---
title: "Discussion on \"I implemented backpropagation in 58 lines of NumPy. No libraries.\" | Hashnode"
description: "Discussion on \"I implemented backpropagation in 58 lines of NumPy. No libraries.\". Day 2: the 1986 paper turned into actual code, line by line"
url: "https://hashnode.com/posts/backpropagation-numpy-from-scratch/69d6bbe7707c1ce7687ff0e0"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [I implemented backpropagation in 58 lines of NumPy. No libraries.](https://papersbyhand.hashnode.dev/backpropagation-numpy-from-scratch?utm_source=hashnode&utm_medium=feed)

Day 2: the 1986 paper turned into actual code, line by line

Apr 2 · 4 min read · papersbyhand.hashnode.dev  
Read the article

-   [#numpy](https://hashnode.com/tag/numpy)
-   [#machine-learning](https://hashnode.com/tag/machine-learning)
-   [#xor](https://hashnode.com/tag/xor)
-   [#ai](https://hashnode.com/tag/ai)
-   [#beginners](https://hashnode.com/tag/beginners)
-   [#coding](https://hashnode.com/tag/coding)

## Written by

[Soham Kute](https://hashnode.com/@paperhand)

@paperhand · Mumbai, India

  
[View profile](https://hashnode.com/@paperhand)

building...

Day 3: I will add a loss curve, visualize the gradients in each layer, and push the full code to GitHub. You will see why gradients in early layers are much smaller than gradients in late layers. This matters for every deep model that exists.

## Responses

Markdown supported · ⌘↵ to post

No responses yet.
