Les belles promesses

Pierre Lemaitre

Description

1. Deep Learning Crash Course: CNNs, Transformers & PyTorch Code

Dive into the heart of modern AI with this hands-on crash course that takes you from classic convolutional networks to the revolutionary Transformer architecture powering today's large language models. Built for developers who learn by doing, each section pairs crystal-clear explanations with ready-to-run PyTorch code. Start with AlexNet's breakthrough on ImageNet, master residual connections in ResNet, then construct a full Transformer block from scratch. Train a vision model on real images, fine-tune a language model on custom text, and deploy both using PyTorch Lightning. Every concept includes a complete Colab notebook, annotated line-by-line, so you can experiment instantly. By the final page you'll have trained, evaluated, and exported production-ready models—without wading through unnecessary theory. Perfect for intermediate Python users ready to level up to state-of-the-art deep learning.

2. Fine-Tune BERT & GPT: NLP with Hugging Face

Transformers aren't magic—they're code you can control. This focused guide strips away the hype and walks you through fine-tuning BERT and GPT-style models using the Hugging Face ecosystem. Begin with tokenization, attention masks, and dataset pipelines, then move to real tasks: sentiment classification, named-entity recognition, and text generation. Each chapter delivers a complete, end-to-end script: load a pre-trained checkpoint, prepare your data with Datasets, train with Trainer, evaluate with industry metrics, and push to the Hugging Face Hub. Learn to manage GPU memory, apply LoRA for parameter-efficient tuning, and avoid common pitfalls like catastrophic forgetting. By the end you'll own a custom model that outperforms generic APIs on your domain-specific data. Ideal for data scientists and engineers who need practical NLP results fast.

3. YOLOv8 & GANs: Real-Time Vision with Python

See computer vision in action with two powerhouse techniques: ultralight object detection and stunning image synthesis. This compact volume puts YOLOv8 and modern GANs in your hands with zero fluff. Train a YOLO detector on custom objects in under an hour, export to ONNX for edge deployment, and run real-time inference on webcam feeds. Then flip to generative models: build a DCGAN to create faces, implement CycleGAN for style transfer, and explore latent-space interpolation. Every example ships with a standalone script, pretrained weights, and a Colab link so you can iterate immediately. Master non-max suppression, anchor-free detection, and progressive growing a GAN training schedule. Whether you're prototyping for robotics or generating art, you'll finish with deployable vision tools that work in the real world.

4. Your First AI Model: Predict Churn in 30 Lines

Turn raw data into a working predictive model in one afternoon. This beginner-friendly walkthrough builds a customer-churn classifier from scratch using only Scikit-learn, Pandas, and Streamlit. Load a CSV, explore distributions with one-liners, engineer features with pipelines, and train a Random Forest in thirty readable lines. Visualize decision paths, tune hyperparameters with GridSearchCV, and interpret results with SHAP values—all inside a single notebook. Deploy your model as an interactive web app in one click. Every step is annotated, every pitfall explained, and every line copy-paste ready. No prior machine learning required; just basic Python. By the final deploy you'll understand the full data-to-dashboard lifecycle and have a portfolio piece you can show any hiring manager.

Spécifications