1.1 Compare and Contrast Various AI Types and Techniques
Types of AI
- Generative AI AI systems capable of creating new content (text, images, code, media). Used for threat simulation and security content generation.
- Machine Learning (ML) Subset of AI enabling systems to learn from data without explicit programming. Foundation for cybersecurity detection systems.
- Statistical Learning Framework for inference and prediction from data; underpins ML algorithms.
- Transformers Neural network architecture for sequential data (LLMs, NLP), utilizing attention mechanisms.
- Deep Learning Subset of ML using multi-layer neural networks; excels at unstructured data (images, text) for threat detection.
- GANs Generator vs. Discriminator networks. Used for attacks (deepfakes) and defense (synthetic data).
- NLP Understanding/generating human language (chatbots, log analysis).
Model Training & Prompting
- Supervised Learning Training with labeled datasets (e.g., malware classification).
- Unsupervised Learning Finding hidden patterns in unlabeled data (anomaly detection).
- Reinforcement Learning Learning by interaction/rewards (adaptive security).
- Federated Learning Distributed training preserving privacy.
- System Prompts Defining AI behavior and guardrails.
- Pruning & Quantization Techniques to reduce model size/compute needs.
1.2 Data Security in Relation to AI
Data Processing & Integrity
- Data Cleansing: Removing errors to prevent "garbage-in-garbage-out".
- Data Verification: Confirming accuracy and consistency.
- Data Lineage/Provenance: Tracking origin and transformation for auditing/compliance.
- Data Balancing: Ensuring equal representation to prevent bias.
- Data Augmentation: Artificially expanding datasets for robustness.
Technologies
- Structured vs. Unstructured Data: Databases/CSV vs Images/Text.
- Watermarking: Embedding markers for content tracing and authenticity.
- RAG (Retrieval-Augmented Generation): Using Vector Storage and Embeddings for semantic meaning and context.
1.3 Security Throughout the AI Life Cycle
- Model Development: From business case alignment and data collection (trustworthiness/authenticity) to model selection.
- Deployment & Validation: Access controls, rollback procedures, and ongoing security validation.
- Human-centric Design: Human-in-the-Loop (active intervention), Human Oversight (monitoring), and Human Validation (QA).
