Day 3 of Music Genre Classification Bootcamp

Day 3/10 – Audio Classification Bootcamp
Building a Basic CNN Model + Training Loop (Step-by-Step)

Github : https://github.com/Hariswar8018/Music-Genre-Classification-Deep-Learning-Project

Welcome to Day 3 of the 10 Days Audio Classification Bootcamp 🌍
Today we move from preprocessing to building and training a real Convolutional Neural Network (CNN) for audio classification.
This session covers:
Model architecture design
Choosing loss & optimizer
Understanding Conv2D layers
Writing a proper training loop
Adding CUDA for GPU acceleration
This is where theory meets practice.

Timeline :
0:00 – Start
0:46 – Loss Function: Cross Entropy Loss
2:01 – Choosing Optimizer: Adam
2:42 – Important Points During NN Design
3:15 – Metric: F1 Score
4:30 – How a DL Model Looks Structurally
5:06 – Making Neural Network Model Class
6:25 – Activation Function: ReLU
7:28 – Conv2D Layer (Working & Format)
10:17 – Conv2D In-Channels
14:55 – Conv2D Out-Channels
15:55 – BatchNorm2d Layer
16:32 – Pooling Layer (Max + Average)
17:57 – Dropout Layer
18:51 – Flatten Layer
19:47 – Linear Layer
21:19 – Secret Reveal 🙂
21:35 – Adaptive Average Pooling
22:37 – Important Points Before Joining NN
22:45 – Joining the NN Together
23:45 – Initializing Model, Loss, Optimizer
24:44 – Training Loop (Basic Version with Break)
26:02 – Adding Optimizer Step & Backpropagation
26:15 – Full Training Loop
26:54 – Adding CUDA Support
28:23 – Summary
30:00 – Training Done
30:37 – End
🧠 What You Will Learn
Why CrossEntropyLoss is ideal for classification
Why Adam optimizer is commonly used
How Conv2D really works (channels, filters, feature maps)
How to structure a CNN class in PyTorch
How training loops actually function
How backpropagation fits into the loop
How to move model & tensors to GPU (CUDA)
🎯 Why This Video Is Important
Many learners:
Copy CNN architectures
Copy training loops
Copy optimizer code
But do not understand why each part exists.
Today you build:
Architecture clarity
Training clarity
Optimization clarity
That is real Deep Learning.
🌍 About This Bootcamp
This 10-day bootcamp is structured for a global audience, moving step by step:
Day 1 → Audio Foundations
Day 2 → Feature Extraction
Day 3 → CNN + Training Loop
Next → Evaluation, tuning & improvement
No shortcuts. Real implementation.
👨‍💻 About Me
I teach Deep Learning & Generative AI, and I also work as:
💙 Flutter App Developer (Freelancing)
🤖 AI / ML Engineer
My approach is always: Clarity → Structure → Practical Execution.
🔗 Connect With Me
LinkedIn: https://www.linkedin.com/in/ayusman-samasi/�
GitHub: https://github.com/Hariswar8018�
Kaggle: https://www.kaggle.com/samasiayushman�
☕ Support (Optional)
If this bootcamp helps you:
UPI: ayusmansamasi@ybl
Buy Me a Coffee: https://buymeacoffee.com/hariswarsax�
Your support helps me continue building structured AI education 🌱
✨ Day 4 will move toward improving performance and evaluation techniques.

Видео Day 3 of Music Genre Classification Bootcamp канала Ayus Dev 🌈