AI surrogates

The stochastic-rs-ai crate ships neural-network volatility surrogates for fast pricing / calibration. The crate is feature-gated behind ai; it pulls in candle-core for the inference runtime.

Status: experimental. The surrogates are useful for warm-starting a calibrator or for sub-millisecond price approximations in a feedback loop, not as the primary pricing surface for production trading.

Models shipped

How they work

Each surrogate maps (model parameters, strike grid, expiry grid) → implied-vol surface. The surrogate is a small MLP (3–5 hidden layers) trained offline on a Fourier-pricer-generated dataset. At inference time the pipeline is:

1. Normalise the inputs via BoundedScaler / StandardScaler.
2. Forward-pass through the MLP.
3. Denormalise the output to a flat IV grid.
4. Hand the grid to ImpliedVolSurface::from_flat_iv_grid for downstream pricing or smile / skew analytics.

Example — Heston surrogate inference

use stochastic_rs::ai::volatility::heston::HestonNn;
use stochastic_rs::quant::vol_surface::implied::ImpliedVolSurface;

let nn = HestonNn::<f64>::load("heston.safetensors")?;

let params = HestonParams {
    kappa: 2.0, theta: 0.04, sigma: 0.3, rho: -0.5, v0: 0.04,
};
let strikes  = vec![80.0, 90.0, 100.0, 110.0, 120.0];
let expiries = vec![0.25, 0.5, 1.0, 2.0];

// Inference returns a flat (strikes × expiries) IV grid in <1ms
let iv_grid = nn.predict_surface(&params, &strikes, &expiries);

let surface = ImpliedVolSurface::<f64>::from_flat_iv_grid(
    &strikes, &expiries, &iv_grid, /* spot */ 100.0,
);
println!("ATM 1y IV = {:.4}", surface.iv_at(100.0, 1.0));

Training round-trip

Every surrogate ships a train_save_load_<model> integration test that:

1. Loads a known training-set archive
2. Trains the MLP for a few epochs
3. Saves to safetensors
4. Reloads and asserts inference parity

This guards the on-disk format from silent regressions.

Adding a surrogate

See the vol-surrogate-nn SKILL. It covers StochVolModelSpec, BoundedScaler / StandardScaler conventions, gzip-npy training-set loading, the round-trip test, and predict_surface integration with ImpliedVolSurface::from_flat_iv_grid.