KI-Glossar
Das vollständige Wörterbuch der Künstlichen Intelligenz
Differentiable Neural Computer
Extension of Neural Turing Machines with improved memory mechanisms and a more complex architecture capable of learning algorithms and reasoning on structured data.
Key-Value Memory Networks
Memory architecture storing information as key-value pairs, enabling efficient retrieval based on similarity between keys and input queries.
Memory Addressing
Process of selecting memory locations for read or write operations, using content or location-based techniques for precise access.
Read/Write Operations
Fundamental operations for accessing external memory, implemented in a differentiable manner to allow gradient backpropagation during training.
External Memory Matrix
Two-dimensional data structure storing information external to the main neural network, accessible through controlled read/write operations.
Memory Controller
Neural network responsible for generating read/write operations and memory addresses, acting as interface between main processing and external memory.
Working Memory
Dynamic short-term memory used to temporarily store relevant information during the processing of a current task.
Memory Interpolation
Technique for combining or interpolating between multiple memory locations to generate hybrid or generalized representations.
Memory Compression
Process of reducing the required memory space while preserving essential information, often used to manage large memories.
Memory Retrieval
Mechanism for searching and extracting relevant information stored in external memory based on specific queries or contexts.
Memory Consolidation
Process of transferring and stabilizing information from short-term memory to long-term memory for permanent storage.