从粗到细 实践中，一般先进行初步范围搜索，然后根据好结果出现的地方，再缩小范围进行更精细的搜索。 先参考相关论文，以论文中给出的参数作为初始参数。 如果找不到参考，那么只能自...
发简信
IP属地：浙江
-
Knowledge transfer summary-2
所谓“知识精炼”我的理解就是将一个训练好的复杂模型的“知识”迁移到一个结构更为简单的网络中，或者通过简单的网络去学习复杂模型中“知识”或模仿复杂模型的行为。当然“知识”的定义...
-
Speeding-up CNN using CP-Decomposition
Approach We propose a simple two-step approach for speeding up convolution layers withi...
-
Exploiting Linear Structure Within CNNApproach Matrix Decomposition Higher Order Tensor Approximations Monochromatic Convolut...
-
Dynamic Network SurgeryApproach Han song recently propose to compress DNN by deleting unimportant parameters a...
-
Speed up CNN with Low Rank ExpansionsApproach Experiment References:Speeding up Convolutional Neural Networks with Low Rank ...
-
Learning Structured Sparsity in DNNApproach The optimization target of learning the filter-wise and channel-wise structure...
-
Fixed-point Factorized NetworksApproach Fixed-point Factorization Full-precision Weights Recovery The quantized weight...
-
Quantized CNN for Mobile DevicesApproach Firstly, we introduce an efficient test-phase computation process with the net...
-
Network SketchingApproach Approximating the Filters Speeding-up the Sketch Model Experiment References:N...
-
Incremental network quantizationApproach We present INQ which incorporates three interdependent operations: weight part...
-
Soft weight-sharing for DNNApproach where p(w) is the prior over w and p(D|w) is the model likelihood. After re-tr...
-
Deep CompressionApproach We introduce “deep compression”, a three stage pipeline: pruning, trained quan...
-
-
-
-
Network pruning summary
一直以来，网络剪枝都是模型压缩中的重要方法。按照被剪对象的粒度来分，大致可以分为三类： 针对权重剪枝，最具代表性的工作是韩松发表在NIP'15上的文章 “Learning b...
-
Dynamic Network Surgery
Approach Han song recently propose to compress DNN by deleting unimportant parameters a...
-
Pruning CNN for resource efficient inferenceApproach The proposed scheme for pruning consists of the following steps: Fine-tune the...
