AI & ML
Smoothing and Learning for Movement-Robust RSS-Based Spoofing Detection
arXiv cs.CR0 views
arXiv:2611.23854v1 Announce Type: new Abstract: Physical-layer spoofing detection identifies an attacker who impersonates a legitimate wireless user by checking whether transmissions claiming the same identity originate from different locations, using the received signal strength (RSS) measured at several receivers. A recent scheme by Romero et al. makes this approach robust to user movement by combining a neural-network position-change detector (PCD) with community detection on a graph of frames and a fixed threshold on a hand-crafted test statistic. I study two modifications to this scheme. First, I smooth the sequence of PCD scores with a median or Gaussian temporal filter before thresholding. Second, I replace the fixed decision threshold with a supervised classifier operating on six statistical features of the region sequence produced by community detection. On real indoor RSS measurements from 52 transmitter locations, Gaussian filtering raises PCD accuracy from 0.870 to 0.921 when the PCD is trained on 10 locations and from 0.905 to 0.925 when trained on 40. On synthetic attack and no-attack sequences built from the same measurements, evaluated on a held-out test set of 1,679 sequences, the best classifier (XGBoost) reaches an accuracy of 0.949 and an F1 score of 0.940, and random forests reach an AUC of 0.99. The detector separates static attackers almost perfectly but misses about a third of attacks in which two transmitters move simultaneously, which produce region sequences similar to those of a single moving user.
Source:https://arxiv.org/abs/2611.23854PreviousWeighing the Envelope: Component Weight and Arbitration in Envelope-Layer Defenses for a Single LLM AgentNextWhen Do Decision Models Help LLM-Driven Network Discovery?
Related posts
Who called me about stolen ID and loans? Handle it like an unauthenticated security alert
0dev.to
Gemini 4 Models Compared on Reasoning, Coding, and Context
0dev.to
Beyond Frozen Weights: How Test-Time Training (TTT) Turns Inference into Live Learning
0dev.to

Why Turn-Based Voice AI Feels Broken: Building a 52ms Barge-In Loop (HardCall)
0dev.to
Harnesses! The other kind
0dev.to
Apple may be developing its own AI agent that runs your Mac
0TechCrunch