MARS-LEARN
🚀 MARS-LEARN
🌱 Lecturer: Kao Panboonyuen
📚 Main Resources
🧠 Deep Learning Architecture Masterclass
🚀 From Simple CNNs to Modern Vision Architectures
This masterclass is designed to bridge the gap between introductory deep learning and real-world AI engineering.
Participants will move beyond basic datasets such as MNIST and CIFAR-10 and explore challenging image classification problems using modern TensorFlow workflows.
🎯 Learning Objectives
By the end of this session, participants will be able to:
- Build and train deep convolutional neural networks (CNNs)
- Understand why deeper architectures outperform shallow models
- Apply modern training strategies:
- Batch Normalization
- Dropout
- Residual Connections (ResNet)
- Learning Rate Scheduling
- Data Augmentation
- Transfer Learning
- Compare performance across multiple architectures
- Analyze overfitting, generalization, and optimization behavior
- Train production-style image classifiers using TensorFlow/Keras
🧪 Hands-On Masterclass Notebook
🔥 TensorFlow Image Classification Masterclass
Topics covered include:
- CNN Fundamentals
- Deep CNN Architectures
- ResNet-style Networks
- Transfer Learning with EfficientNet
- Data Augmentation Pipelines
- Fine-Tuning Strategies
- Performance Evaluation
- GPU Training Optimization
🗂️ Datasets Used in This Masterclass
This course intentionally uses datasets that are more challenging than MNIST/CIFAR while remaining practical for GPU training in Google Colab.
| Dataset | Focus Area |
|---|---|
| Fashion-MNIST | CNN Baselines |
| CIFAR-100 | Deep Architecture Comparison |
| Oxford-IIIT Pets | Transfer Learning |
| Tiny ImageNet | Advanced Classification |
| Food-101 | Real-World Visual Complexity |
📚 Main Resources
-
Slides on Basic AI to MARS
-
(Machine Learning Lab) Titanic Machine Learning
-
(Deep Learning Lab) Dental Classification
-
(Deep Learning Masterclass) TensorFlow Image Classification Architecture
📊 MLflow for MARS (MLOps & Experiment Tracking)
We introduce MLflow for MARS (Motor AI Recognition Solution) as a key component for building a scalable and production-ready MLOps pipeline for computer vision tasks such as vehicle detection.
This material demonstrates how MARS can manage:
- Experiment tracking (hyperparameters, metrics, runs)
- Model versioning and reproducibility
- Training comparison for YOLO-based detection systems
- End-to-end MLOps workflow from training to deployment
It also includes a practical comparison between MLflow and Weights & Biases for real-world AI development workflows in MARS.
📄 Download Slides (MLflow for MARS)
View Slides (PDF)
🎯 About MARS-LEARN
MARS (Motor AI Recognition Solution) is an AI company specializing in car insurance solutions. MARS-LEARN is a dedicated session for updating AI knowledge, providing foundational learning resources in AI and data modeling for MARS team members. This repository includes curated materials, hands-on labs, and interactive tutorials to help students and professionals enhance their machine learning and deep learning skills. Whether you’re a beginner or an advanced practitioner, MARS-LEARN serves as a comprehensive learning hub.
🔥 Stay Updated!
New materials will be added regularly, so keep checking for updates. Happy coding! 🚀