More Layers Increase Accuracy
programmer, can you give me more layers To increase accuracy? -Yeeess. Actually underfits like a boss vanishing_gradient_time f(x) = 1 / (1 + e⁻ˣ) tanh x = (eˣ − e⁻ˣ) / (eˣ + e⁻ˣ)
A four-panel comic shows a programmer asking for more neural-network layers to increase accuracy. The network becomes increasingly complex, then still underfits, while the final panel references vanishing gradients and activation functions, joking about how adding layers can make deep networks harder to train. It is sent as a reaction to overengineering machine-learning models or discovering that more complexity does not solve the problem.
neural network diagram / схема нейронной сетиprogrammer / программистwoman with glasses / женщина в очкахcomputer code / компьютерный кодneural network layers / слои нейронной сетиmathematical graph / математический графикred crosses / красные крестикиblue circles / синие кружкиgreen foliage / зелёная растительностьwhite background / белый фонformula / формулаvanishing gradient / исчезающий градиент
#neural network#нейронная сеть#machine learning#машинное обучение#deep learning#глубокое обучение#programmer#программист#underfitting#недообучение#vanishing gradient#исчезающий градиент#layers#слои#accuracy#точность#overengineering#переусложнение#coding#кодинг#reaction#реакция#comic#комикс#technical meme#технический мем
- @profunctor_io / 1558 2018-12-24
- Type
- image
- Size
- 500×629 · 71.9 KiB
- Mood
- sarcastic, absurd
- Language
- en
- Added