mlsanityData to 7 Oct 2026

Streaming Knowledge Compilation: Proactive Materiality-Scored Pinning for Time-Evolving LLM Wikis

arXiv 2606.09877 ↗First posted 3 Jun 2026cs.LG cs.CE cs.CLcited by 1

On the map: Efficient LLM inference and serving → KV cache optimization

Abstract (arXiv)

LLM wiki systems compile knowledge into pre-filled KV caches for efficient inference, but assume a static corpus -- an assumption that fails whenever the underlying information landscape evolves. We formalize Streaming Knowledge Compilation: given a document stream, a fixed token budget, and future queries unknown at ingestion time, maintain a compiled wiki that minimizes cumulative regret against an offline oracle with perfect foresight. The enabling insight is a materiality signal φt(k,n)∈[0,1] that scores document importance for entity k at time t, acting as a query-relevance surrogate for proactive pinning before queries arrive; we prove an O(√Tlog K) regret bound where varepsilon=mathbbE[|φt-hatφt|] is the only domain-specific quantity. We instantiate in two domains: finance, where φt is abnormal stock volatility predicted by frozen Llama 3.1 8B classification head (AUROC = 0.728 on 76K articles, strict temporal split; 1.49× higher realized forward volatility for predicted-material articles); and Wikipedia, where φt is the Abnormal Edit Ratio (AER), a cross-sectionally normalized edit velocity -- showing the same algorithm generalizes beyond the finance domain. End-to-end QA evaluation on 173 matched pairs (finance) and 119 (Wikipedia) reveals a pervasive LLM-as-judge confound on post-training knowledge, establishing that regret analysis -- not absolute QA scores -- is the reliable evaluation metric for compiled knowledge systems. Finance cumulative regret converges to -20.0 (-0.12/step); Wikipedia to +16.0 (+0.13/step), with the positive sign confirming that Wikipedia edit content is genuinely post-training -- richer context consistently improves scores (No Wiki 3.80 vs. Oracle 4.74) -- and eliminates this confound. The O(√Tlog K) guarantee applies to any domain where knowledge gaps can be predicted from streaming signals.

Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Placed in KV cache optimization, one of 12 directions in Efficient LLM inference and serving. Builds on 20 of its 52 references that are in the map. First posted on arXiv 3 Jun 2026.

Builds on

20 of 52 references.

Closest ancestors · 1

Same direction, newest first

  1. 2026WiCER: Wiki-memory Compile, Evaluate, Refine Iterative Knowledge Compilation for LLM Wiki Systems8 May

Other references · 19

  1. 2018BERT: Pre-training of Deep Bidirectional Transformers for Language Understandingcited by 40,204
  2. 2020Retrieval-Augmented Generation for Knowledge-Intensive NLP Taskscited by 9,160
  3. 2024From Local to Global: A Graph RAG Approach to Query-Focused Summarizationcited by 1,411
  4. 2016Understanding intermediate layers using linear classifier probescited by 1,279
  5. 2023BloombergGPT: A Large Language Model for Financecited by 923
  6. 2019Introduction to Online Convex Optimizationcited by 793
  7. 2024RAPTOR: Recursive Abstractive Processing for Tree-Organized Retrievalcited by 519
  8. 2019FinBERT: Financial Sentiment Analysis with Pre-trained Language Modelscited by 378
  9. 2023FinGPT: Open-Source Financial Large Language Modelscited by 244
  10. 2024Are Language Models Actually Useful for Time Series Forecasting?cited by 163
  11. 2023Can ChatGPT Forecast Stock Price Movements? Return Predictability and Large Language Modelscited by 125
  12. 2022StreamingQA: A Benchmark for Adaptation to New Knowledge over Time in Question Answering Modelscited by 123
7 more
  1. 2024From News to Forecast: Integrating Event Analysis in LLM-Based Time Series Forecasting with Reflectioncited by 93
  2. 2024WISE: Rethinking the Knowledge Memory for Lifelong Model Editing of Large Language Modelscited by 90
  3. 2023The Wall Street Neophyte: A Zero-Shot Analysis of ChatGPT Over MultiModal Stock Movement Prediction Challengescited by 59
  4. 2024Don't Do RAG: When Cache-Augmented Generation is All You Need for Knowledge Taskscited by 38
  5. 2025FinAgentBench: A Benchmark Dataset for Agentic Retrieval in Financial Question Answeringcited by 29
  6. 2024Fine-Tuning Large Language Models for Stock Return Prediction Using Newsflowcited by 8
  7. 2024RiskLabs: Predicting Financial Risk Using Large Language Model based on Multimodal and Multi-Sources Datacited by 4

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Training a Knowledge Base: Supervised Structure Learning for Agent-Curated Document Storescited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Mitigating Hallucination in Financial Retrieval-Augmented Generation via Fine-Grained Knowledge Verification
  2. 2026FinFIRST: Benchmarking Search Agents for Financial Information Retrieval, Sourcing and Traceability
  3. 2025Forget BIT, It is All about TOKEN: Towards Semantic Information Theory for LLMs
  4. 2026TRACE: Temporal Rule-Anchored Chain-of-Evidence on Knowledge Graphs for Interpretable Stock Movement Prediction
  5. 2025From Static to Dynamic: A Streaming RAG Approach to Real-time Knowledge Base
  6. 2025Beyond Exponential Decay: Rethinking Error Accumulation in Large Language Models
  7. 2026FinPerMA: A Theory-Informed, Event-Grounded Personalized-Memory Benchmark for LLM Agents
  8. 2026Absorbing Complexity: An Interaction-Native Knowledge Harness for Financial LLM Agents