Batchnorm has other parameters - in addition to weight and bias - that are not counted in torchsummary, such as running_mean and running_var:

https://pytorch.org/docs/stable/_modules/torch/nn/modules/batchnorm.html

HIBERT: Document Level Pre-training of Hierarchical Bidirectional Transformers for Document Summarization
Xingxing Zhang, Furu Wei, Ming Zhou
to appear in ACL 2019
https://arxiv.org/abs/1905.06566

0. 論文

Document Similarity for Texts of Varying Lengths via Hidden Topics
Hongyu Gong* Tarek Sakakini* Suma Bhat* Jinjun Xiong †

1. どんなもの？

文章の長さが大きく違うような文章同士の類似度を測る手法

2. 先行研究と比べてどこがすごい？

本文と要約のように長さや情報量が大きく異なる文章の類似度を測ることができる

3. 技術や手法のキモはどこ？

Word Mover's Distanceのような手法では長さが大きく異なる文章間の類似度計測には向いていない。
そこで長い方の文章(e.g. 本文)から潜在的内トピックをベクトルで取り出

The documentation suggests, that you can use wildcards in the "ignores" configuration. As a matter of fact, the values in this array are compared on a 1:1 basis. No regex included.

Feature request: Ignores configuration should support wildcards or regular expressions.

Vincent Dumoulin, Francesco Visin

We introduce a guide to help deep learning practitioners understand and manipulate convolutional neural network architectures. The guide clarifies the relationship between various properties (input shape, kernel shape, zero padding, strides and output shape) of convolutional, pooling and transposed convolutional layers, as well as the relationship between convo

[Tutorial] Asynchronous Flutter chat client with Go chat server which are powered by gRPC (simple…

Aleksandr Sokolovskii Jan 13 I opened Flutter for myself few weeks ago. It dramatically changes mobile app development. I would call it now — development with…



January 14, 2019 at 01:32PM

via Instapaper [https://medium.com/@amsokol.com/tutorial-asynchron