-
A Better Match for Drivers and Riders: Reinforcement Learning at Lyft
Authors:
Xabi Azagirre,
Akshay Balwally,
Guillaume Candeli,
Nicholas Chamandy,
Benjamin Han,
Alona King,
Hyungjun Lee,
Martin Loncaric,
Sebastien Martin,
Vijay Narasiman,
Zhiwei,
Qin,
Baptiste Richard,
Sara Smoot,
Sean Taylor,
Garrett van Ryzin,
Di Wu,
Fei Yu,
Alex Zamoshchin
Abstract:
To better match drivers to riders in our ridesharing application, we revised Lyft's core matching algorithm. We use a novel online reinforcement learning approach that estimates the future earnings of drivers in real time and use this information to find more efficient matches. This change was the first documented implementation of a ridesharing matching algorithm that can learn and improve in rea…
▽ More
To better match drivers to riders in our ridesharing application, we revised Lyft's core matching algorithm. We use a novel online reinforcement learning approach that estimates the future earnings of drivers in real time and use this information to find more efficient matches. This change was the first documented implementation of a ridesharing matching algorithm that can learn and improve in real time. We evaluated the new approach during weeks of switchback experimentation in most Lyft markets, and estimated how it benefited drivers, riders, and the platform. In particular, it enabled our drivers to serve millions of additional riders each year, leading to more than $30 million per year in incremental revenue. Lyft rolled out the algorithm globally in 2021.
△ Less
Submitted 13 November, 2023; v1 submitted 20 October, 2023;
originally announced October 2023.
-
Complete mathematical characterization of two simple toggle-switch biological systems
Authors:
Xavier Richard,
Benoît Richard,
Christian Mazza,
Jan Roelof van der Meer
Abstract:
Prokaryotic gene expression is dynamic and noisy, and can lead to phenotypic variations. Previous work has shown that the selection of such variation can be modeled through basic reaction networks described by O.D.E. displaying a bistable behavior. While previous mathematical studies have shown that mono- or bistable behavior depends on the rate of the reactions in the system, no analytical soluti…
▽ More
Prokaryotic gene expression is dynamic and noisy, and can lead to phenotypic variations. Previous work has shown that the selection of such variation can be modeled through basic reaction networks described by O.D.E. displaying a bistable behavior. While previous mathematical studies have shown that mono- or bistable behavior depends on the rate of the reactions in the system, no analytical solution of the curves delineating the actual parameter conditions that result in mono or bistability has so far been provided. In this work we provide the first explicit analytical solution for the boundary curve that separates the parameter space defining domains where double positive and double negative feedback loops become bistable.
△ Less
Submitted 7 April, 2023; v1 submitted 6 April, 2023;
originally announced April 2023.
-
A situated agent-based model to reveal irrigators' options behind their actions under institutional arrangements in Southern France
Authors:
Bastien Richard,
Bruno Bonté,
Olivier Barreteau,
Isabelle Braud
Abstract:
There has been little exploration of the explicit simulation of the set of options of actors in agent-based models and its evolution over time. This study proposes to use affordances as intermediate entities between agents' environment and agent actions. We illustrated the approach on a typical gravity-fed network in the South-East of France to explore how the abandonment of traditional sharing of…
▽ More
There has been little exploration of the explicit simulation of the set of options of actors in agent-based models and its evolution over time. This study proposes to use affordances as intermediate entities between agents' environment and agent actions. We illustrated the approach on a typical gravity-fed network in the South-East of France to explore how the abandonment of traditional sharing of water changes the irrigators' options to irrigate. We simulated a typical dry year irrigation season under two institutional arrangements (i.e. traditional coordination through daily slots and its abandonment). Simulation results are consistent with field surveys, and reveal an increase in the number of internal conflicts among irrigators as the counterpart of the abandonment of traditional sharing of water. They also highlight the consequences of the heterogeneity of the irrigators' interests within the collective institution. The sensitivity analysis of the model allowed identification of optimal modalities of coordination, and a potential compromise between past and current institutional arrangements. The key benefits of using affordances in ABM lie in the study of their population dynamics for characterizing the interaction situations between actors and their environment and for better understanding the model dynamics.
△ Less
Submitted 14 October, 2022;
originally announced November 2022.
