Real-world deployment validation of LDAC-Net

Investigate whether LDAC-Net transfers effectively to uncontrolled real-world gas-sensing settings and determine its on-device latency and power consumption during actual deployment.

Background

The paper evaluates LDAC-Net on the controlled SmellNet and eNose-Drift datasets, including cross-dataset transfer between different MOX sensor arrays and drift regimes. However, both corpora were collected offline under controlled conditions. The authors explicitly identify deployment in uncontrolled real-world environments as unresolved and also note that the practical computational and energy costs of the model have not yet been assessed on hardware.

References

Transfer to uncontrolled real-world settings remains untested, and the on-device latency and power of the model, including the small extra cost of the per-window WSAN pass, have yet to be measured in an actual deployment.

— LDAC-Net: A Learnable Multi-Lag Differencing Attention-Convolution Network for Drift-Robust Recognition with Low-Cost MOX Gas Sensors  (2608.25646 - Zhang et al., 26 Aug 2026) in Section 6.3, p. 12