Search

Evolutionary Algorithms (EA) have been successfully used for the optimization of neural networks for policy search, but they still remain sample inefficient and underperforming in some cases compared to gradient-based reinforcement learning (RL). Various methods combine the two approaches, many of them training a RL algorithm on data from EA evaluations and injecting the RL actor into the EA population. However, when using Evolution Strategies (ES) as the EA, the RL actor can drift genetically far from the the ES distribution and injection can cause a collapse of the ES performance. Here, we highlight the phenomenon of genetic drift where the actor genome and the ES population distribution progressively drift apart, leading to injection having a negative impact on the ES. We introduce Genetic Drift Regularization (GDR), a simple regularization method in the actor training loss that prevents the actor genome from drifting away from the ES. We show that GDR can improve ES convergence on problems where RL learns well, but also helps RL training on other tasks, , fixes the injection issues better than previous controlled injection methods.

Evolution Strategies (ES) are effective gradient-free optimization methods that can be competitive with gradient-based approaches for policy search. ES only rely on the total episodic scores of solutions in their population, from which they estimate fitness gradients for their update with no access to true gradient information. However this makes them sensitive to deceptive fitness landscapes, and they tend to only explore one way to solve a problem. Quality-Diversity methods such as MAP-Elites introduced additional information with behavior descriptors (BD) to return a population of diverse solutions, which helps exploration but leads to a large part of the evaluation budget not being focused on finding the best performing solution. Here we show that behavior information can also be leveraged to find the best policy by identifying promising search areas which can then be efficiently explored with ES. We introduce the framework of Quality with Just Enough Diversity (JEDi) which learns the relationship between behavior and fitness to focus evaluations on solutions that matter. When trying to reach higher fitness values, JEDi outperforms both QD and ES methods on hard exploration tasks like mazes and on complex control problems with large policies.

In evolutionary policy search, neural networks are usually represented using a direct mapping: each gene encodes one network weight. Indirect encoding methods, where each gene can encode for multiple weights, shorten the genome to reduce the dimensions of the search space and better exploit permutations and symmetries. The Geometric Encoding for Neural network Evolution (GENE) introduced an indirect encoding where the weight of a connection is computed as the (pseudo-)distance between the two linked neurons, leading to a genome size growing linearly with the number of genes instead of quadratically in direct encoding. However GENE still relies on hand-crafted distance functions with no prior optimization. Here we show that better performing distance functions can be found for GENE using Cartesian Genetic Programming (CGP) in a meta-evolution approach, hence optimizing the encoding to create a search space that is easier to exploit. We show that GENE with a learned function can outperform both direct encoding and the hand-crafted distances, generalizing on unseen problems, and we study how the encoding impacts neural network properties., Comment: 9 pages, 8 figures

The exquisite precision of atom interferometers has sparked the interest of a large community for use cases ranging from fundamental physics to geodesy and inertial navigation. However, their practical use for onboard applications is still limited, not least because rotation and acceleration are intertwined in a single phase shift in free-fall atom interferometers, which makes the extraction of a useful signal more challenging. Moreover, the spatial separation of the wave packets due to rotations leads to a loss of signal. Here we present an atom interferometer operating over a large range of random angles, rotation rates and accelerations. An accurate model of the expected phase shift allows us to untangle the rotation and acceleration signals. We also implement a real-time compensation system using two fibre-optic gyroscopes and a tip-tilt platform to rotate the reference mirror and maintain the full contrast of the atom interferometer. Using these theoretical and practical tools, we reconstruct the fringes and demonstrate a single-shot sensitivity to acceleration of 24 $\mu$g, for a total interrogation time of 2T = 20 ms, for angles and rotation rates reaching 30$^\circ$ and 14 $^\circ$/s respectively. Our hybrid rotating atom interferometer unlocks the full potential of quantum inertial sensors for onboard applications, such as autonomous navigation or gravity mapping., Comment: 20 pages, 5 figures

