Whitebox LM · M13 · preprint

The state-compression ladder

How much of learning a dynamical neuron can be decoupled from the neuron itself?

PedalCore / Whitebox LM Hodgkin–Huxley surrogate study code: github.com/PedalCore/state-compression-ladder

We ask how many dynamical degrees of freedom a neuron's computation actually requires, and — more sharply — which parts of learning a self-coupled dynamical system can be parallelised and which cannot. Using a Hodgkin–Huxley teacher whose four state variables are known, we build an instrument (a compression ladder) and drive it through eleven preregistered rounds, an external code audit, and a single-shot held-out evaluation.

Two results survive. (1) A compact primitive: a frozen, parallel-trained vector field plus a ~34-parameter, single-scalar bounded correction reaches ≈0.88 held-out spike-timing F1 with the correct anodal-break rebound and near-perfect firing-rate curve. (2) A boundary: local dynamics, state representation, and even own-trajectory generation all parallelise — but high-fidelity closed-loop learning was reached only by long-horizon sequential credit assignment. Freshness, short-block coupling, and every offline distribution/label/objective correction we tried fell short.

The neuron is usually the cheap part of a network — a weight and a nonlinearity. This study starts from the opposite premise: that a neuron is a small stateful dynamical system, and asks what that state is for, how little of it is needed, and — the question that turned out to matter most — how much of the difficulty of learning it is irreducibly sequential.

Working against a known-dimensional teacher lets every claim be an instrument reading rather than an assertion. What follows is organised as the eight questions the programme actually answered, in order.

Q1 How much state is needed?

We built a collision diagnostic: two representations that map to nearby points must demand consistent futures, or the representation is insufficient. Applied to learned encoders of the HH state, sufficiency drops sharply and then plateaus — a knee at roughly three effective coordinates, consistent with the classical fast-m reduction (V, h, n) rather than the full four. The instrument measures a minimal realisation for a model class under an input distribution, not a canonical dimension — an important scoping that recurs throughout.

Q2 Can voltage history recover hidden state?

Instantaneous voltage is ambiguous (collision ratio 0.35). A rolling 12 ms window of five voltage samples reaches the same sufficiency floor as the full-state encoders (0.14). Informationally, recent history substitutes for the hidden gating variables — a measured Takens-style result on the exact data.

Q3 Can local dynamics be learned in parallel?

Yes, and cheaply. Training a vector field on independent (state, input) → dV/dt samples is roughly 15× faster than sequential rollout training and reproduces the firing-rate curve and the anodal-break rebound. An integrator-isolation control (RK4 vs Euler on identical weights) came back null: discretisation is not the bottleneck.

Q4 Why does event fidelity plateau?

The parallel-trained field sits at ≈0.74 spike F1 despite excellent waveform statistics. Purpose-built metrics found the cause, and it was none of the obvious candidates. Spike-miss F1 is flat across the rollout — not accumulating drift. Misses are episodic: P(miss | previous miss) = 0.69 against a base rate of 0.29. Episodes are entered rarely, then healed by the input in ≈5.4 ms. Initiation is a slow inter-spike under-excitability drift (n rising, h falling) visible ~5 ms before the missed spike. The plateau is an event-decision problem, not a stability failure.

Q5 Can a tiny bounded correction solve it?

Freeze the good field; add a small recurrent correction, hard-bounded to the empirically located drift/decision voltage band so it cannot rewrite the physics. It crosses the original F1>0.9 gate that every prior model failed. The correction is inspectable: its magnitude is near-zero everywhere except the drift and decision bands — the credit state doing exactly the job the causal anatomy predicted.

modelcorr. paramsstate/neuronheld-out F1 (s0/s1)f-I (Hz)rebound
trust-hybrid, kc=839680.879 / 0.8880.5 / 0.3✓ / ✓
trust-hybrid, kc=1321 scalar0.861 / 0.8360.6 / 9.9✓ / ✓
observable parallel baseline≈0.264
sequential GRU (long-horizon)93080.86914.5
Held-out performance (single-shot, locked corpus, new RNG seed). The frozen field plus a bounded correction reaches ≈0.88 spike-timing F1 with correct rebound; a single scalar of state and 32 shared parameters holds ≈0.85. Numbers are reported once on a corpus generated after all code fixes and never used for any decision (dev↔test gap ≈0.02–0.06).

Q6 Does the correction really need much state?

No. Compressing the corrector from 396 parameters to 32 parameters and one scalar of runtime state costs almost nothing. And in complete coordinates, a matched-parameter static correction performs the same as a recurrent one — so recurrence, at matched size, adds nothing when the coordinates already expose the state.

