Aionoscope: Debugging Latent-State Accessibility in Time-Series Representations
2026-07-01 • Machine Learning
Machine LearningArtificial Intelligence
AI summaryⓘ
The authors created Aionoscope, a tool to check if time-series models keep important details about signals like timing or frequency in their hidden representations. Unlike usual evaluations that only look at overall prediction accuracy, Aionoscope tests whether the model’s internal data really shows fine details about the signal. They tested 37 models and found that while models can usually tell what kind of signal is present, they often fail to capture exact timing or amplitude information. This shows that a model may seem good at recognizing signals but miss critical details needed for deeper analysis.
time-series modelslatent staterepresentation learningevent timingphaseamplitudefrequencylinear probesynthetic datadiagnostic tools
Authors
Alexander Chemeris, Ming Jin, Randall Balestriero
Abstract
Time-series models are often evaluated by what they can forecast or classify, but those scores do not show whether their representations preserve the process state a user may want to inspect: event timing, phase, amplitude, frequency, or regime variables. We introduce Aionoscope, a generator-based diagnostic tool for debugging latent-state accessibility in frozen time-series representations. Aionoscope separates process generation from observation rendering, producing seeded synthetic streams with exact categorical and dense labels across mixture complexity and nuisance variation. We instantiate Aionoscope as Primitive Process Mixtures and evaluate 37 model-plus-adapter systems with a common pooled linear-probe protocol. The main result is a mismatch between coarse and fine-grained accessibility. Most systems make component presence easy to recover, but expose dense process state much less reliably: the highest observed dense-probe row reaches 0.689 mean masked $R^2$, while a dense-feature oracle reaches 0.999. This is the failure mode Aionoscope is designed to surface: a representation can look informative at the level of "what kind of signal is present" while hiding the timing, phase, amplitude, frequency, or regime variables needed for debugging.