-
Entanglement masquerading in the CMB
Authors:
Arsalan Adil,
Andreas Albrecht,
Rose Baunach,
R. Holman,
Raquel H. Ribeiro,
Benoit J. Richard
Abstract:
The simplest single-field inflation models capture all the relevant contributions to the patterns in the Cosmic Microwave Background (CMB) observed today. A key assumption in these models is that the quantum inflationary fluctuations that source such patterns are generated by a particular quantum state -- the Bunch-Davies (BD) state. While this is a well-motivated choice from a theoretical perspec…
▽ More
The simplest single-field inflation models capture all the relevant contributions to the patterns in the Cosmic Microwave Background (CMB) observed today. A key assumption in these models is that the quantum inflationary fluctuations that source such patterns are generated by a particular quantum state -- the Bunch-Davies (BD) state. While this is a well-motivated choice from a theoretical perspective, the question arises of whether current data can rule out other, also well motivated, choices of states. In particular, as we previously demonstrated in arXiv:2104.13410 [hep-th], entanglement is naturally and inevitably dynamically generated during inflation given the presence of a "rolling" spectator scalar field -- and the resulting entangled state will yield a primordial power spectrum with potentially measurable deviations compared to the canonical BD result. For this work we developed a perturbative framework to allow a systematic exploration of constraints on (or detection of) entangled states with Planck CMB data using Monte Carlo techniques. We have found that most entangled states accessible with our framework are consistent with the data. One would have to expand the framework to allow a greater variety of entangled states in order to saturate the Planck constraints and more systematically explore any preferences the data may have among the different possibilities.
△ Less
Submitted 2 July, 2023; v1 submitted 20 November, 2022;
originally announced November 2022.
-
An optical neural network using less than 1 photon per multiplication
Authors:
Tianyu Wang,
Shi-Yuan Ma,
Logan G. Wright,
Tatsuhiro Onodera,
Brian Richard,
Peter L. McMahon
Abstract:
Deep learning has rapidly become a widespread tool in both scientific and commercial endeavors. Milestones of deep learning exceeding human performance have been achieved for a growing number of tasks over the past several years, across areas as diverse as game-playing, natural-language translation, and medical-image analysis. However, continued progress is increasingly hampered by the high energy…
▽ More
Deep learning has rapidly become a widespread tool in both scientific and commercial endeavors. Milestones of deep learning exceeding human performance have been achieved for a growing number of tasks over the past several years, across areas as diverse as game-playing, natural-language translation, and medical-image analysis. However, continued progress is increasingly hampered by the high energy costs associated with training and running deep neural networks on electronic processors. Optical neural networks have attracted attention as an alternative physical platform for deep learning, as it has been theoretically predicted that they can fundamentally achieve higher energy efficiency than neural networks deployed on conventional digital computers. Here, we experimentally demonstrate an optical neural network achieving 99% accuracy on handwritten-digit classification using ~3.2 detected photons per weight multiplication and ~90% accuracy using ~0.64 photons (~$2.4 \times 10^{-19}$ J of optical energy) per weight multiplication. This performance was achieved using a custom free-space optical processor that executes matrix-vector multiplications in a massively parallel fashion, with up to ~0.5 million scalar (weight) multiplications performed at the same time. Using commercially available optical components and standard neural-network training methods, we demonstrated that optical neural networks can operate near the standard quantum limit with extremely low optical powers and still achieve high accuracy. Our results provide a proof-of-principle for low-optical-power operation, and with careful system design including the surrounding electronics used for data storage and control, open up a path to realizing optical processors that require only $10^{-16}$ J total energy per scalar multiplication -- which is orders of magnitude more efficient than current digital processors.
△ Less
Submitted 27 April, 2021;
originally announced April 2021.
-
Does Planck Actually "See" the Bunch-Davies State?
