-
Probabilistic Photonic Computing with Chaotic Light
Authors:
Frank Brückerhoff-Plückelmann,
Hendrik Borras,
Bernhard Klein,
Akhil Varri,
Marlon Becker,
Jelle Dijkstra,
Martin Brückerhoff,
C. David Wright,
Martin Salinga,
Harish Bhaskaran,
Benjamin Risse,
Holger Fröning,
Wolfram Pernice
Abstract:
Biological neural networks effortlessly tackle complex computational problems and excel at predicting outcomes from noisy, incomplete data, a task that poses significant challenges to traditional processors. Artificial neural networks (ANNs), inspired by these biological counterparts, have emerged as powerful tools for deciphering intricate data patterns and making predictions. However, convention…
▽ More
Biological neural networks effortlessly tackle complex computational problems and excel at predicting outcomes from noisy, incomplete data, a task that poses significant challenges to traditional processors. Artificial neural networks (ANNs), inspired by these biological counterparts, have emerged as powerful tools for deciphering intricate data patterns and making predictions. However, conventional ANNs can be viewed as "point estimates" that do not capture the uncertainty of prediction, which is an inherently probabilistic process. In contrast, treating an ANN as a probabilistic model derived via Bayesian inference poses significant challenges for conventional deterministic computing architectures. Here, we use chaotic light in combination with incoherent photonic data processing to enable high-speed probabilistic computation and uncertainty quantification. Since both the chaotic light source and the photonic crossbar support multiple independent computational wavelength channels, we sample from the output distributions in parallel at a sampling rate of 70.4 GS/s, limited only by the electronic interface. We exploit the photonic probabilistic architecture to simultaneously perform image classification and uncertainty prediction via a Bayesian neural network. Our prototype demonstrates the seamless cointegration of a physical entropy source and a computational architecture that enables ultrafast probabilistic computation by parallel sampling.
△ Less
Submitted 31 January, 2024;
originally announced January 2024.
-
Shadows and Barriers
Authors:
Martin Brückerhoff,
Martin Huesmann
Abstract:
We show an intimate connection between solutions of the Skorokhod Embedding Problem which are given as the first hitting time of a barrier and the concept of shadows in martingale optimal transport. More precisely, we show that a solution $τ$ to the Skorokhod Embedding Problem between $μ$ and $ν$ is of the form $τ= \inf \{t \geq 0 : (X_t,B_t) \in \mathcal{R}\}$ for some increasing process…
▽ More
We show an intimate connection between solutions of the Skorokhod Embedding Problem which are given as the first hitting time of a barrier and the concept of shadows in martingale optimal transport. More precisely, we show that a solution $τ$ to the Skorokhod Embedding Problem between $μ$ and $ν$ is of the form $τ= \inf \{t \geq 0 : (X_t,B_t) \in \mathcal{R}\}$ for some increasing process $(X_t)_{t \geq 0}$ and a barrier $\mathcal{R}$ if and only if there exists a time-change $(T_l)_{l \geq 0}$ such that for all $l \geq 0$ the equation
$$\mathbb{P}[B_τ \in \cdot , τ\geq T_l] = \mathcal{S}^ν(\mathbb{P}[B_{T_l} \in \cdot , τ\geq T_l])$$
is satisfied, i.e.\ the distribution of $B_τ$ on the event that the Brownian motion is stopped after $T_l$ is the shadow of the distribution of $B_{T_l}$ on this event in the terminal distribution $ν$.
This equivalence allows us to construct new families of barrier solutions that naturally interpolate between two given barrier solutions. We exemplify this by an interpolation between the Root embedding and the left-monotone embedding.
△ Less
Submitted 5 March, 2021;
originally announced March 2021.
-
Instability of Martingale optimal transport in dimension d $\ge$ 2
Authors:
Martin Brückerhoff,
Nicolas Juillet
Abstract:
Stability of the value function and the set of minimizers w.r.t. the given data is a desirable feature of optimal transport problems. For the classical Kantorovich transport problem, stability is satisfied under mild assumptions and in general frameworks such as the one of Polish spaces. However, for the martingale transport problem several works based on different strategies established stability…
▽ More
Stability of the value function and the set of minimizers w.r.t. the given data is a desirable feature of optimal transport problems. For the classical Kantorovich transport problem, stability is satisfied under mild assumptions and in general frameworks such as the one of Polish spaces. However, for the martingale transport problem several works based on different strategies established stability results for R only. We show that the restriction to dimension d = 1 is not accidental by presenting a sequence of marginal distributions on R 2 for which the martingale optimal transport problem is neither stable w.r.t. the value nor the set of minimizers. Our construction adapts to any dimension d $\ge$ 2. For d $\ge$ 2 it also provides a contradiction to the martingale Wasserstein inequality established by Jourdain and Margheriti in d = 1.
△ Less
Submitted 18 January, 2021;
originally announced January 2021.
-
Shadow martingales -- a stochastic mass transport approach to the peacock problem
Authors:
Martin Brückerhoff,
Martin Huesmann,
Nicolas Juillet
Abstract:
Given a family of real probability measures $(μ_t)_{t\geq 0}$ increasing in convex order (a peacock) we describe a systematic method to create a martingale exactly fitting the marginals at any time. The key object for our approach is the obstructed shadow of a measure in a peacock, a generalization of the (obstructed) shadow introduced in \cite{BeJu16,NuStTa17}. As input data we take an increasing…
▽ More
Given a family of real probability measures $(μ_t)_{t\geq 0}$ increasing in convex order (a peacock) we describe a systematic method to create a martingale exactly fitting the marginals at any time. The key object for our approach is the obstructed shadow of a measure in a peacock, a generalization of the (obstructed) shadow introduced in \cite{BeJu16,NuStTa17}. As input data we take an increasing family of measures $(ν^α)_{α\in [0,1]}$ with $ν^α(\mathbb{R})=α$ that are submeasures of $μ_0$, called a parametrization of $μ_0$. Then, for any $α$ we define an evolution $(η^α_t)_{t\geq 0}$ of the measure $ν^α=η^α_0$ across our peacock by setting $η^α_t$ equal to the obstructed shadow of $ν^α$ in $(μ_s)_{s \in [0,t]}$. We identify conditions on the parametrization $(ν^α)_{α\in [0,1]}$ such that this construction leads to a unique martingale measure $π$, the shadow martingale, without any assumptions on the peacock. In the case of the left-curtain parametrization $(ν_{\text{lc}}^α)_{α\in [0,1]}$ we identify the shadow martingale as the unique solution to a continuous-time version of the martingale optimal transport problem.
Furthermore, our method enriches the knowledge on the Predictable Representation Property (PRP) since any shadow martingale comes with a canonical Choquet representation in extremal Markov martingales.
△ Less
Submitted 8 September, 2022; v1 submitted 18 June, 2020;
originally announced June 2020.
