Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

ZK-SenseLM: Verifiable Large-Model Wireless Sensing with Selective Abstention and Zero-Knowledge Attestation

Hasan Akgul, Mari Eplik, Javier Rojas, Aina Binti Abdullah, Pieter van der Merwe

TL;DR

ZK-SenseLM integrates a robust RF/CSI encoder with a policy-grounded decision layer and end-to-end zero-knowledge proofs to yield auditable wireless sensing at the edge. The approach combines masked spectral modeling, phase-consistency regularization, and cross-modal alignment to produce uncertainty-aware decisions that can be verifiably tied to a registered model and bounded time window. A four-stage ZK proving pipeline (feature commitment, version/threshold binding, time-window binding, and decision correctness) enables compact proofs and fast verification, with micro-batching and gateway offloading reducing prover load. Federated and DP-enabled training maintain privacy without weakening verifiability, and a selective abstention mechanism allows safe operation under distribution shift. Across HAR, presence/intrusion, respiration proxies, and RF fingerprinting tasks, ZK-SenseLM demonstrates strong utility and calibration, robustness to perturbations, effective open-set handling, and practical verifiability costs suitable for edge deployments and audits. The work advances practical, privacy-preserving, and accountable wireless sensing by marrying robust RF representations with formal cryptographic guarantees and zero-trust control.

Abstract

ZK-SenseLM is a secure and auditable wireless sensing framework that pairs a large-model encoder for Wi-Fi channel state information (and optionally mmWave radar or RFID) with a policy-grounded decision layer and end-to-end zero-knowledge proofs of inference. The encoder uses masked spectral pretraining with phase-consistency regularization, plus a light cross-modal alignment that ties RF features to compact, human-interpretable policy tokens. To reduce unsafe actions under distribution shift, we add a calibrated selective-abstention head; the chosen risk-coverage operating point is registered and bound into the proof. We implement a four-stage proving pipeline: (C1) feature sanity and commitment, (C2) threshold and version binding, (C3) time-window binding, and (C4) PLONK-style proofs that the quantized network, given the committed window, produced the logged action and confidence. Micro-batched proving amortizes cost across adjacent windows, and a gateway option offloads proofs from low-power devices. The system integrates with differentially private federated learning and on-device personalization without weakening verifiability: model hashes and the registered threshold are part of each public statement. Across activity, presence or intrusion, respiratory proxy, and RF fingerprinting tasks, ZK-SenseLM improves macro-F1 and calibration, yields favorable coverage-risk curves under perturbations, and rejects tamper and replay with compact proofs and fast verification.

ZK-SenseLM: Verifiable Large-Model Wireless Sensing with Selective Abstention and Zero-Knowledge Attestation

TL;DR

ZK-SenseLM integrates a robust RF/CSI encoder with a policy-grounded decision layer and end-to-end zero-knowledge proofs to yield auditable wireless sensing at the edge. The approach combines masked spectral modeling, phase-consistency regularization, and cross-modal alignment to produce uncertainty-aware decisions that can be verifiably tied to a registered model and bounded time window. A four-stage ZK proving pipeline (feature commitment, version/threshold binding, time-window binding, and decision correctness) enables compact proofs and fast verification, with micro-batching and gateway offloading reducing prover load. Federated and DP-enabled training maintain privacy without weakening verifiability, and a selective abstention mechanism allows safe operation under distribution shift. Across HAR, presence/intrusion, respiration proxies, and RF fingerprinting tasks, ZK-SenseLM demonstrates strong utility and calibration, robustness to perturbations, effective open-set handling, and practical verifiability costs suitable for edge deployments and audits. The work advances practical, privacy-preserving, and accountable wireless sensing by marrying robust RF representations with formal cryptographic guarantees and zero-trust control.

Abstract

ZK-SenseLM is a secure and auditable wireless sensing framework that pairs a large-model encoder for Wi-Fi channel state information (and optionally mmWave radar or RFID) with a policy-grounded decision layer and end-to-end zero-knowledge proofs of inference. The encoder uses masked spectral pretraining with phase-consistency regularization, plus a light cross-modal alignment that ties RF features to compact, human-interpretable policy tokens. To reduce unsafe actions under distribution shift, we add a calibrated selective-abstention head; the chosen risk-coverage operating point is registered and bound into the proof. We implement a four-stage proving pipeline: (C1) feature sanity and commitment, (C2) threshold and version binding, (C3) time-window binding, and (C4) PLONK-style proofs that the quantized network, given the committed window, produced the logged action and confidence. Micro-batched proving amortizes cost across adjacent windows, and a gateway option offloads proofs from low-power devices. The system integrates with differentially private federated learning and on-device personalization without weakening verifiability: model hashes and the registered threshold are part of each public statement. Across activity, presence or intrusion, respiratory proxy, and RF fingerprinting tasks, ZK-SenseLM improves macro-F1 and calibration, yields favorable coverage-risk curves under perturbations, and rejects tamper and replay with compact proofs and fast verification.

Paper Structure

This paper contains 140 sections, 19 equations, 4 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. From Device-Free RF Sensing to Learning-Centric Pipelines
  4. Standardization and Systemization: IEEE 802.11bf and Beyond
  5. Healthcare, Physiology, and Human-Centric Ambient Intelligence
  6. RF Fingerprinting, Identity, and Provenance
  7. Attacks and Robustness in Wireless Sensing
  8. Cross-Modal and Through-Wall Perception
  9. Edge Learning and Systems for Scalable Deployment
  10. Zero Trust for IoT and Access Control Context
  11. Zero Knowledge and Verifiable Machine Learning
  12. Privacy, Ethics, and Policy Grounding
  13. Open Problems and Positioning of This Work
  14. Relationship to Our Prior and Contemporary Work
  15. Summary
  16. ...and 125 more sections

Figures (4)

  • Figure 1: System overview
  • Figure 2: Latency–energy Pareto under batching and gateway offload. Micro-batching (B=1,4,8,16) bends the curve favorably; offloading proving to a gateway reduces on-device energy while keeping verification fast.
  • Figure 3: Coverage–risk trade-offs across models. ZK-SenseLM traces curves toward the lower-left region (lower selective risk at comparable coverage), especially under mild distribution shift.
  • Figure 4: Reliability diagram for ZK-SenseLM: calibration remains close to the diagonal on clean data and degrades gracefully under domain shift.