Thuật ngữ AI
Từ điển đầy đủ về Trí tuệ nhân tạo
Multimodal episodic memory
Memory system capable of storing and retrieving specific experiences combining multiple sensory modalities such as text, images, and audio. It allows for the reconstruction of complete events from multimodal information fragments.
Cross-modal encoding
Neuro-inspired process that transforms information from one sensory modality into representations compatible with other modalities. This encoding enables the harmonious integration of heterogeneous data into a unified memory.
Multimodal recall
Information retrieval mechanism that can be triggered by cues from any modality (visual, auditory, textual). The system generates coherent responses by automatically combining relevant modalities.
Multimodal hierarchical storage
Memory architecture organized into successive levels optimized for different temporal scales and complexities of multimodal information. Each level processes and stores data according to its relevance and access frequency.
Representation fusion
Technique that combines feature spaces from different modalities into a unified and coherent representation. The fusion preserves inter-modal correlations while reducing information redundancy.
Multimodal semantic memory
Structured knowledge base storing abstract concepts and their relationships across multiple representative modalities. It enables reasoning and inference on generalized multimodal knowledge.
Multimodal memory neural network
Deep learning architecture integrating specialized memory modules to process and store data of different natures. These networks use addressing and writing mechanisms adapted to each modality.
Multimodal indexing
Spatial organization system for information that enables rapid access to multimodal data based on their semantic content. The indexing combines similarity metrics specific to each modality.
Multimodal short-term memory
Temporary buffer of limited capacity that actively maintains recent multimodal information for immediate processing. It enables real-time integration of heterogeneous sensory streams.
Multimodal long-term memory
Large-capacity persistent storage retaining multimodal knowledge and experiences over extended durations. It uses consolidation mechanisms to transfer and optimize information from short-term memory.
Multimodal working memory
Active system that temporarily manipulates and transforms multimodal information to accomplish complex tasks. It simultaneously coordinates multiple information streams for ongoing reasoning.
Intermodal association
Process of establishing statistical and semantic links between information from different modalities. These associations enable cross-modal completion and prediction in memory.
Multimodal consolidation
Neuro-inspired process that gradually stabilizes and integrates multimodal traces into long-term memory. Consolidation strengthens intermodal correlations while eliminating redundant noise.
Multimodal similarity retrieval
Search algorithm that identifies relevant information based on their semantic similarity across all modalities. Retrieval combines heterogeneous metrics for optimal result ranking.
Multimodal memory architecture
Computational structure organizing storage, access, and management components for multimodal data. The architecture defines information flows and interaction protocols between different memory modules.
Multimodal declarative memory
Explicit memory system storing facts and events consciously accessible through multiple sensory representations. It supports direct recall and verbalization of multimodal knowledge.
Multimodal Procedural Memory
Implicit storage of skills and procedures involving the coordination of multiple sensory and motor modalities. This memory guides the automatic execution of complex multimodal tasks.