Künstliche Intelligenz (KI/AI) + Cloud

Cloud Computing ist die natürliche Heimat moderner KI — GPU-Verfügbarkeit, Skalierbarkeit, Pay-as-you-go-Pricing und vorgefertigte AI-Services machen Cloud unverzichtbar. Dieser Kurs verbindet KI-Wissen mit praktischer Cloud-Implementierung. KI-Grundlagen (1 Woche): Machine Learning-Konzepte (Supervised, Unsupervised, Reinforcement), Deep Learning-Basics (Neural Networks, Backpropagation, CNN für Vision, Transformer für NLP/Multimodal), Generative AI (LLMs, Diffusion Models). Cloud-Grundlagen (1 Woche): IaaS/PaaS/SaaS, Shared Responsibility Model, Multi-Cloud vs. Single-Cloud, Identity & Access Management (IAM), VPC/VNet, Storage (Object, Block, File). AWS AI Services (3 Wochen): SageMaker (Studio, Training Jobs, Endpoints, Pipelines, Feature Store, Model Monitor), Bedrock (Foundation Models von Anthropic Claude, Meta Llama, Mistral, AI21, Amazon Titan; Knowledge Bases mit RAG; Agents; Guardrails), Comprehend (Sentiment, Entities, Topics), Rekognition (Image/Video Analysis, Faces, Celebrities), Textract (OCR), Polly (Text-to-Speech), Transcribe (Speech-to-Text), Translate, Lex (Conversational AI), Personalize (Recommendations), Forecast. Azure AI Services (3 Wochen): Azure Machine Learning (Workspaces, Compute, Pipelines, Designer, AutoML, Endpoints, MLflow), Azure AI Services / Cognitive Services (Vision, Language, Speech, Document Intelligence, Translator, Search), Azure OpenAI Service (GPT-4o, o1, embeddings, DALL-E, Whisper), Azure AI Foundry (ehemals AI Studio) für End-to-End-AI-Development. GCP AI (2 Wochen): Vertex AI (Workbench, Training, Endpoints, Pipelines, Model Garden mit Gemini/Llama/Claude), Document AI, Translation API, Speech-to-Text, Vision AI, Natural Language API, Recommendations AI, Dialogflow CX für Conversational AI. Architekturmuster (1 Woche): Serverless ML (AWS Lambda + SageMaker Endpoints, Azure Functions + ML Endpoints), Edge AI (AWS Greengrass, Azure IoT Edge, Google Edge TPU), Hybrid AI (On-Premise Training, Cloud Serving). Cost Optimization: Spot/Preemptible Instances, GPU-Sharing (MIG), Inference Optimization (Quantization, Distillation, Pruning), Caching.