Marking Sense of Hierarchical Log Data
summary
Argues that flow-level logs plus occasional program logs can be turned into strings of fork/join actions and compressed with a grammar variational autoencoder to produce a low-dimensional latent variable Z that clusters transaction types and reveals intent; uses a sequential fork-join simplification, proposes NS3 simulation and real sets (credit-card, student exercise logs), offers an unsupervised pipeline for diagnosing slow transactions and a synecdoche engine to surface representative example pairs rather than aggregates but reports no empirical results yet.
AI-written summary · about
chapters (12)
- 0:00Intro
- 1:00Data Centers
- 4:00Questions
- 6:22Log Messages
- 8:26App Dynamics
- 11:08Machine Learning
- 14:15Neural Networks
- 16:06Autoencoder
- 17:52Zedd
- 18:45Autoencoders
- 19:20Why
- 26:54Selector Key Engine
related talks
44mIntroduction to Sequence Analytics
40mClojure Emerging Data Ecosystem
34mHow to understand user behaviour using Multiple Linear Regression?
15mStatically Analyzing the Dataflow of R Programs
34mBuilding Data Driven Products with Ruby
59mData-driven program analysis
49mUnderstanding Customer Journeys with Deep Learning
29m5 Simple Steps to Create Meaningful Features from Clickstream Data