Studies in signal processing, machine learning, and forecasting. Applications span tropical-Pacific climate and financial markets.
Nine sequential reports develop and test an exactly invertible forecasting system; a standalone publication and a shared metric charter sit beside the main sequence.
Three studies move from representation diagnostics to a pre-registered signal scan and then a walk-forward comparison of compact and foundation-model forecasters.
21 documents · Newest first
Signal Processing · Machine Learning · Climate
A methods-validation study of the leakage-safe pipeline running from monthly tropical-Pacific sea-surface temperature, through a complete three-level dual-tree complex wavelet pyramid, into the official Earthformer backbone and back to reconstructed forecast maps.
Signal Processing · Machine Learning · Climate
Nine controlled ablation families on 335 observed training windows: architecture access, trainability boundary, amplitude and skill-aware objectives, and persistence-residual target parameterizations.
Signal Processing · Machine Learning · Climate
A V0–V7 masked-coefficient pretraining ladder over ERA5 and three historical SST reconstruction products, paired with the installation of the project's statistical governance and a two-batch transfer-protocol study.
Signal Processing · Machine Learning · Climate
Supervised forecast-task pretraining across eleven CMIP6 source models, the predeclared transfer consultation that selects PB-1, and the project's single pristine hash-gated read of the sealed 2016–2020 period.
Signal Processing · Machine Learning · Climate
A controlled test of whether upper-300 m ocean heat content adds transferable information to the sea-surface-temperature pathway, against equal-budget twins that keep the pretraining lineage and zero only the subsurface input.
Signal Processing · Machine Learning · Climate
A generative ladder built around a frozen deterministic center: per-coefficient Gaussian heads, spatially correlated latent-noise heads, a conditional diffusion head, and its regularized and simulation-pretrained variants.
Signal Processing · Machine Learning · Climate
Six rungs of explicit index calibration under leave-one-start-year-out folds: bandwise error decomposition, flat EMOS with ECC-Q, a model-analog trajectory baseline, state-conditional variance, extended features, and coherent lowpass field lifting.
Signal Processing · Machine Learning · Climate
Temporally joint diffusion for whole forecast trajectories, a predeclared correction-structure audit that convicts the objective rather than the architecture, and the second and final authorized read of the sealed test period.
Signal Processing · Machine Learning · Climate
Five strictly separate ERA5 pipelines change only the anomaly reduction: fixed, updated-fixed, causal trailing, relative sea-surface temperature, or detrended. Each is freshly retrained under one frozen recipe.
Signal Processing · Machine Learning · Climate
The editorial control document defines the series-wide evidence hierarchy and separates engineering checks from held-out forecast claims. It also fixes the metric identifiers used in every report and the comparability rules that keep incompatible estimators out of a shared column.
Machine Learning · Probabilistic Forecasting · Financial Markets
A completed comparative study of regime-adaptive probabilistic forecasting on the BTCUSDT perpetual surface: a from-scratch multi-resolution encoder against frozen and LoRA-fine-tuned Chronos-T5 and Moirai-MoE lanes, evaluated under strict purged walk-forward splits with rolling conformal recalibration.
Signal Processing · Machine Learning · Climate
Two 14-month tropical-Pacific SST forecast products built in an exactly invertible 37-channel dual-tree complex-wavelet space: a CMIP6-pretrained persistence-residual Earthformer, and a 32-member residual-diffusion ensemble sampled around its frozen center.
Signal Processing · Time–Frequency Analysis · Financial Markets
Two decompositions of the same intraday series disagree about the low-frequency band by more than the cross-asset difference they are being used to measure.
Machine Learning · Probabilistic Forecasting · Financial Markets
A pre-registered scan asks where the full BTCUSDT perpetual store (1m–1d bars, 2019–2026) carries learnable signal across timeframes, horizons, and target definitions. It uses small standardized probes under a purged chronological protocol.
Machine Learning · Probabilistic Forecasting · Financial Markets
This investigation asks where the predictable structure in BTCUSDT actually lives. Prompted by a production quantile head that settled on a near-constant median over frozen Moirai-MoE embeddings, it uses a fast GPU harness over a 158,528-window cached embedding store under purged and embargoed splits.
Statistical Inference · Time Series · Financial Markets
A statistically real effect with a break-even cost of a quarter of a basis point, and a bid–ask-bounce artefact at the shortest horizon.
Natural Language Processing · Statistical Inference · Financial Markets
An event study that returns a null, and residuals that locate the missing reaction in a different window.
Change-Point Detection · Time Series · Financial Markets
Three named detectors scored against a fixed-threshold baseline under a matching rule fixed before any metric was computed. The baseline wins on both axes.
Signal Processing · Forecasting · Financial Markets
A controlled comparison of raw and wavelet features against matched linear and neural baselines for multi-step return forecasting.
Machine Learning · Replication · Financial Markets
A real-time machine-learning pipeline reproduced on public data: the headline Sharpe survives six-factor adjustment and collapses under value weighting.
Machine Learning · Forecasting · Financial Markets
A three-feature model beats chance on next-second mid-price direction and loses money in every cost cell once the spread is paid.