Federated DDoS Detection
Master’s research on federated DDoS detection for service-provider customers, with evidence reviews, engineering checks, and bounded network-lab checks.
Nov 3, 2026
This section brings together my research projects, starting with my M.A.Sc. thesis at ÉTS. Each project has a stable home for its research question, references, code, and supporting materials. Later projects can join the same collection without replacing the earlier work.
November 18, 2026, controlled laboratory study executed and scored. The pre-registered study of design v2.1 has moved from design to results. Between 15 and 17 November the laboratory was built, independently audited and qualified, and two acquisition blocks were collected on 16 November. The packet conservation checks caught three generator faults, which were corrected, and physical conservation then held for all 164 attack episodes of the two blocks. The study was scored once, on 17 November, after independent audits of both analysis pipelines. The departures from the registered procedure are disclosed in the private draft.
On 17 November the research questions were revised with the supervisor. Packet sampling becomes the central question, federated transfer of an unseen attack type the second, the conservation checks a methods question, the specificity diagnostic an exploratory fourth, router cost a supporting fifth, and the conditional Sellier-inspired question becomes the sixth. The revised fifth question was measured in two parts with a stopping rule fixed before measurement. Training and exchange cost was measured in 40 trials, and the forwarding cost of metering and inference inside the emulated border router in 90 trials, with no failed trial in either part.
These results come from one laboratory in one virtual machine with generated benign traffic, so they describe a controlled evaluation and not a provider network. The private English draft now reports the methods and results; the researcher’s reading, native Word checks, French translation and the deposit remain open.
November 15, 2026, pre-registered controlled laboratory study, design v2.1. A new controlled laboratory study was designed and frozen before any build, traffic or data. It asks how packet sampling at a provider’s border routers changes what a router-side detector can see during flood-shaped DDoS attacks (RQ2), and how detection at a recipient point of presence changes when a second, contributing point of presence collaborates (RQ1 and RQ4). The study is pre-registered and separate from the earlier controlled track, which stopped at its pre-registered separability gate. That earlier result stays reported unchanged. No laboratory component of the new study has been built or started, and no traffic has been generated. Building the laboratory, qualifying it and acquiring data each need a separate release by the researcher.
The frozen design, topology, propositions and figure are published byte-identical to the frozen files, and their README states the status and limits of the design. The Lean project now holds 196 checked theorems in six modules, including 124 for the design propositions P1 to P12. The checker passes with the standard axioms only (propext, Classical.choice and Quot.sound). The formal-verification README states the limits of that check. For example, P4 is proved for an integer size only, P6 (d), P7 (b) and P9 are proved on finite sample spaces, P10 has its covariance identities only, and the keyed packet selector is treated as independent selection without proof. These are mathematical statements about the design, not measurements. The frozen design establishes what will be measured and how results will be judged. It does not establish detection performance, the effect of sampling, the benefit of collaboration or the behavior of a real provider network.
The next planned task is to write the offline laboratory build artifacts from the frozen design, using benign test fixtures only, and to have them independently audited. The researcher will then be asked for a go-ahead before the qualification and peak-load dry run. The laboratory virtual machine stays stopped until that release.
Earlier November 14, 2026, conditional source reconstruction and fit preparation. A completed reconstruction of previously used development evidence has passed an independent saved-output audit. Sampling and timeout views were replayed against retained ingress evidence, preserving the existing training and development assignments and common evaluation opportunities. Missing sampled observations remain coverage outcomes, not inferred classification decisions. This qualification is limited to the historical strict-conditioned cohort. It does not admit a wider source population, independent episodes, natural provider or customer roles, or untouched calibration and final-evaluation sets. New study fitting and scoring remain unrun.
Development training-controller checks and a separate statistical federated-method plan have also been audited. Actual-source fitting still depends on trusted readers, preserved training context and a bounded release. The methods retain different supervision and weighting, so these checks establish neither matched training work nor detection efficacy. The private English draft now records these boundaries; human reading and acceptance remain open.
Earlier November 14, 2026, ingress functional boundary and source screening. An independently audited finite native check exercised invented ingress packets, forwarding observations and flow-meter replay under the declared observation policy. The saved evidence also supports cleanup of the resources created for that check. This check included no learning or inference and does not establish detection efficacy, endpoint acceptance, deployment capacity, latency or source-data admission. A separate annotation-availability screen has completed; packet association, source eligibility, data-role assignment and fitting remain open gates.
The research plan retains four required questions and a conditional Sellier-inspired fifth question. The required work covers matched independent local, pooled-central and FedAvg learning with a separate statistical federated-system comparison; paired packet-sampling and timeout changes; router deployment with separate forwarding, inference and resource costs; and contributor-policy compatibility versus complementary DDoS behavior. The finite functional checks do not complete these evaluations.
