摘要: 1. The State of Sparsity in Deep Neural Networks(2019 optional) 工作 We perform a comprehensive evaluation of variational dropout , l0 regularization (L 阅读全文
posted @ 2022-10-08 08:27 撬动地球的coder 阅读(108) 评论(0) 推荐(0)
摘要: 1. Theano: A CPU and GPU Math Compiler in Python(2010 optional) 动机 Python is slow, one reason is that Python uses full-fledged Python objects on the h 阅读全文
posted @ 2022-09-01 17:34 撬动地球的coder 阅读(182) 评论(0) 推荐(0)
摘要: 1. Neural Adaptive Video Streaming with Pensieve (require 2017 MIT) 动机 Adaptive bitrate (ABR) algorithms are the primary tool that content providers u 阅读全文
posted @ 2022-08-29 08:08 撬动地球的coder 阅读(62) 评论(0) 推荐(0)
摘要: 1. On Large-Batch Training for Deep Learning: Generalization Gap and Sharp Minima(Optional 2017 Northwestern University) 动机 SGD and its variants理论属性: 阅读全文
posted @ 2022-08-09 18:08 撬动地球的coder 阅读(186) 评论(0) 推荐(0)
摘要: 1. Large Scale Distributed Deep Networks(Optional 2012, Google Jeffrey Dean) 动机: we consider the problem of training a deep network with billions of p 阅读全文
posted @ 2022-07-22 08:37 撬动地球的coder 阅读(245) 评论(0) 推荐(0)
摘要: 1. Interstellar: using halide’s scheduling language to analyze dnn accelerators (formerly: dnn dataflow choice is overrated. 动机 CNN存在大量的数据reuse,这为局部性优 阅读全文
posted @ 2022-07-07 18:40 撬动地球的coder 阅读(252) 评论(0) 推荐(0)
摘要: 简单记录下论文中某些要点,对应的实现方式。主要函数都在fix_quant_ops.py中实现。 1. 激活值fractional lengthlen grid search实现步骤 def fraclen_gridsearch(input, wl, align_dim, signed): err_l 阅读全文
posted @ 2022-06-16 17:26 撬动地球的coder 阅读(115) 评论(0) 推荐(0)