AI Insight
This paper introduces the Trajectory-based Quantization Sensitivity Score (TQS), a new metric for optimizing neural network compression that treats the network as a dynamical system to predict how quantization errors accumulate over time. The approach separates sensitivity analysis from the actual quantization process, enabling mixed-precision optimization without requiring calibration data or expensive computational approximations. The authors demonstrate that this dynamical-systems framework provides an effective method for deploying compressed models in resource-limited environments.
Why it matters
This work could improve the deployment of AI models on edge devices and mobile platforms by making it easier to compress neural networks while maintaining performance. The ability to analyze quantization sensitivity without calibration data or access to internal network structure makes the approach practical for real-world applications where computational resources and data availability are limited.
arXiv:2606.13300v2 Announce Type: replace
Abstract: We introduce the Trajectory-based Quantization Sensitivity Score (TQS), a metric that reframes post-training quantization (PTQ) through the lens of dynamical-systems stability. By modeling the network’s rollout as a discrete-time dynamical system, TQS characterizes how quantization-induced errors propagate and amplify over the rollout horizon. Unlike conventional PTQ methods, where sensitivity analysis is often coupled to the quantization procedure, TQS enables a priori sensitivity estimation decoupled from quantizer selection and bit-width assignment. This separation allows for quantization budget planning even for black-box or compiled networks with fused operators. Building on this, we present TQS-PTQ, a flexible mixed-precision framework that requires no calibration data or costly second-order approximations. Our experiments show that a dynamical-systems perspective provides a robust, high-performing pathway for low-precision deployment in resource-constrained settings.
Source: Quantizing Time-Series Models As Dynamical Systems: Trajectory-Based Quantization Sensitivity Score