TQBTHEQUANTBATEMAN
TQB/ learn/ frontier/ neural sdesEN · DARK
Frontierresearchresearch

Neural SDEs

Stochastic differential equations with learned functional components.

Reviewed 2026-08-10TheQuantBateman ResearchReading note
01Intuition

Build the mental model first.

Keep continuous-time stochastic structure while learning flexible drift or diffusion maps from data.

ONE-LINE DEFINITION

Stochastic differential equations with learned functional components.

02Mathematics

Now make it exact.

dXt=μθ(Xt,t)dt+σθ(Xt,t)dWtdX_t=\mu_\theta(X_t,t)dt+\sigma_\theta(X_t,t)dW_t
Notation and units

Decimal rates and volatilities, year-fraction time and continuous compounding unless stated otherwise.

03Assumptions

Every model has a price.

01

Educational conventions are stated explicitly and may simplify market quotation or settlement details.

02

Rates are continuously compounded unless the section says otherwise.

03

Inputs are deterministic in the base model.

“An unstated convention is a future reconciliation break.”— THEQUANTBATEMAN
04Market use

Why a quant cares.

Research-stage modelling with challenges in identifiability, stability and governance.

IntuitionMathematicsImplementationDesk risk
05Desk view
FRONT OFFICE VIEW

The hedge has opinions.

Start with the quote convention, then ask which Frontier risk survives the hedge. A number without its convention is merely well-dressed ambiguity.

Ask Bateman about this model
06Related

Continue through the graph.