3 min readGuide
Mixture of Experts in LLMs: Capacity Through Selective Routing
How MoE LLMs work: routing, active parameters, memory, load balance and latency and quality evaluation across languages.
Research Notes on Federated Learning, Cybersecurity and Cyberdefense
Research notes and technical guides on federated learning, distributed AI, cyberdefense, trustworthy systems, and privacy-preserving platforms.
Topic: Trustworthy AI Clear topic
14 articles. Page 2 of 3.
3 min readGuide
Golden Sets for LLM Evaluation: Small, Curated and Difficult to Cheat
Build golden sets for LLMs and RAG with examples, adjudication, leakage-aware splits and bilingual security cases.
3 min readGuide
Metrics for LLMs, RAG and Cybersecurity Systems
LLM and RAG metrics: retrieval recall, citations, unsafe actions, calibration, latency and uncertainty. Define fair comparisons.
5 min readResearch note
How to Measure Robustness in Adversarial DFL Experiments
Evaluate DFL robustness with F1, false rejection, overhead and recovery. Interpret Flighter and compare clean and attacked conditions.
5 min readGuide
Designing a Decentralized Federated Learning System
Design a DFL system with topology, message contracts, mixing weights, recovery policies and per-client evaluation.
5 min readResearch note
From Monitoring to Mitigation: A DFL Cyberdefense Lifecycle with LLM Explanations
Connect telemetry, federated learning, alerts and LLM-assisted mitigation with traceable evidence, validation and reviewed feedback.