Authors:
Rose Baunach,
Nadia Bolis,
R. Holman,
Stacie Moltner,
Benoit J. Richard
Abstract:
To what extent can the Planck satellite observations be interpreted as confirmation of the quantum part of the inflationary paradigm? Has it "seen" the Bunch-Davies state? We compare and contrast the Bunch-Davies interpretation with one using a so-called entangled state in which the fluctuations of a spectator scalar field are entangled with those of the metric perturbations $ζ$. We first show how…
▽ More
To what extent can the Planck satellite observations be interpreted as confirmation of the quantum part of the inflationary paradigm? Has it "seen" the Bunch-Davies state? We compare and contrast the Bunch-Davies interpretation with one using a so-called entangled state in which the fluctuations of a spectator scalar field are entangled with those of the metric perturbations $ζ$. We first show how a spectator scalar field $Σ$, with an expectation value $σ(t)$ that evolves in time, will generically generate such a state. We then use this state to compute the power spectrum $P_ζ(k)$ and thence the temperature anisotropies $C_l$ in the Cosmic Microwave Background (CMB). We find interesting differences from the standard calculations using the Bunch-Davies (BD) state. We argue that existing data may already be used to place interesting bounds on this class of deviations from the BD state and that, for some values of the parameters of the state, the power spectra may be consistent with the Planck satellite data.
△ Less
Submitted 13 July, 2021; v1 submitted 27 April, 2021;
originally announced April 2021.
-
BESS Optimal Sizing Methodology -Degree of Impact of Several Influencing Factors
Authors:
Benoît Richard,
Xavier Le Pivert,
Yves-Marie Bourien
Abstract:
Battery Energy Storage Systems (BESS) are more and more competitive due to their increasing performances and decreasing costs. Although certain battery storage technologies may be mature and reliable from a technological perspective, with further cost reductions expected, the economic concern of battery systems is still a major barrier to be overcome before BESS can be fully utilized as a mainstre…
▽ More
Battery Energy Storage Systems (BESS) are more and more competitive due to their increasing performances and decreasing costs. Although certain battery storage technologies may be mature and reliable from a technological perspective, with further cost reductions expected, the economic concern of battery systems is still a major barrier to be overcome before BESS can be fully utilized as a mainstream storage solution in the energy sector. Since the investment costs for deploying BESS are significant, one of the most crucial issues is to optimally size the battery system to balance the trade-off between using BESS to improve energy system performance and to achieve profitable investment. Determining the optimal BESS size for a specific application is a complex task because it relies on many factors, depending on the application itself, on the technical characteristics of the battery system and on the business model framework. This paper describes a generic simulation-based analytical method which has been developed to determine the BESS optimal size by taking into account both the application and the storage performance over its lifetime. Its implementation and the associated results are presented for two different BESS use cases: A smoothing and peak shaving application for PV injection and an off-grid hybrid microgrid case. In order to provide a better understanding of the most influencing drivers to consider during a BESS sizing procedure, several sensitivity analyses have been carried out on these two illustrative cases. The use of comparative scenarios led to quantify the degree of impact on optimal sizing results of several factors among the following topics: control strategy, forecast quality, degradation of battery performance due to ageing, precision of technical modelling.
△ Less
Submitted 13 November, 2020;
originally announced November 2020.
-
Generating Entangled Inflationary Quantum States
Authors:
R. Holman,
Benoit J. Richard
Abstract:
Entangled inflationary quantum states offer an interesting alternative to the standard Bunch-Davies vacuum. While they can be viewed from the point of view of a theory of effective initial states, it has been unclear exactly how such states might be generated. Using a model involving two scalar fields $Φ,Σ$ we show that if {\em both} fields have non-zero time dependent expectation values $φ,σ$ res…
▽ More
Entangled inflationary quantum states offer an interesting alternative to the standard Bunch-Davies vacuum. While they can be viewed from the point of view of a theory of effective initial states, it has been unclear exactly how such states might be generated. Using a model involving two scalar fields $Φ,Σ$ we show that if {\em both} fields have non-zero time dependent expectation values $φ,σ$ respectively, and if the interaction potential $V(Φ,Σ)$ when evaluated at these expectation values has $\partial_φ\partial_σV(φ, σ)\neq 0$, this can serve as a source term for the entanglement kernel describing the state within the Gaussian approximation around the expectation values. We also show how cubic interactions between the metric perturbation $ζ$ and a scalar $Σ$ can be used to generate a source term for entanglement.
△ Less
Submitted 1 February, 2019;
originally announced February 2019.
