AI-ordlista
Den kompletta ordlistan över AI
RAG Fine-tuning
Process of specifically adapting language models to optimize their performance in retrieval-augmented generation tasks, by adjusting network weights on relevant contextual data.
Retrieval-Augmented Generation
Hybrid architecture combining information retrieval and text generation, where the model consults an external knowledge base before producing more accurate and factual responses.
Knowledge Distillation for RAG
Technique for transferring knowledge from a complex RAG model (teacher) to a lighter model (student), preserving retrieval-based reasoning capabilities.
Embedding Fine-tuning
Adaptation of vector representations to improve the relevance of retrieved documents, by optimizing embeddings according to the specific RAG application domain.
Retrieval Encoder Adaptation
Modification of retrieval encoders to better understand and index domain-specific documents, thus improving the accuracy of the search phase in RAG.
Prompt Engineering for RAG
Optimized design of input instructions to effectively integrate retrieved information into the generation process, maximizing response coherence and relevance.
Vector Database Fine-tuning
Optimization of vector database parameters to accelerate similarity queries and improve the quality of retrieved documents in RAG systems.
Cross-Encoder Optimization
Adjustment of cross-encoder models for precise reranking of retrieved documents, improving the selection of the most relevant information for generation.
Multi-Modal RAG Adaptation
Extension of RAG techniques to process and integrate different types of data (text, images, audio) in the retrieval and augmented generation process.
Domain-Specific RAG
Specialization of RAG systems for specific domains (medical, legal, technical) by adapting models and knowledge bases to specific terminologies and concepts.
Few-Shot Learning for RAG
Learning technique allowing the RAG model to quickly adapt to new tasks with very few examples, leveraging contextual retrieval capabilities.
Retrieval-Generation Alignment
Optimization process aimed at aligning semantic representations between retrieval and generation phases to ensure better consistency in produced responses.
Hybrid Search Fine-tuning
Combined optimization of vector and semantic search in RAG systems, adjusting weights between different methods to maximize result relevance.
RAG Performance Calibration
Systematic adjustment of model parameters to balance confidence between retrieved information and internal knowledge, avoiding hallucinations while remaining consistent.