IT solutions & software
Custom websites, mobile applications and management systems, built end to end.
PhD Candidate · LCSI Laboratory · ESI Algiers
Iyad Assaad
Nekka
I build anomaly detectors for graphs that change over time — and I make them say why they flagged what they flagged.
stream
initialising evidence pool…
01 — Research
Detecting what shouldn't be there,
in networks that never sit still.
I am a PhD candidate in Artificial Intelligence at the Higher National School of Computer Science (ESI, formerly INI and CERI) in Algiers, within the LCSI Laboratory. My thesis is on anomaly detection in dynamic graphs using deep learning, and specifically on graph neural networks.
Transactions between accounts, connections between hosts, ratings exchanged between traders — these arrive as a stream, not a snapshot. Deep detectors have become good at scoring them. They have not become good at explaining themselves, and an analyst who cannot see which counterparty or which moment triggered an alert cannot act on it. That gap is what my work addresses.
-
Explainable detection
Detectors whose explanation is the input to the decision, not a model fitted afterwards to guess at it.
-
Benchmarking & evaluation
Unified protocols for dynamic-graph anomaly detection, and the leakage channels that inflate published numbers.
-
Graph neural networks
Temporal transformers, enclosing-subgraph representations, and sparse selection over evidence.
-
Applied intelligence
Intelligent systems, smart-city applications, image processing and computer vision, with a growing focus on medical AI.
02 — Selected work
Papers
Statuses below are current as of this build. Verify before citing.
-
EVIDENT: An Ante-Hoc Evidence Bottleneck for Intrinsically Explainable Anomaly Detection in Dynamic Graphs
In preparationA detector whose anomaly score is computed only from a sparse evidence set it selects inside its own forward pass, so sufficiency holds by construction rather than by measurement.
-
Deep Learning for Anomaly Detection in Dynamic Graphs: A Verified Survey, Taxonomy and Benchmarking
In preparationA unified benchmark across detectors and corpora, and the finding that rankings established under synthetic anomaly injection do not survive contact with real labels.
-
Dual Spatial-Temporal Shapley Attribution for Explainable Anomaly Detection in Dynamic Graphs
Under reviewA post-hoc framework that wraps a frozen transformer detector and returns, per flagged edge, which neighbour nodes and which historical snapshots drove the alert.
-
Post-hoc explainability for dynamic-graph detectors
SeriesArchitecture-matched attribution for structural, contrastive, recurrent and semi-supervised detectors — each pairing a specific detector with an explanation procedure evaluated under a shared fidelity protocol.
Full list on ORCID , Google Scholar and ResearchGate .
03 — Outside the lab
ALIAN Business Services
I hold a Computer Science engineering degree and a Master's from the University of Batna 2. Alongside the PhD I founded and run ALIAN Business Services, based in Sétif.
Real estate brokerage
Property and consulting services, in collaboration with certified land experts (experts fonciers).
Administrative support
Visa application assistance and form-filling, handled properly the first time.
Find the office by searching “ALIAN by Iyad Assaad Nekka” on Google Maps.
04 — Contact
Get in touch
For research collaboration, reviewing, or anything about the work above, email is best.
ORCID