Earlier November 14, 2026, development comparisons and separate native toy. A frozen FLNET development comparison has been executed. Reused observations, different transforms and optimizer choices limit causal and final-evaluation claims. A separate common two-counter flow-meter and training pilot retained an unfavorable fixed-decision finding. Its learned arms and count baseline produced no positive alerts. An earlier router-local component check exercised a source-trained model on synthetic probes inside a border-router process, together with forwarding, metering and collector-input parity. It establishes that finite component path, not detection efficacy or networked federated training. A distinct two-PoP native toy then exercised invented federated updates, model delivery and loading, and toy-model inference. Its benign toy records were flagged as positive. These false positives remain part of the record; the invented updates do not establish source-backed federated training. These observations establish neither a causal federated-learning advantage nor operational detection or scalability. Paired observation-policy experiments, workload and source-validity checks, human acceptance and French submission gates remain open. Reusable public packaging of the new implementations remains pending; earlier published source links retain their original scope.
November 12, 2026, frozen CSV scorer publication. The published score-only CSV interface provides export and loading interfaces for neural and AF-BKM-family model snapshots, owner-local TRAIN transform recovery and summary-only merging. Local and hosted checks of the integrated repository each discovered 328 unit tests, with 327 executed and one optional TShark test skipped. The engineering self-test and fabricated workflows passed. These checks use fabricated inputs; they do not establish detection accuracy, calibration or compatibility with live traffic.
The raw-data work supports only a narrow offline connection and counter reference from one inspected capture. Capture completeness, historical analyzer settings and independent attack episodes remain unresolved, so it has not been accepted as a controlled sampling/timeout benchmark. Final protocol selection, functional verification, and measured overlap and capacity checks remain open for the approved lab architecture and IP plan.
Earlier November 12, 2026, loader checkpoint. The checked role-aware loader reproduced the development records and their existing training, validation and calibration assignments. The required suite passed 77 unit tests, its engineering self-test and synthetic workflow. Two bounded independent code critiques found no consequential implementation defect. A separate feature-value and multiplicity audit remains pending before fitting. No learned preprocessing, model fitting, calibration or model scoring was performed.
This preparation check does not establish training capacity or deployment performance. The candidate retains source observers and reserved-test boundaries; stored records do not establish independent customers or a common original sampling/timeout policy.
November 5, 2026. The private Sellier-inspired chapter now has explanatory figures and an indexed author-discussion guide with anchored Word comments. Source checks distinguish published statements, illustrative calculations and the proposed federated adaptation. A bounded xAI public-mathematics critique was checked against the primary texts; it did not review the private chapter. Author clarification and a numerical implementation remain pending. No new training or detection result is reported.
November 4, 2026, historical design. The earlier provider design uses provider-owned collection and scoring at points of presence (PoPs), with central coordination. These PoPs are candidate learning participants, whereas customers remain reporting and evaluation groups. No detector appliances are assumed on customer premises. The planned comparison covers independent local, permitted pooled-central and federated training. It keeps the same observation view, feature definitions and PoP inference placement. Pooled and federated training use the same permitted fitting evidence. Independent local training retains its local evidence. The comparison remains unrun. Three additional research papers and two RFCs have been acquired, and selected full-paper FlowSentry sections now replace the earlier abstract-only check. No EVE topology or scientific code was changed.
The earlier English manuscript checkpoints remain private draft history. Human acceptance and broader empirical work remained pending at those cutoffs. Chapter 1 awaits human review. The provider reference diagram is now placed in Chapter 1, beside an explicit explanation of the illustrative topology. Chapter 4 refers back to the same figure. This is a manuscript presentation update; it does not change the architecture or establish lab readiness. The retained reading bank contains 98 works. The working Zotero collection has 103 parent references, including five supporting records. The public reference selection now contains 23 entries with primary links and a downloadable BibTeX file. These are separate records, not measures of research completion.
Three bounded public-source xAI architecture critique calls are recorded with grok-4.7-build. They did not review the new private Chapter 1 text. Their advice was checked against source evidence and study requirements; agreement does not establish novelty or validity. The code page records the generic design, prior art and targeted reading limits.
Earlier foundation work established DDoS, the learning task, model illustrations, gradient training and federated coordination before the study detail. Its separate Antigravity and xAI source-selection critiques retain their original scope and recorded models. The current acquisitions do not establish a new full-text appraisal of the whole bank, an exhaustive systematic review or independent replication. The collection-year chart remains a historical 87-record snapshot, not a measure of field-wide growth or novelty.
Required source-data checks, the final learner and protocol, and novelty assessment remain open. Sellier-inspired training remains an optional, bounded part of the planned work. English revisions precede French translation and proofreading. The reviewed pilot source remains unchanged.
The website connects the parts of each project. Original papers retain their authors, publication records, and DOI or repository links. Software links identify the source repository, and reported results should identify the exact code version and experiment record used to produce them.
Public project pages contain material prepared for sharing. Working drafts and private notes remain in the private thesis workspace, which requires an authorized account.
Substantive research progress is recorded on the project page with its date, evidence, limits, and next step. Published results will also identify their source and code versions.