Important
the way to learn are from books to implementations. Target 50 books
All code are written with plain python (.py), you must know about Modular Programming. All Production-Ready are used as feature with API behaviour (more or less). You think notebook are served to production-use?
The Advice to Learn from Andrej Karpathy or from George Hotz
Both of two are kinda contradictive, but you can take it as auxiliary
The books reference will be added later :)
| Area | Description | Topics/Tools/Technologies | Estimated Hours |
|---|---|---|---|
| Supervised Learning | Learning from labeled data to make predictions. | Linear Regression, Decision Trees, SVM, Random Forests, Scikit-learn, XGBoost, LightGBM | 800 |
| Unsupervised Learning | Learning from unlabeled data to find patterns. | K-Means, PCA, Hierarchical Clustering, Autoencoders, Scikit-learn, TensorFlow, PyTorch | 600 |
| Data Engineering | Data ingestion, transformation, and storage. | Apache Kafka, Apache Nifi, Flume, Apache Spark, Apache Beam, Talend, Hadoop, Amazon S3, Google BigQuery | 500 |
| Model Development | Model creation, training, and validation. | Jupyter, VS Code, TensorFlow, PyTorch, Scikit-learn, Git, DVC | 800 |
| Neural Networks | Study of artificial neurons and architectures. | CNNs, RNNs, LSTMs, GANs, Transformers, TensorFlow, PyTorch, Keras | 1000 |
| Natural Language Processing (NLP) | Processing and understanding human language. | Tokenization, Named Entity Recognition, Sentiment Analysis, NLTK, SpaCy, Hugging Face Transformers | 800 |
| Computer Vision | Enabling machines to interpret and understand visual data. | Image Classification, Object Detection, Image Segmentation, OpenCV, TensorFlow, PyTorch | 800 |
| CI/CD | Automating build, test, and deployment phases. | Jenkins, GitLab CI/CD, CircleCI, Apache Airflow, Kubeflow Pipelines | 500 |
| Model Management | Managing versions, metadata, and lifecycle of models. | MLflow, DVC, Neptune.ai, TFX, Metaflow | 400 |
| Infrastructure Management | Handling the underlying infrastructure for MLOps. | Terraform, Ansible, AWS, GCP, Azure, Kubernetes, OpenShift | 500 |
| Reinforcement Learning | Learning through rewards and penalties. | Q-Learning, Deep Q-Networks, Policy Gradients, OpenAI Gym, Ray RLlib, TensorFlow Agents | 600 |
| Model Deployment | Deploying models to production environments. | Docker, Kubernetes, Docker Swarm, TensorFlow Serving, TorchServe, FastAPI | 600 |
| Monitoring | Tracking model performance and system health. | ELK Stack (Elasticsearch, Logstash, Kibana), Prometheus, Grafana, Nagios, Seldon Core | 500 |
| Embedded Machine Learning | Implementing ML algorithms on resource-constrained devices. | TensorFlow Lite, PyTorch Mobile, CoreML, Edge Impulse, TFLite Micro, ARM CMSIS-NN | 800 |
| Edge AI | Deploying AI models on edge devices for real-time processing. | Nvidia Jetson, Intel Movidius, Google Coral, TensorFlow Lite, OpenVINO, AWS Greengrass | 800 |
| Low-Power Machine Learning | Developing ML models efficient in terms of power consumption. | TinyML, Quantized Neural Networks (QNNs), ARM Cortex-M, RISC-V, Microcontrollers | 600 |
| Real-Time Processing | Ensuring ML models can process data in real-time on embedded systems. | Real-time Operating Systems (RTOS), Stream Processing, FreeRTOS, Zephyr OS, Apache NiFi | 500 |
| Model Optimization | Reducing model size and improving efficiency. | Quantization, Pruning, Knowledge Distillation, TensorFlow Model Optimization Toolkit, ONNX Runtime | 500 |
| Sensor Fusion | Combining data from multiple sensors for more accurate predictions. | Kalman Filters, Bayesian Networks, Arduino, Raspberry Pi, Nvidia Jetson | 400 |
| Connectivity | Robust communication between embedded devices and cloud. | MQTT, CoAP, LoRaWAN, BLE, AWS IoT, Azure IoT, Google Cloud IoT Core | 400 |
| Security | Ensuring data privacy and security in embedded AI applications. | Secure Boot, Encryption, Anomaly Detection, Arm TrustZone, Secure Elements | 400 |
| Ethics and Fairness | Ensuring ethical AI and fairness in ML models. | Bias Mitigation, Explainability, Fairness Metrics, IBM AI Fairness 360, Google What-If Tool | 300 |
| Advanced ML Topics | More complex and specialized areas in ML. | Meta-Learning, Federated Learning, Few-Shot Learning, Transfer Learning, TFLite, Edge TPU, ONNX | 600 |
| Explainable AI (XAI) | Making AI decisions interpretable by humans. | SHAP, LIME, Explainable Boosting Machine (EBM), InterpretML | 400 |
| AutoML | Automated model selection and hyperparameter tuning. | AutoKeras, TPOT, H2O.ai, Google Cloud AutoML | 400 |
| Adversarial Machine Learning | Techniques to make models robust against adversarial attacks. | FGSM, PGD, Adversarial Training, CleverHans | 400 |
| Scalable Machine Learning | Approaches for scaling ML algorithms and infrastructure. | Apache Spark MLlib, Dask-ML, TensorFlow on Kubernetes, Horovod | 500 |
| Graph Neural Networks (GNNs) | Learning from data structured as graphs. | Graph Convolutional Networks (GCNs), Graph Attention Networks (GATs), DGL, PyTorch Geometric | 500 |
| Advanced Optimization Algorithms | Advanced optimization techniques for ML models. | Evolutionary Algorithms, Bayesian Optimization, Hyperopt, Optuna | 400 |
| Data Augmentation and Synthetic Data | Increasing the diversity of training data. | SMOTE, Data Augmentation, GANs for Synthetic Data, Augmentor, Imgaug | 400 |
- C, Rust, Python, Mojo
- The Tensor and its behaviour
- Machine Learning Compilers
- SNPE, coremltools, TensorRT, armnn
- more will be elaborated