Don’t Fear the Bit Flips: Robust Linear Prediction Through Informed Channel Coding
Don’t Fear the Bit Flips: Robust Linear Prediction Through Informed Channel Coding
复制标题
不要担心位翻转：通过知情通道编码进行稳健的线性预测
DOI：
--
复制
发表时间：
2017
期刊：
影响因子：
--
通讯作者：
Guy Van den Broeck
中科院分区：
文献类型：
--
作者：
Frederic Sala;S. Kabir;L. Dolecek;Guy Van den Broeck
传统上，数据存储系统在独立层中提供了错误纠正和数据完整性技术，其目的是平均保护所有数据。在机器学习系统的背景下，此策略不合适：一些功能中的错误可能被证明非常重要（相对于算法输出），而许多错误可能几乎没有效果。这项工作采取了不同的方向：我们允许ML算法与错误纠正方案进行对话，目的是使算法稳健地对存储噪声进行鲁棒性。我们引入了一些新的问题，提供了一个有效的解决方案，以估计线性模型的算法输出的噪声变化，并展示如何优化错误校正代码以最大程度地减少固定开销的误差效应。
Traditionally, data storage systems provide errorcorrection and data integrity techniques in an independent layer, with the goal of protecting all the data equally regardless of the application. In the context of machine learning systems, this strategy is not appropriate: errors in a few features may prove to be critically important (with respect to the algorithm output), while many errors may have little or no effect. This work takes a different direction: we allow ML algorithms to talk to error-correction schemes, with the goal of making algoritms robust to storage noise. We introduce several novel problems, provide an efficient solution to estimate the noiseinduced change in the algorithm output for linear models, and show how to optimize errorcorrection codes to minimize error effects for fixed overhead.