DCT breaks down the given input data into independent components of which the ones that don’t contribute much to the original input can be discarded, based on the tolerance for loss in quality. The JPEG etc. A neural network model learns W and b tensors which are stored with the model. Hence, they contribute significantly to its size. Of the two tensors, W naturally dominates the size owing to its higher

parameters which could be searched as well. transformation parameters in data augmentation layer contribute to performance improvements while others like learning rate, batch size or momentum are geared

given weights matrix with 8 bits of precision. We will leave the biases untouched since they do not contribute significantly to the layer’s footprint. num_bits = 8 weights_dequantized, weights_reconstruction_error_quant

quality for deployment. Both of these learning techniques help to improve training efficiencies and contribute to model quality and smaller footprints. This chapter has various exercises and projects to substantiate

是否正常。假设我们已经修改了文件~/d2l-en/chapter_appendix_tools/how-to-contribute.md中的一个拼 写错误。你可以检查你更改了哪些文件。 此时，Git将提示chapter_appendix_tools/how-to-contribute.md文件已被修改。 mylaptop:d2l-en me$ git status On branch master directory) modified: chapter_appendix_tools/how-to-contribute.md 16.5. 为本书做贡献 765 在确认这是你想要的之后，执行以下命令： git add chapter_appendix_tools/how-to-contribute.md git commit -m 'fix typo in git documentation'