Oy 3 — Deep Learning

🎯 Bu oydagi maqsad

Oy oxirida siz quyidagilarni qila olasiz:

• Neural network nima ekanini, qanday ishlashini tushunasiz
• PyTorch'da o'z neural network'ingizni quryasiz va o'rgatasiz
• TensorFlow/Keras bilan tanishasiz
• CNN bilan image classification qila olasiz
• RNN/LSTM bilan sequence data'ni qayta ishlay olasiz
• Transfer learning'ni qo'llay olasiz

Haftalik taqsimot

Boblar tartibi

1. Neural Networks asoslari — perceptron, backprop, intuition
2. PyTorch asoslari — tensor, autograd, nn.Module
3. TensorFlow va Keras — alternativ framework
4. Training texnikalari — optimizers, regularization, callbacks
5. CNN — Convolutional Networks — rasm classification
6. RNN, LSTM, GRU — sequence data
7. Mashqlar

Oy oxirida nima qila olasiz?

• PyTorch'da nn.Module yozish va training loop qurish
• MNIST, CIFAR-10 kabi datasetlarda 95%+ accuracy
• Pretrained model (ResNet, EfficientNet) ni fine-tune qilish
• FastAPI orqali GPU-powered prediction servis
• ML model'larni torch.save / torch.jit bilan production'ga olib chiqish

Backend Dev uchun maslahat

DL = "Layered functions + Automatic differentiation". Sizga 2 ta narsakerak:

1. Model arxitekturasi — qatlamlarni yig'ish (LEGO kabi)
2. Training loop — for-each-batch: forward → loss → backward → optimizer

Birinchi marta murakkab tuyuladi, lekin 2-3 ta misol yozgandan keyin "patternni" sezasiz.

Hardware haqida

DL — bu CPU emas, GPUuchun yaratilgan. Variantlar:

1. Mac M1/M2/M3 — MPS backend (PyTorch 2.0+) — kichik modellar uchun yetarli
2. Local NVIDIA GPU(RTX 3060+) — CUDA + cuDNN o'rnatish
3. Google Colab — bepul T4 GPU (12 soat/sessiya) — TAVSIYA
4. Kaggle Notebooks — bepul P100 GPU (30 soat/hafta)
5. **Pullik:**Lambda Labs, vast.ai, RunPod — soatiga $0.20-2

**Maslahat:**Lokal mashqlar uchun CPU/M-chip, capstone uchun Colab/Kaggle GPU.

Boshlash

Neural Networks asoslari ga o'ting.