Neural Networks from Scratch Lecture 1: Coding a neuron and layers

Neural Networks from Scratch Lecture 1: Coding a neuron and layers

When students learn about neural networks, they typically use TensorFlow. It's a brilliant library through which neural network codes can be run in 10-15 lines as follows:

Using TensorFlow, any student deploy a neural network project using a matter of minutes.

However, this is not the best way to understand and master neural networks. Let us understand why.

Let's consider a practical example of Zomato delivery or Uber Eats, where a student is tasked to work on a food recommendation system.


  • Limited Flexibility: Prebuilt libraries like TensorFlow provide a lot of functionality out of the box, but they also limit the student's ability to customize the system according to specific requirements. For instance, in a food recommendation system, the student might need to incorporate unique features such as nutritional information, user preferences, or specific dietary restrictions. Without understanding the underlying mechanics of the system, the student will not be able to adapt the prebuilt library to these specific needs.
  • Troubleshooting: By not understanding how the system works, students may not be able to identify and troubleshoot issues effectively. Furthermore, they may not be able to adapt the system to new or unexpected challenges that arise during deployment.
  • Innovation: Without a deep understanding of the underlying principles, the student may struggle to adapt the library to new scenarios.

We have started a new Youtube series called "Building Neural Networks from Scratch". This series is entirely free.

In this series, we will show you how to build an entire neural network, right from the very basics.

Here is the content we plan to cover:

  • Coding our first neuron
  • Coding our first layers
  • Coding activation functions
  • Coding the loss function
  • Coding the forward pass
  • Coding the backward pass
  • Applications on real datasets

The first lecture is now available for free on Youtube. You can access it here:


If you've not followed our Youtube channel till now, you're missing out on the below playlists:

(1) Machine Learning Teach by Doing

(2) AI Researcher Bootcamp

(3) Mastering Generative AI

(4) DSA fundamentals

(5) Mastering Data Science

Here's the link to our Youtube channel:

Stay tuned for more exciting ML content on this newsletter!


Great initiative! I too am designing my own course on Physics-informed Neural Nets (PINNs) and I am very interested in the "ab-initio" approach to NNs. Coming from a Physics background, the only way I learn is by breaking things down and understanding the basics. I would love to connect with you and chat further!

Anoushka Tripathi

Winner @DIR-V Symposium Hackathon|FPGA Trainee @SSPL DRDO, Ministry of Defence, Govt. of India|Founder @Bharatiya Silicon Innovators|RISC V Design & Verification|Final year|VLSI Engineer|Bhāratīya

9 个月

Thank you so much for this amazing initiative

Mehran Sahil

Graduate Research Assistant & John E. Goldberg Fellow @ Purdue University | x Junior Engineer (Civil) @ Astral Constructors | GIKI'23 | Silver Medalist | S.M.ASCE | x Vice President @ ICE GIK Chapter

9 个月

I liked your videos on Machine learning Rajat Dandekar

要查看或添加评论,请登录

Vizuara的更多文章

社区洞察

其他会员也浏览了