-
A Spinodal Solution to Swampland Inflationary Constraints
Authors:
R. Holman,
B. Richard
Abstract:
We show how the inclusion of non-perturbative, {\em dynamical} quantum effects on the evolution of the inflaton can allow for an inflationary phase that is both consistent with cosmological constraints {\em and} avoids the problems associated with the so-called swampland conditions. In particular, for concave potentials such as those preferred by the Planck data, spinodal instabilities associated…
▽ More
We show how the inclusion of non-perturbative, {\em dynamical} quantum effects on the evolution of the inflaton can allow for an inflationary phase that is both consistent with cosmological constraints {\em and} avoids the problems associated with the so-called swampland conditions. In particular, for concave potentials such as those preferred by the Planck data, spinodal instabilities associated with tachyonic masses for long wavelength modes induce a second round of inflation, which in essence decouples the tree-level potential from the inflationary phase. We illustrate these points using natural inflation as an example.
△ Less
Submitted 13 May, 2019; v1 submitted 14 November, 2018;
originally announced November 2018.
-
Spinodal Backreaction During Inflation and Initial Conditions
Authors:
Benoit J. Richard,
McCullen Sandora
Abstract:
We investigate how long wavelength inflationary fluctuations can cause the background field to deviate from classical dynamics. For generic potentials, we show that, in the Hartree approximation, the long wavelength dynamics can be encapsulated by a two-field model operating in an effective potential. The latter is given by a simple Gaussian integral transformation of the original inflationary pot…
▽ More
We investigate how long wavelength inflationary fluctuations can cause the background field to deviate from classical dynamics. For generic potentials, we show that, in the Hartree approximation, the long wavelength dynamics can be encapsulated by a two-field model operating in an effective potential. The latter is given by a simple Gaussian integral transformation of the original inflationary potential. We use this new expression to study backreaction effects in quadratic, hilltop, flattened, and axion monodromy potentials. We find that the net result of the altered dynamics is to slightly modify the spectral tilt, drastically decrease the tensor-to-scalar ratio, and to effectively smooth over any features of the potential, with the size of these deviations set by the initial value of power in large scale modes and the shape of the potential during the entire evolution.
△ Less
Submitted 13 April, 2015;
originally announced April 2015.
-
Spinodal Instabilities and Super-Planckian Excursions in Natural Inflation
Authors:
Andreas Albrecht,
R. Holman,
Benoit J. Richard
Abstract:
Models such as Natural Inflation that use Pseudo-Nambu-Goldstone bosons (PNGB's) as the inflaton are attractive for many reasons. However, they typically require trans-Planckian field excursions $ΔΦ>M_{\rm Pl}$, due to the need for an axion decay constant $f>M_{\rm Pl}$ to have both a sufficient number of e-folds {\em and} values of $n_s,\ r$ consistent with data. Such excursions would in general…
▽ More
Models such as Natural Inflation that use Pseudo-Nambu-Goldstone bosons (PNGB's) as the inflaton are attractive for many reasons. However, they typically require trans-Planckian field excursions $ΔΦ>M_{\rm Pl}$, due to the need for an axion decay constant $f>M_{\rm Pl}$ to have both a sufficient number of e-folds {\em and} values of $n_s,\ r$ consistent with data. Such excursions would in general require the addition of all other higher dimension operators consistent with symmetries, thus disrupting the required flatness of the potential and rendering the theory non-predictive. We show that in the case of Natural Inflation, the existence of spinodal instabilities (modes with tachyonic masses) can modify the inflaton equations of motion to the point that versions of the model with $f<M_{\rm Pl}$ can still inflate for the required number of e-folds. The instabilities naturally give rise to two separate phases of inflation with different values of the Hubble parameter $H$ one driven by the zero mode, the other by the unstable fluctuation modes. The values of $n_s$ and $r$ typically depend on the initial conditions for the zero mode, and, at least for those examined here, the values of $r$ tend to be unobservably small.
△ Less
Submitted 16 April, 2015; v1 submitted 22 December, 2014;
originally announced December 2014.
