Fusion 2026: interpretable sensing and temporal AI
I took part in the 29th International Conference on Information Fusion (Fusion 2026) in Trondheim, Norway, held on 23–26 June 2026. Our two conference papers address interpretability in wireless sensing and temporal foundation models.
Two conference papers
Towards Causally Interpretable Wi-Fi CSI-Based Human Activity Recognition with Discrete Latent Compression and LTL Rule Extraction
CHARL-TRE connects discrete latent compression, causal discovery and deterministic Linear Temporal Logic rules to make activity classification from Wi-Fi signals more inspectable. The evaluation remains preliminary and bounded.
Preliminary Insights in Chronos Frequency Data Understanding and Reconstruction
This study uses controlled sinusoidal inputs, probes and concept-erasure interventions to examine how Chronos represents frequency information across its decoder layers, and where that information becomes less accessible.