arXiv:2608.01612

A Framework for Stochastic Differentiable Programming

Guillaume Verdon, Leo Tyrpak, Owen Lockwood, Seth Morton, Alexander Neagoe, Anton Sugolov, Ian MacCormack, Mirko Amico

Published Aug 3, 2026• 1 citation

Abstract

We introduce Parametrized Stochastic Circuits (PSCs), a gate-based intermediate representation for programmable stochastic dynamics in which typed local stochastic kernels with tunable parameters compose over explicit binary, categorical, and continuous wires, and \texttt{torx}, an open-source JAX framework for constructing, executing, and differentiating them. PSCs' data types and stochastic kernels are chosen to align closely with the native operations exposed by emerging probabilistic hardware. In this way, stochastic algorithms can be designed directly in terms of the operations the hardware executes natively, so that the energy advantage arising at this level is not lost on mappings that introduce substantial decomposition, communication, or control overhead. We demonstrate the framework on a variety of example applications such as random walks on graphs, discrete diffusion, stochastic graph networks, jump diffusion and Ising sampling. We also report a hardware experiment in which probabilistic bits on the X0 subthreshold CMOS test chip, hosted by the XTR-0 desktop platform, provide physical randomness for Metropolis-Hastings and importance-sampling estimators, yielding estimates consistent with a software pseudorandom baseline.

DOI: 10.48550/arxiv.2608.01612

Related papers

Thermalizing Stochastic Programs

Mirko Amico, Andraž Jelinčič, Colin Oscar Nancarrow +6 more

We present a set of tools for mapping general stochastic programs to thermodynamic hardware designed for energy-efficient stochastic sampling. Given a target stochastic program expressed as a Directed Factor Graph (DFG) of stochastic channels, or equivalently as a Parametrized Stochastic Circuit (PSC), we first introduce a method to approximately compile each factor in the DFG to an Energy-Based Model (EBM) that is native to the hardware. We then analyze how the error of the compiled DFG accumulates from the per-factor errors, and introduce two training refinements, context matching and trajectory-level REINFORCE post-training, which can reduce the residual error left by training each factor in isolation. The \texttt{thermalizers} framework takes a stochastic program expressed in the \texttt{torx} library and replaces its factors with thermodynamic kernels implemented and sampled using the \texttt{thrml} library. We demonstrate it on several example applications, including a market simulator that learns the joint day-to-day dynamics of a panel of financial time series from recorded market history alone, a probabilistic model from mathematical ecology, Gibbs sampling of an EBM the hardware cannot natively express, and a sequential Bayesian design loop over a Gaussian stochastic circuit.

arXiv:2608.01615Aug 3, 20260 citations

First hardware demonstration of a continuous-state probabilistic computer with programmable energy-defined sampling

Chao Fang, Hsuan Yao Tu, Yihan He +3 more

Probabilistic computing has proved useful for combinatorial optimization, but conventional implementations rely on binary stochastic units and sigmoid-based updates. Extending these systems to continuous variables requires binary encoding. Here we introduce a continuous probabilistic computer (CPC) whose continuous probabilistic bits sample bounded states from conditional distributions defined by local energies. Separating random-source generation from conditional sampling provides a route to using physical random sources with different native distributions. A field-programmable gate array (FPGA) prototype updates finite-precision states using digital randomness. The same framework supports product constraints, memory recovery and solution-space exploration. For the tested 7-bit factorization case, fixed-temperature FPGA CPC reduces mean core solving time approximately 33,000-fold relative to the evaluated quadratic unconstrained binary optimization software implementation. The FPGA recovers stored MNIST patterns without pretrained networks, achieving final success rates of 100%. Software CPC reaches 90% reference coverage of the feasible area in approximately 18% of the time required by restarted Adam.

arXiv:2610.31339Oct 22, 20260 citations

A Stochastic-Computing-Based Framework for Quantum Circuit Simulation

Martin Lukáč, Naoya Onizawa, Shinobu Nagayama

Quantum computing is considered a promising platform for next-generation large-scale computing. Although fully scalable quantum computers are still under development, quantum accelerators already offer a practical means to test, evaluate, and better understand quantum computation. Existing simulators provide valuable tools, but they typically rely on deterministic arithmetic and often incur significant computational overhead. To address these limitations, we developed a quantum computing simulator based on bipolar stochastic arithmetic, designed with two goals in mind. First, stochastic computation enables flexible, on-demand control of simulation precision and runtime. Second, stochastic circuits generally require significantly less hardware area than their deterministic counterparts. In our approach, quantum states and operations are encoded using bipolar stochastic bitstreams, allowing quantum circuits to be simulated using lightweight stochastic logic. We implement a representative set of quantum circuits and evaluate the simulator with respect to accuracy and required stream length. The results demonstrate that stochastic simulation yields satisfactory accuracy for small quantum circuits in software, and with improved error-correction techniques, it has the potential to serve as an efficient alternative to conventional quantum simulators.