Q7 Then when is state actually required?

Exactly when the coordinates hide it. The sharpest contrast in the study: given the same 12 ms voltage history, a static map fails utterly — 0.004 F1 even with oracle-perfect history — while a recurrent model reaches 0.87+. Identical information, opposite outcomes. Internal state is a computational resource for reconstructing hidden state under partial observation, not a universal enhancer. A selective (Mamba-style) affine accumulator matched the GRU at a fraction of the training cost — the right machinery is cheap and parallel-trainable.

Q8 Why does the observable parallel model still fall short — and what remains?

Everything else was exonerated by preregistered control: state amount, state form, capacity, field-error magnitude, error placement, label alignment, and the state-shaping objective (a JEPA-style latent-future target changed nothing). The observable ceiling is ≈0.264. What remains is the coupling between a model's parameters and the trajectory distribution those parameters induce.

We priced it directly. Offline training on the model's own distribution — at any mixture up to 100% — collapsed. A block-coupling experiment (in the style of test-time training) then separated the two candidate resources with a built-in invariant: at block size 1, where no through-time credit exists, the detached and backprop-through-time arms are byte-identical, validating the harness. Above it, neither fresh-state resynchronisation nor short-block through-time credit crossed the gap. Only long-horizon sequential training ever did.

The block-coupling discriminator (held-out F1 vs coupling block size). Same generated block, same targets, same budget for both arms; the only difference is whether gradient flows through earlier self-generated states. The b=1 invariant holds exactly (bars coincide). Neither arm rises meaningfully above the ≈0.24 baseline at any block size — short-horizon coupling of either kind is insufficient. The dashed line marks long-horizon sequential training (0.869), the only regime that crosses.
Most of learning a self-coupled dynamical primitive decouples and parallelises.
The irreducible remainder is long-horizon contemporaneous credit — the feedback between parameter updates and the distribution they induce.
What parallelises, and what does not. Each stage of learning the primitive was tested for whether it can be trained on independent / detached / offline data. Only the closed-loop credit path resisted every decoupling attempt.

§ What this buys, in hardware terms

The decomposition has a clean deployment reading. The dynamics are trained once, paying the one-time sequential cost. Thereafter the primitive freezes: a shared ~34-parameter law plus one scalar of state per neuron. Downstream networks train connectivity and readouts around the frozen cell with no coupling problem at all, because a frozen field's rollout distribution no longer moves. Bounded per-population adaptation stays parallel; only rewriting the core dynamics reawakens the sequential cost. Coupled training is a cost you pay to mint the primitive — not one you repay on every use.

Reproducibility & validation

An independent code-path review found indexing (an off-by-one in the flagship target), a chunk-boundary regularisation error, current-inconsistent phase labels, and a data-budget confound in the coupling sweep. All were corrected and the affected experiments rerun. The corrected runs preserved every headline conclusion — and tightened one: the single-scalar corrector improved with correct targets. The observable baseline was corrected upward (≈0.20 → ≈0.264). A closed-form scan implementation was unit-tested against an explicit recurrence to 1×10⁻⁷ in value and gradient. Final performance was then measured once on a freshly generated locked corpus. Repeatedly-scored figures are labelled development metrics; the held-out numbers above are the reported test performance.

Where this goes next

The primitive is proven; the next question is whether richer dynamics buy computation per unit hardware. That becomes a measurement — Information Processing Capacity (linear and nonlinear orders) per state-bit, per gate, per picojoule — comparing the frozen primitive against LIF, AdEx, and a matched timescale-spectrum control. Beyond it: a four-neuron re-taskability test (fixed weights, changing only a context vector) as the bridge toward synchronisation-as-computation. Connectivity cost, which scales roughly quadratically, is measured separately from neuron cost — the real hardware argument may be that a richer neuron needs fewer neurons and therefore far fewer connections.


A whitebox LM research note. Full lab notebook — including preregistrations, failed recipes, published corrections, and every intermediate result in chronological order — plus a clean reproducibility path (repro/) at github.com/PedalCore/state-compression-ladder. Eleven development rounds and this synthesis were shaped throughout by an external reviewer's critiques; the audit above is theirs. Part of soundlark.studio — see also the white-box lab and thinking in ticks.

AI-generated research. The experiments, code and write-ups on this site were produced by AI agents (Claude and Codex) working semi-autonomously, directed at a high level by a human. Nothing here is peer reviewed and none of it is affiliated with an academic institution. Measurements are reported as taken — including the nulls, the caveats and the corrections — but the work has had no external review.