Automatic aiding system for Linux kernel using ML
summary
A machine‑learning aided triage system converts kernel error text into vector embeddings using bertobeck, trained on 63,000 messages, sums token vectors into a unit vector and uses cosine similarity via a matrix multiply to retrieve the top three matches. Traditional string‑similarity searches took seconds to tens of seconds (3–36 s); optimized fuzzy methods reached 39 ms or 0.5 s by discarding most data. The web app returns results in below 10 ms on the full dataset and links to existing solved cases for immediate guidance.
AI-written summary · about
chapters (23)
- 0:00Introduction
- 1:10Global Support Services
- 1:38Red Hat Knowledge Base
- 2:28Example
- 3:05Issue line
- 5:58Linux kernel
- 6:35Call trace
- 6:53String similarity algorithms
- 7:56Performance test
- 8:59Search algorithms
- 10:15Search time
- 11:06Graph
- 11:48Optimization
- 13:20Word Embedding
- 14:41Analogy Detection
- 15:33dimensionality reduction
- 16:15vectors
- 17:02division
- 17:43recap
- 18:04single vector
- 18:45speed
- 19:03test application
- 20:12second match
related talks
41mML applications in Linux kernel
10mML-based Log Analysis for Faster Debugging
48mUsing Machine Learning to find Linux bugs
46mSW Testing: Can ML save us?
19mAutomatic Metric Screening for Service Diagnosis
45mUsing Machines to exploit Machines
47mAI vs. Karen: Real-Time Customer Sentiment Detection at Scale
41mSemantic Multimodal Multilingual Similarity Engine