| layout | page |
|---|---|
| title | Papers |
UCEF: Universal Context Extension Framework — Breaking the Context Barrier with Hyperbolic Geometry and Quantum-Inspired Selection
Honglin He
arXiv preprint, 2026
GitHub · Documentation
We propose UCEF, a model-agnostic framework combining hyperbolic retrieval, quantum-inspired selection, and adaptive compression with quality feedback to extend any LLM's effective context window. Evaluated on 8 LongBench tasks with GLM-4-flash and DeepSeek-v3, UCEF achieves up to +19.3% ROUGE-L improvement over standard RAG baselines.
- Nickel & Kiela, "Poincaré Embeddings for Learning Hierarchical Representations", NeurIPS 2017
- Cao et al., "HyperbolicRAG", arXiv 2025
- Madhu et al., "HypRAG: Hyperbolic Dense Retrieval for RAG", arXiv 2026
- Kuznetsov et al., "Quantum-Inspired Self-Attention in a Large Language Model", arXiv 2026
- Li et al., "ATACompressor: Adaptive Task-Aware Compression", arXiv 2026