Replication maintains copies across nodes for high availability, disaster recovery, and load distribution. It doesn't eliminate backups—they serve different purposes.

Modern languages use Introsort which starts with Quicksort (O(n log n) avg) but switches to Heapsort (O(n log n) worst) to avoid worst-case scenarios.

Overfitting occurs when a model memorizes training data including noise, reducing generalization. Regularization and cross-validation help prevent it.

NoSQL databases prioritize flexible schemas and horizontal scaling. They may use eventual consistency (BASE model) rather than strict ACID.

Sharding partitions data based on a key (e.g., user ID) across servers. This horizontal scaling improves performance and handles large datasets efficiently.

Activation functions (ReLU, Sigmoid, Tanh) add non-linearity, allowing neural networks to learn non-linear relationships. Without them, networks become linear classifiers.

Microservices decompose applications into loosely-coupled, independently deployable services. This enables scalability, flexibility, and technology diversity.

Connection pooling maintains a pool of reusable connections, reducing the cost of opening/closing connections repeatedly. This significantly improves application performance.

Insertion Sort has O(n) best-case complexity for nearly sorted data. Other algorithms don't exploit this, maintaining O(n log n) regardless of initial order.

Least privilege minimizes security risk by restricting access. Users get only permissions needed for their role, reducing damage from compromised accounts.