Presentation
Mixed-Precision Quantization for Deep Vision Models with Integer Quadratic Programming
DescriptionQuantization is a widely used technique to compress neural networks. Assigning uniform bit-widths across all layers can result in significant accuracy degradation at low precision and inefficiency at high precision. Mixed-precision quantization (MPQ) addresses this by assigning varied bit-widths to layers, optimizing the accuracy-efficiency trade-off. Existing sensitivity-based methods for MPQ assume that quantization errors across layers are independent, which leads to suboptimal choices. We introduce CLADO, a practical sensitivity-based MPQ algorithm that captures cross-layer dependency of quantization error. CLADO approximates pairwise cross-layer errors using linear equations on a small data subset. Layerwise bit-widths are assigned by optimizing a new MPQ formulation based on cross-layer quantization errors using an Integer Quadratic Program. Experiments with CNN and transformer models on ImageNet demonstrate that CLADO achieves state-of-the-art mixed-precision quantization performance. Code is available at \href{https://anonymous.4open.science/r/CLADO_DAC2025-BC55/README.md}{CLADO anonymous GitHub repo.}
Event Type
Research Manuscript
TimeTuesday, June 2411:15am - 11:30am PDT
Location3001, Level 3
AI1: AI/ML Algorithms