Speaker: Asif Qamar
LinkedIn: https://www.linkedin.com/in/asifqamar/
Technology Leader | AI/Data Scientist | Computer Scientist | Educator | Theoretical Particle Physicist
Key Highlights of the session :
Agentic Retrieval-Augmented Generation: A Survey on Agentic RAG
https://arxiv.org/html/2501.09136v1
- Semantic Embeddings & Vector Space: Semantic embeddings project text into a vector space, ensuring related topics (e.g., wildflowers and California poppies) are grouped together, while unrelated topics (e.g., car repairs) are far apart.
- Contrastive Loss for Embeddings: Transformer encoders, trained with contrastive loss functions, optimize semantic embeddings to better reflect the relationships between terms.
- Keyword vs. Semantic Search: Keyword search relies on matching specific words, while semantic search considers contextual relationships, providing more relevant results but with increased token costs.
- Naive RAG (Retrieval-Augmented Generation): Keyword search in naive RAG can result in fragmented or irrelevant answers, especially when words have multiple meanings (e.g., "bank" as a river or financial institution).
- Modular RAG: A hybrid approach, combining various search indices (e.g., neural search, structured databases, and APIs), to improve retrieval accuracy and efficiency.
- Advanced RAG Techniques: Integration of LLMs with local codebase retrieval systems (e.g., Cursor in VS Code) to provide more effective, localized search experiences and code suggestions.
- Agent-Based RAG: Future RAG systems will involve AI agents that dynamically manage queries and refine search results, enhancing contextual understanding and improving the retrieval process.
- Knowledge Graphs for Augmented Search: Knowledge graphs represent relationships between entities in the form of "triplets" (entity-relationship pairs), enhancing search accuracy and providing context.
- Automating Knowledge Graph Creation: LLMs can help automate the creation of knowledge graphs, extracting relevant titles from documents, though human review might still be necessary for high-quality results.
- RAG in SQL Generation: RAG can aid in generating SQL queries by using LLMs to infer relationships within structured data (e.g., tables, columns), improving the accuracy of SQL query generation for complex enterprise data.
Join over 2000 professionals
who have developed expertise in AI/ML
Become part of SupportVectors to learn about in-depth technical abilities and further your career.
https://supportvectors.ai