2021年2月8日 · Quantization enables efficient acceleration of deep neural networks by reducing model memory footprint and exploiting low-cost integer math hardware units. Quantization maps floating-point weights and activations in a trained model to …

Quantized inference accelerates compute-bound operations, conserve memory bandwidth for memory-bound operations, and reduce on-chip storage size. VS-Quant extensions highlighted in gray. With VS-Quant, we see energy overhead over corresponding per-channel scaled configuration due to additional multipliers and wider accumulation.

We propose fine-grained per-vector scaled quantization (VS-Quant) to mitigate quantization-related accuracy loss. In contrast to coarse-grained per-layer/per-output-channel scal-ing, VS-Quant employs a scale factor for each vector of elements (V 1 1) in the activation and/or weight tensor as shown in Figure 1.

Quantization enables efficient acceleration of deep neural networks by reducing model memory footprint and exploiting low-cost integer math hardware units. Quantization maps floating-point weights and activations in a trained model to low-bitwidth integer values using scale factors.

2022年1月28日 · The signal-to-noise ratio (SNR) often drives the choice of distributed acoustic sensing (DAS) parameters in vertical seismic profiling (VSP). We compare this established approach for data quality assessment with metrics comparing DAS data products to …

LLMQuant是由一群来自世界顶尖高校和量化金融从业人员组成的前沿社区，致力于探索人工智能（AI）与量化（Quant）领域的无限可能。 我们的团队成员来自剑桥大学、牛津大学、哈佛大学、苏黎世联邦理工学院、北京大学、中科大等世界知名高校，外部顾问来自Microsoft、HSBC、Citadel、Man Group、Citi、Jump Trading、国内顶尖私募等一流企业。

2023年3月31日 · To associate your repository with the quantv-for-sp topic, visit your repo's landing page and select "manage topics." GitHub is where people build software. More than 150 million people use GitHub to discover, fork, and contribute to over 420 million projects.

准确提取Q值是研究地层吸收衰减特性的关键.对三种Q值反演方法（质心频率法、振幅衰减法、频谱比法）进行对比分析，旨在为衰减特性的求取提供参考.针对零偏VSP数据进行计算，对比总结薄层、频带宽度及低衰减层、波场成分、界面干扰等条件下Q值反演与层位揭示的准确性及差异性：对于相对较薄的层位，质心频率法几乎能准确揭示所有地层，而其他两种方法在薄层分界面处出现异常，误差超过200%；高频成分对频率域方法影响较大，尤其是对于低衰减层的反演；只 …

2024年11月19日 · We demonstrate that direct control over the VSP can significantly improve efficiency and present an unbiased volume rendering algorithm based on an existing resampling framework for precise control over the VSP.

2021年4月5日 · Quantization enables efficient acceleration of deep neural networks by reducing model memory footprint and exploiting low-cost integer math hardware units. Quantization maps floating-point weights and activations in a trained model to …