Research archive
Thirteen open research pages covering distributional neurons, variational computation, Transformer integration, internal activity, latent-state readout and uncertainty-aware control.
August 10, 2026 (v1) · Zenodo · Preprint
Separates Bayesianity, variationality, stochasticity and distribution-valued computation; formalizes a unit-level variational-neuron specification criterion and tests it through controlled EVE experiments, including a fresh exact-capacity 50-seed comparison.
2026 (v1) · Research Square · Research Article
EVE makes uncertainty a neuron-level computational primitive and evaluates it from a single-unit microscope to dense networks and Transformer feed-forward computation.
July 14, 2026 (v1) · Research Square · Research Article
Develops the architectural thesis that distributional representation can reside inside the neuron rather than only in weights, outputs or global latent variables.
June 30, 2026 (v1) · Research Square · Research Article
Freezes the learned EVE posterior and varies only the readout rule to test whether a neuron-level latent state carries predictive information beyond its mean.
May 4, 2026 (v1) · Research Square · Method Article
Defines a local variational neuron and a measurement protocol for KL activity, posterior mean energy, out-of-range units and optional autoregressive persistence.
2026-02-20 · arXiv · Open
Introduces a computational unit formulated as a local variational primitive with a prior, amortized posterior and local ELBO.
2026-03-14 · arXiv · Open
Maps the effects of latent dimensionality, local capacity control and temporal persistence on the measurable operating regime of EVE.
2026-03-30 · arXiv · Open
Integrates EVE units into Transformer feed-forward computation and evaluates predictive quality, calibration and internal uncertainty signals.
2026-04-14 · arXiv · Open
Explores internal uncertainty as an operational signal for regulation, checkpoint retention and inference-time intervention.
2026-07-27 · Zenodo · Open
EVE represents selected hidden activations as input-conditioned distributions, making internal uncertainty observable for reliability monitoring and uncertainty-aware computation.
2026-07-29 · Zenodo · Open
EVE exposes unit-level posterior activity during language-model computation and compares predictive quality, calibration and tail risk with deterministic, dropout and ensemble references.
2026-07-28 · Zenodo · Open
This paper introduces a measurement panel for local posterior activity and relates internal probabilistic statistics to predictive behavior, calibration and robustness.
2026-07-29 · Zenodo · Open
A controlled five-seed study evaluates EVE as a small-scale mechanism for measurable internal distributional activity and uncertainty-aware readouts.