CECO-LAD

v2 Cloud-Edge Collaborative Log Anomaly Detection
Idle
📄 Raw Logs log lines from source files
›
🔍 Parse Fields timestamp · level · component · message
›
🗂️ Sessions event sequences — input to the AI
›
⚡ Edge fast local scan (Q-BAT, 3 models)
›
🔀 Routing uncertain sessions → cloud
›
☁️ Cloud deep re-analysis (BAT, 81 models)
›
✅ Result anomaly predictions
Console Output
Waiting for pipeline output…
🔗 Pipeline
📊 Results
⚙️ System
📄Raw Log Lines

Data is loading…

— Go to page
Session
Source raw log lines (~0)
LineTimestampLevelMessage
Preprocessor output — sliding context window matrix (— events · each row = one event · columns = context steps back · · = NO_EVENT padding)
The preprocessor applies a sliding context window of length 10 over each session. For each event, the model sees the 10 preceding event IDs as context (filled with NO_EVENT=−1337 when no preceding event exists). These rows are fed row-by-row into the AI model (windows of 100 events at a time).
Event # Current t−9t−8t−7t−6 t−5t−4t−3t−2 t−1
🔍

Run inference to see the anomaly timeline.

Detection Thresholds

Each AI model learns a threshold — the boundary between "normal" and "anomaly". These values are calibrated on training data using a statistical method (EM-GMM).

⚙️

Run evaluate or inference to see thresholds.

Run Log Files
Select a log file above.
Forward sequence — sliding context window (size = 10 preceding events)