-
Feeling de Sitter
Authors:
Andreas Albrecht,
R. Holman,
Benoit J. Richard
Abstract:
We address the following question: To what extent can a quantum field tell if it has been placed in de Sitter space? Our approach is to use the techniques of non-equilibrium quantum field theory to compute the time evolution of a state which starts off in flat space for (conformal) times $η<η_0$, and then evolves in a de Sitter background turned on instantaneously at $η=η_0$. We find that the answ…
▽ More
We address the following question: To what extent can a quantum field tell if it has been placed in de Sitter space? Our approach is to use the techniques of non-equilibrium quantum field theory to compute the time evolution of a state which starts off in flat space for (conformal) times $η<η_0$, and then evolves in a de Sitter background turned on instantaneously at $η=η_0$. We find that the answer depends on what quantities one examines. We study a range of them, all based on two-point correlation functions, and analyze which ones approach the standard Bunch-Davies values over time. The outcome of this analysis suggests that the nature of the equilibration process in this system is similar to that in more familiar systems.
△ Less
Submitted 14 January, 2015; v1 submitted 9 October, 2014;
originally announced October 2014.
-
Conduction Electron Scattering and Spin-Flip** at Sputtered Co/Ni Interfaces
Authors:
H. Y. T. Nguyen,
R. Acharyya,
E. Huey,
B. Richard,
R. Loloee,
W. P. Pratt Jr.,
J. Bass,
Shuai Wang,
Ke Xia
Abstract:
Current-perpendicular-to-plane magnetoresistance (CPP-MR) measurements let us quantify conduction electron scattering and spin-flip** at a sputtered ferromagnetic/ferromagnetic (F1/F2 = Co/Ni) interface, with important consequences for CPP-MR and spin-torque experiments with perpendicular anisotropy. We use ferromagnetically coupled ([Ni/Co]xn)Ni multilayers, and Py-based, symmetric double excha…
▽ More
Current-perpendicular-to-plane magnetoresistance (CPP-MR) measurements let us quantify conduction electron scattering and spin-flip** at a sputtered ferromagnetic/ferromagnetic (F1/F2 = Co/Ni) interface, with important consequences for CPP-MR and spin-torque experiments with perpendicular anisotropy. We use ferromagnetically coupled ([Ni/Co]xn)Ni multilayers, and Py-based, symmetric double exchange-biased spin-valves (DEBSVs) containing inserts of ferromagnetically coupled ([Co/Ni]xn)Co or ([Ni/Co]xn)Ni multilayers, to derive Co/Ni interface specific resistances AR(Co/Ni)(Up) = 0.03 (+0.02)(-0.03) f-ohm-m^2 and AR(Co/Ni)(down) = 1.00 +/- 0.07 f-ohm-m^2, and interface spin-flip** parameter delta(Co/Ni) = 0.35 +/- 0.05. The specific resistances are consistent with our no-free-parameter calculations for an interface thickness between 2 and 4 monolayers (ML) that is compatible with expectations.
△ Less
Submitted 20 December, 2010;
originally announced December 2010.
-
Spin-Diffusion Lengths in Dilute Cu(Ge) and Ag(Sn) Alloys
Authors:
Q. Fowler,
B. Richard,
A. Sharma,
N. Theodoropoulou,
R. Loloee,
W. P. Pratt Jr.,
J. Bass
Abstract:
We use current-perpendicular-to-plane (CPP) exchange-biased spin-valves to directly measure spin diffusion lengths lsf for N = Cu(2.1 at.%Ge) and Ag(3.6 at.%Sn) alloys. We find lsf(Cu2%Ge)) = 117+10-6 nm and lsf(Ag4%Sn)= 39 +/- 3 nm. The good agreement of this lsf(Cu2%Ge) with the value lsf(Cu2%Ge) = 121 +/- 10 nm derived from an independent spin-orbit cross-section measurement for Ge in Cu, qua…
▽ More
We use current-perpendicular-to-plane (CPP) exchange-biased spin-valves to directly measure spin diffusion lengths lsf for N = Cu(2.1 at.%Ge) and Ag(3.6 at.%Sn) alloys. We find lsf(Cu2%Ge)) = 117+10-6 nm and lsf(Ag4%Sn)= 39 +/- 3 nm. The good agreement of this lsf(Cu2%Ge) with the value lsf(Cu2%Ge) = 121 +/- 10 nm derived from an independent spin-orbit cross-section measurement for Ge in Cu, quantitatively validates the use of Valet-Fert theory for CPP-MR data analysis to layer thicknesses several times larger than had been done before.
From the value of lsf(Ag4%Sn), we predict the ESR spin-orbit cross-section for Sn impurities in Ag.
△ Less
Submitted 12 September, 2008;
originally announced September 2008.