This paper introduces an Electric Vehicle Charging Station (EVCS) model that incorporates real-world constraints, such as slot power limitations, contract threshold overruns penalties, or early disconnections of electric vehicles (EVs). We propose a formulation of the problem of EVCS control under uncertainty, and implement two Multi-Stage Stochastic Programming approaches that leverage user-provided information, namely, Model Predictive Control and Two-Stage Stochastic Programming. The model addresses uncertainties in charging session start and end times, as well as in energy demand. A user's behavior model based on a sojourn-time-dependent stochastic process enhances cost reduction while maintaining customer satisfaction. The benefits of the two proposed methods are showcased against two baselines over a 22-day simulation using a real-world dataset. The two-stage approach demonstrates robustness against early disconnections by considering a wider range of uncertainty scenarios for optimization. The algorithm prioritizing user satisfaction over electricity cost achieves a 20% and 36% improvement in two user satisfaction metrics compared to an industry-standard baseline. Additionally, the algorithm striking the best balance between cost and user satisfaction exhibits a mere 3% relative cost increase compared to the theoretically optimal baseline - for which the nonanticipativity constraint is relaxed - while attaining 94% and 84% of the user satisfaction performance in the two used satisfaction metrics.

Microgrids are local energy systems that integrate energy production, demand, and storage units. They are generally connected to the regional grid to import electricity when local production and storage do not meet the demand. In this context, Energy Management Systems (EMS) are used to ensure the balance between supply and demand, while minimizing the electricity bill, or an environmental criterion. The main implementation challenges for an EMS come from the uncertainties in the consumption, the local renewable energy production, and in the price and the carbon intensity of electricity. Model Predictive Control (MPC) is widely used to implement EMS but is particularly sensitive to the forecast quality, and often requires a subscription to expensive third-party forecast services. We introduce four Multistage Stochastic Control Algorithms relying only on historical data obtained from on-site measurements. We formulate them under the shared framework of Multistage Stochastic Programming and benchmark them against two baselines in 61 different microgrid setups using the EMSx dataset. Our most effective algorithm produces notable cost reductions compared to an MPC that utilizes the same uncertainty model to generate predictions, and it demonstrates similar performance levels to an ideal MPC that relies on perfect forecasts.

Robust and accurate acceleration tracking remains a challenge in many fields. For geophysics and economic geology, precise gravity mapping requires onboard sensors combined with accurate positioning and navigation systems. Cold-atom-based quantum inertial sensors can potentially provide such high-precision instruments. However, current scalar instruments require precise alignment with vector quantities. Here, we present the first hybrid three-axis accelerometer exploiting the quantum advantage to measure the full acceleration vector by combining three orthogonal atom interferometer measurements with a classical navigation-grade accelerometer triad. Its ultra-low bias permits tracking the acceleration vector over long timescales -- yielding a 50-fold improvement in stability ($6 \times 10^{-8}~g$) over our classical accelerometers. We record the acceleration vector at a high data rate (1 kHz), with absolute magnitude accuracy below 10 $\mu g$, and pointing accuracy of 4 $\mu$rad. This paves the way toward future strapdown applications with quantum sensors and highlights their potential as future inertial navigation units.

The Dedekind zeta functions of infinitely many non-Galois cubic fields have negative central values.

We present a new electro-optic modulation technique that enables a single laser diode to realize a cold-atom source and a quantum inertial sensor based on matter-wave interferometry. Using carrier-suppressed dual single-sideband modulation, an IQ modulator generates two optical sidebands from separate radio-frequency (rf) signals. These sidebands are controlled independently in frequency, phase, and power using standard rf components. Our laser source exhibits improved rejection of parasitic sidebands compared to those based on phase modulators, which generate large systematic shifts in atom interferometers. We measure the influence of residual laser lines on an atom-interferometric gravimeter and show agreement with a theoretical model. We estimate a reduction of the systematic shift by two orders of magnitude compared to previous architectures, and reach a long-term sensitivity of 15 ng on the gravitational acceleration with an interrogation time of only T = 20 ms. Finally, we characterize the performance of our integrated laser system, and show that it is suitable for mobile sensing applications including gravity surveys and inertial navigation., Comment: 14 pages, 11 figures

We report in this paper electro-optical results on InGaN/GaN based green micro light-emitting diodes ({\mu}LEDs). Current light-voltage measurements reveal that the external quantum efficiency (EQE) behavior versus charge injection does not follow the ABC model prediction. Light-emission homogeneity investigation, carried out by photoluminescence mapping, shows that the Quantum Confinement Starck Effect (QCSE) is less significant at the edges of {\mu}LEDs. Electroluminescence shows a subsequent color green-to-blue deviation at high carrier injection levels. The extracted spectra at different current injection levels tend to show the appearance of discrete wavelength emissions. These observations may enhance the hypothesis that higher-energy excited-levels in InGaN quantum wells may also contribute to the blue shift, solely attributed to QCSE lessening under intense electric field magnitudes. We hereby present first results dealing with green {\mu}LEDs electro-optical performances with regards to their size.

