c0der0712/cns_project

How c0der0712/cns_project works

This is a TaCo traffic-classifier robustness study: the notebook prepares packet-capture flows and MFR inputs, trains or loads classifiers, runs packet-level attacks, and records study results. The local package implements that evaluation workflow and supporting analyses, including capture-disjoint and masked-identifier experiments, adversarial fine-tuning, probes, transfer, calibration, and shortcut/window analyses. Edges rely on README statements and the supplied source excerpts or SOURCE INDEX; unsampled notebook and module internals are not asserted beyond those references.

Main entry points: attacks.py Attack search · data_io.py Flow loading · suite.py Attack suite · model_io.py Model objectives

Components 26

Study workflow

Experiment notebook CNS_Project.ipynb
Flow and MFR prep CNS_Project.ipynb
Classifier training CNS_Project.ipynb
Study results Results

Attack evaluation

Attack search Robust_evaluation/robust/attacks.py
Attack suite Robust_evaluation/robust/suite.py
Attack evaluation Robust_evaluation/robust/eval_attacks.py
Adversarial fine-tuning Robust_evaluation/robust/adv_finetune.py

Robustness analyses

Capture-disjoint split Robust_evaluation/robust/make_capture_split.py
Masked dataset Robust_evaluation/robust/mask_dataset.py
Linear probes Robust_evaluation/robust/linear_probe.py
Paired comparison Robust_evaluation/robust/paired_compare.py
Similarity calibration Robust_evaluation/robust/rq1_calibration.py
Foundation transfer Robust_evaluation/robust/foundation_transfer.py
Shortcut analysis Robust_evaluation/robust/shortcut_analysis.py
Window-group analysis Robust_evaluation/robust/window_groups.py

Evaluation core

Threat budgets Robust_evaluation/robust/threat_model.py
Flow loading Robust_evaluation/robust/data_io.py
Packet edit operators Robust_evaluation/robust/pcap_ops.py
MFR conversion Robust_evaluation/robust/pipeline.py
Identifier masking Robust_evaluation/robust/masking.py
Model objectives Robust_evaluation/robust/model_io.py
Confidence statistics Robust_evaluation/robust/stats.py

Other components

Researcher
ISCXVPN captures
TaCo models

Connections 44

Researcher → Experiment notebook: runs README.md
ISCXVPN captures → Flow and MFR prep: provides captures README.md
TaCo models → Classifier training: provides model code README.md
Experiment notebook → Flow and MFR prep: prepares flows README.md
Experiment notebook → Classifier training: trains classifiers README.md
Experiment notebook → Attack evaluation: runs evaluations README.md
Experiment notebook → Study results: writes outputs README.md
Flow loading → Packet edit operators: loads flows Robust_evaluation/robust/data_io.py
Flow loading → MFR conversion: converts flows Robust_evaluation/robust/data_io.py
MFR conversion → Packet edit operators: reads flow structure Robust_evaluation/robust/pipeline.py
MFR conversion → Identifier masking: applies masks Robust_evaluation/robust/pipeline.py
Attack search → Packet edit operators: generates edits Robust_evaluation/robust/attacks.py
Attack search → MFR conversion: scores MFRs Robust_evaluation/robust/attacks.py
Attack search → Model objectives: queries objectives Robust_evaluation/robust/attacks.py
Attack search → Threat budgets: obeys budgets Robust_evaluation/robust/attacks.py
Attack suite → Attack search: dispatches attacks Robust_evaluation/robust/suite.py
Attack suite → Flow loading: consumes items Robust_evaluation/robust/suite.py
Attack suite → Model objectives: builds objectives Robust_evaluation/robust/suite.py
Attack suite → MFR conversion: converts flows Robust_evaluation/robust/suite.py
Attack suite → Threat budgets: uses budgets Robust_evaluation/robust/suite.py
Attack evaluation → Attack suite: runs suite Robust_evaluation/robust/eval_attacks.py
Attack evaluation → Confidence statistics: summarizes outcomes Robust_evaluation/robust/eval_attacks.py
Adversarial fine-tuning → Attack search: crafts training examples Robust_evaluation/robust/adv_finetune.py
Adversarial fine-tuning → Attack suite: evaluates defense Robust_evaluation/robust/adv_finetune.py
Adversarial fine-tuning → Model objectives: uses objectives Robust_evaluation/robust/adv_finetune.py
Adversarial fine-tuning → Flow loading: loads training flows Robust_evaluation/robust/adv_finetune.py
Adversarial fine-tuning → MFR conversion: converts examples Robust_evaluation/robust/adv_finetune.py
Adversarial fine-tuning → Confidence statistics: reports intervals Robust_evaluation/robust/adv_finetune.py
Masked dataset → Identifier masking: uses masking Robust_evaluation/robust/mask_dataset.py
Linear probes → Model objectives: uses models Robust_evaluation/robust/linear_probe.py
Linear probes → Confidence statistics: aggregates results Robust_evaluation/robust/linear_probe.py
Paired comparison → Confidence statistics: computes differences Robust_evaluation/robust/paired_compare.py
Similarity calibration → Attack search: searches collisions Robust_evaluation/robust/rq1_calibration.py
Similarity calibration → Flow loading: loads test flows Robust_evaluation/robust/rq1_calibration.py
Similarity calibration → Model objectives: evaluates embeddings Robust_evaluation/robust/rq1_calibration.py
Similarity calibration → Confidence statistics: reports statistics Robust_evaluation/robust/rq1_calibration.py
Foundation transfer → Attack search: crafts transfer attacks Robust_evaluation/robust/foundation_transfer.py
Foundation transfer → Flow loading: loads flows Robust_evaluation/robust/foundation_transfer.py
Foundation transfer → Model objectives: evaluates models Robust_evaluation/robust/foundation_transfer.py
Shortcut analysis → Flow loading: analyzes flows Robust_evaluation/robust/shortcut_analysis.py
Shortcut analysis → Model objectives: examines models Robust_evaluation/robust/shortcut_analysis.py
Shortcut analysis → Confidence statistics: summarizes findings Robust_evaluation/robust/shortcut_analysis.py
Window-group analysis → Flow loading: groups flows Robust_evaluation/robust/window_groups.py
Window-group analysis → Model objectives: analyzes models Robust_evaluation/robust/window_groups.py

Drawn by GitDiagram on November 10, 2026 from the repository's file tree, README and source files. Read as Markdown ·