IEEE International Symposium on Multiple-Valued Logic (ISMVL)May 19, 20261 citation

Hybrid diffusion-optimization for quantum synthesis with continuous native gates

Dajun Guo, Chukun Hu, Xiaolu Su

Abstract Quantum circuit synthesis becomes increasingly challenging as system sizes grow and hardware-native gate sets diversify. Denoising diffusion models have recently been applied to quantum circuit synthesis as an alternative to expensive classical optimization. Existing diffusion-based approaches, however, either operate on discrete gate alphabets or separate gate identities and continuous parameters into different generative streams, which raises the question of whether a unified representation can more naturally encode structure–parameter dependencies. We propose a hybrid diffusion–optimization framework for quantum circuit synthesis with continuous native gates. Taking a trapped-ion-inspired native gate set as a representative example, our single-stream diffusion model operates directly on a continuous-parameter gate set { R X ( θ ), R Y ( θ ), R XX ( θ )} and produces fully parameterized circuits in a single diffusion sampling run. A lightweight gradient-based post-optimization of the continuous parameters, based on the process infidelity, then corrects residual stochastic errors. Trained and evaluated on random three-qubit unitaries, the model generates compact circuits that accurately approximate unseen targets. Under identical training conditions, the continuous-gate formulation achieves substantially lower process infidelity than both random-circuit and a discrete-gate diffusion baseline. Furthermore, applying gradient-based post-optimization to the continuous-gate diffusion output yields a significant additional improvement in fidelity compared with the same model without post-optimization, demonstrating that this refinement step is an effective strategy for correcting residual stochastic errors in the diffusion-generated parameters. These results point to a practical route toward continuous-parameter quantum circuit synthesis in native gate sets.

Physica ScriptaMay 7, 20260 citations

Precision-Elastic Persistent Stochastic Execution for Quantum Circuit Simulation

Naoya Onizawa, Martin Lukáč, Shinobu Nagayama +1 more

Quantum circuit simulation is usually evaluated through final numerical accuracy, while the dynamics of stochastic execution itself are less explicitly characterized. This work presents a precision-elastic persistent stochastic execution framework based on integral stochastic computing (ISC), where execution behavior is controlled by stream length N and ISC multiplicity m, where m represents the number of aggregated stochastic sub-streams per cycle (standard SC corresponds to m = 1). We focus on correlation-sensitive propagation under persistent reuse, and show that this regime produces circuit-dependent stochastic behavior and execution uncertainty patterns that are not captured by stage-wise re-encoded execution alone. To characterize this behavior, we use high-m tail descriptors, including the circuit-dependent coefficient γc, as compact indicators of persistent stochastic sensitivity. Across benchmark circuits, persistent execution exhibits reproducible tail regimes and structured cross-circuit variability, while deterministic reduced-precision baselines are used only as trend-consistency references. We further demonstrate adaptive stochastic precision scheduling, where circuit-dependent (N,m) settings satisfy a target fidelity with reduced stochastic workload. These results position persistent ISC as a configurable stochastic execution framework for analyzing execution-induced uncertainty propagation and correlation-sensitive behavior in quantum circuits.

EntropyAug 1, 20260 citations

Protocols for trainable and differentiable quantum generative modeling

Oleksandr Kyriienko, Annie E. Paine, Vincent E. Elfving

We propose an approach for learning probability distributions as differentiable quantum circuits (DQC) that enable efficient quantum generative modeling (QGM) and synthetic data generation. Contrary to existing QGM approaches, we perform training of a DQC-based model, where data is encoded in a latent space with the proposed phase feature map of exponential capacity. This is followed by a trainable quantum circuit, forming the model. We then map the trained model to the bit basis using a fixed unitary transformation, in this case corresponding to a quantum Fourier transform circuit. It allows fast sampling from parametrized distributions using a single-shot readout. Importantly, latent space training provides models that are automatically differentiable, and we show how samples from solutions of stochastic differential equations (SDEs) can be accessed by solving stationary and time-dependent Fokker-Planck equations with a quantum protocol. Our approach opens a route to multidimensional generative modeling with qubit registers explicitly correlated via a (fixed) entangling layer. In this case quantum computers can offer advantage as efficient samplers, which perform complex inverse transform sampling enabled by the fundamental laws of quantum mechanics. On a technical side the advances are multiple, as we introduce the phase feature map, analyze its properties, and develop frequency-taming techniques that include qubitwise training and feature map sparsification. Published by the American Physical Society 2024

arXiv:2202.08253Physical Review ResearchSep 12, 202414 citations