Let $G$ be a split semisimple group over a function field. We prove the temperedness at unramified places of automorphic representations of $G$, subject to a local assumption at one place, stronger than supercuspidality, and assuming the existence of cyclic base change with good properties. Our method relies on the geometry of $\operatorname{Bun}_G$. It is independent of the work of Lafforgue on the global Langlands correspondence.

Long-term inertial navigation is currently limited by the bias drifts of gyroscopes and accelerometers and ultra-stable cold-atom interferometers offer a promising alternative for the next generation of high-end navigation systems. Here, we present an experimental setup and an algorithm hybridizing a stable matter-wave interferometer with a classical accelerometer. We use correlations between the quantum and classical devices to track the bias drift of the latter and form a hybrid sensor. We apply the Kalman filter formalism to obtain an optimal estimate of the bias and simulate experimentally a harsh environment representative of that encountered in mobile sensing applications. We show that our method is more precise and robust than traditional sine-fitting methods. The resulting sensor exhibits a 400 Hz bandwidth and reaches a stability of 10 ng after 11 h of integration., Comment: 7 pages + 4 supplemental, 8 figures

We prove Rietsch's mirror conjecture that the Dubrovin quantum connection for minuscule flag varieties is isomorphic to the character D-module of the Berenstein-Kazhdan geometric crystal. The idea is to recognize the quantum connection as Galois and the geometric crystal as automorphic. We reveal surprising relations with the works of Frenkel-Gross, Heinloth-Ng\^o-Yun, and Zhu on Kloosterman sheaves. The isomorphism comes from global rigidity results where Hecke eigensheaves are determined by their local ramification. As corollaries we obtain combinatorial identities for counts of rational curves and the Peterson variety presentation of the small quantum cohomology ring.

We establish properties of families of automorphic representations as we vary prescribed supercuspidal representations at a given finite set of primes. For the tame supercuspidals constructed by J.-K. Yu we prove the limit multiplicity property with error terms. Thereby we obtain a Sato-Tate equidistribution for the Hecke eigenvalues of these families. The main new ingredient is to show that the orbital integrals of matrix coefficients of tame supercuspidal representations with increasing formal degree on a connected reductive $p$-adic group tend to zero uniformly for every noncentral semisimple element.

In this paper we study quantitative aspects of trace characters $\Theta_\pi$ of reductive $p$-adic groups when the representation $\pi$ varies. Our approach is based on the local constancy of characters and we survey some other related results. We formulate a conjecture on the behavior of $\Theta_\pi$ relative to the formal degree of $\pi$, which we are able to prove in the case where $\pi$ is a tame supercuspidal. The proof builds on J.-K.~Yu's construction and the structure of Moy-Prasad subgroups., Comment: Proceedings of Simons symposium on the trace formula

We study various families of Artin $L$-functions attached to geometric parametrizations of number fields. In each case we find the Sato-Tate measure of the family and determine the symmetry type of the distribution of the low-lying zeros.

We establish the Sato-Tate equidistribution of Hecke eigenvalues on average for families of Hecke--Maass cusp forms on SL(n,R)/SO(n). For each of the principal, symmetric square and exterior square L-functions we verify that the families are essentially cuspidal and deduce the level distribution with restricted support of the low-lying zeros. We also deduce average estimates toward Ramanujan.

We establish the transition behavior of Jacquet-Whittaker functions on split semi-simple Lie groups. As a consequence, we show that for certain finite volume Riemannian manifolds, the local bound for normalized Laplace eigenfunctions does not hold globally., Comment: 55 pages, 3 figures. Corrections in response to referee comments

In [90] the first-named author gave a working definition of a family of automorphic L-functions. Since then there have been a number of works [33], [107], [67] [47], [66] and especially [98] by the second and third-named authors which make it possible to give a conjectural answer for the symmetry type of a family and in particular the universality class predicted in [64] for the distribution of the zeros near s=1/2. In this note we carry this out after introducing some basic invariants associated to a family.