Skip to content

Latest commit

 

History

History
10 lines (9 loc) · 972 Bytes

README.md

File metadata and controls

10 lines (9 loc) · 972 Bytes

denoising-autoencoder

3層のデノイジングオートエンコーダをnumpyのみを用いて実装しました.
偏微分の計算はライブラリなどを使わずに,微分の連鎖律を手で計算することで導出しました.
今回は学習を,層数が3層,ノード数が[100,80,100],データ数が50,データ長が100,学習係数を10,学習回数を10000,という条件の下で行いました.
結果はsrc/resultの中にあります.

I implemented a three-layer Denoising Auto-encoder using only numpy.
The partial derivative was derived by computing the chain law of the derivative by hand, without using a library.
This time, the learning was performed under the conditions that the number of layers is 3, the number of nodes is [100,80,100], the number of data is 50, the data length is 100, the learning coefficient is 10, and the number of learning is 10000.
You can find the result in src/result.