ML-for-ML

2026-08-06

AI Summary

This work addresses the suboptimal end-to-end performance of AI training workloads in shared cloud clusters, which arises from the independent optimization of networking and machine learning system parameters. To overcome this limitation, the paper proposes a cross-layer joint optimization framework that, for the first time, co-models network resource scheduling and distributed machine learning training strategies within a unified time-to-target-loss objective. By jointly tuning parameters across these traditionally siloed layers, the approach breaks down conventional hierarchical design barriers. Experimental results demonstrate that the proposed method achieves up to a 42% improvement in training speed while reaching the same target loss, significantly enhancing overall training efficiency.

Abstract

AI training workloads are growing rapidly, making their time, energy, and infrastructure costs increasingly important. In shared cloud clusters, training and fine-tuning jobs compete with co-running workloads for network resources, while network mechanisms and ML training choices are typically optimized separately: networking controls how bytes move, whereas ML systems control when and how much communication occurs. We argue that this separation leaves end-to-end performance on the table. We present ML-for-ML, a cross-layer perspective in which network-side and ML-side knobs are selected jointly under a shared time-to-target-loss objective. Our preliminary prototype shows that by co-optimizing the ML and network parameters, we reach the target loss up to 42% faster.

Problem

• ML training
• network resource contention
• cross-layer optimization
• shared clusters
• time-to-target-loss

Innovation

• cross-layer optimization
• ML training acceleration
• network-ML co-design
• time-to-target-loss
• resource efficiency