LeNEPA
A small causal encoder that learns time-series representations by predicting its own next latent token — no augmentations, no bidirectional attention, no hand-built difference branch.
Paper: LeNEPA: No-Augmentation Next-Latent Prediction for Time-Series Representation Learning · poster at KDD 2026 MILETS
Aionoscope
Generate signals with known latent factors, freeze a foundation model, and read how it arranges phase, event time, and trend into geometric manifolds — across layers, scale, and training.
Paper: Aionoscope: Debugging Latent-State Accessibility in Time-Series Representations · oral at KDD 2026 MILETS
Open Aionoscope →The Instrument
The generator, the frozen-model probe harness, and the manifold metrics — isometry, neighbourhood, projection, fiber — that score how faithfully a representation lays a factor out.
Read articleThe Shape of a Signal
How phase, event time, and trend become geometric manifolds in the activations of a frozen time-series foundation model.
Read articleFantastic Manifolds and How to Catch Them
The interactive slide deck from the KDD 2026 MILETS oral: the walk from linear probes to representation geometry on live charts driven by real Aionoscope data, with the generator knobs left in for you to turn.
Open slide deck