but lack the ability to retrain those models, placing the burden of performance maintenance on the provider. We introduce a model-agnostic framework that captures the interaction between resource allocation, concept drift dynamics, and deployment timing. We show that optimal training policies depend critically on the aging properties of concept durations. Under sudden concept changes, we derive optimal training policies subject to budget constraints when concept durations follow distributions with Decreasing Mean Residual Life (DMRL), and show that intuitive heuristics are provably suboptimal under Increasing Mean Residual Life (IMRL). We further study model deployment under communication constraints, prove that the associated optimization problem is quasi-convex under mild conditions, and propose a randomized scheduling strategy that achieves near-optimal client-side performance. These results offer theoretical and a
讲义
讲义·推断 依据「原文」自动生成的结构化摘要（推断），非原文表述；以原文为准。
1. 人话版
We study how to allocate resources for training and deployment of machine learning (ML) models under concept drift and limited budgets.
We consider a setting in which a model provider distributes trained models to multiple clients whose devices support local inference but lack the ability to retrain those models, placing the burden of performance maintenance on the provider.
2. 领域脉络
本文类目：cs.LG、cs.NI，属于其所在研究脉络的最新进展。
3. 机制拆解
We introduce a model-agnostic framework that captures the interaction between resource allocation, concept drift dynamics, and deployment timing.
We show that optimal training policies depend critically on the aging properties of concept durations.
4. 证据与数字
摘要未给出量化结果——留意原文的实验与数据。
5. 反例与边界
Under sudden concept changes, we derive optimal training policies subject to budget constraints when concept durations follow distributions with Decreasing Mean Residual Life (DMRL), and show that intuitive heuristics are provably suboptimal under Increasing Mean Residual Life (IMRL).
6. 跨领域连接与意外收获
横跨 2 个类目（cs.LG、cs.NI），关注其在你兴趣板块间的迁移面。
7. 可复用方法
把本文机制与你手头项目对照，找一个两周内能验证的最小实验。
8. 术语表
精读时把不熟的术语记入此处，作为下次回忆的锚点。
主动回忆
先合上内容自己复述，再点「显示」核对，然后如实自评。评分即时进 FSRS 排程。
讲义·推断 依据「原文」自动生成的结构化摘要（推断），非原文表述；以原文为准。
1. 人话版
We study how to allocate resources for training and deployment of machine learning (ML) models under concept drift and limited budgets.
We consider a setting in which a model provider distributes trained models to multiple clients whose devices support local inference but lack the ability to retrain those models, placing the burden of performance maintenance on the provider.
2. 领域脉络
本文类目：cs.LG、cs.NI，属于其所在研究脉络的最新进展。
3. 机制拆解
We introduce a model-agnostic framework that captures the interaction between resource allocation, concept drift dynamics, and deployment timing.
We show that optimal training policies depend critically on the aging properties of concept durations.
4. 证据与数字
摘要未给出量化结果——留意原文的实验与数据。
5. 反例与边界
Under sudden concept changes, we derive optimal training policies subject to budget constraints when concept durations follow distributions with Decreasing Mean Residual Life (DMRL), and show that intuitive heuristics are provably suboptimal under Increasing Mean Residual Life (IMRL).
6. 跨领域连接与意外收获
横跨 2 个类目（cs.LG、cs.NI），关注其在你兴趣板块间的迁移面。
7. 可复用方法
把本文机制与你手头项目对照，找一个两周内能验证的最小实验。
8. 术语表
精读时把不熟的术语记入此处，作为下次回忆的锚点。