arXiv subjects

Marin Soljacic

Publications and source records attributed to Marin Soljacic.

At least 37 records · Page 2Linked to original sources

arXiv2022

On the Importance of Calibration in Semi-supervised Learning

State-of-the-art (SOTA) semi-supervised learning (SSL) methods have been highly successful in leveraging a mix of labeled and unlabeled data by combining techniques of consistency regularization and pseudo-labeling. During pseudo-labeling, the model's predictions on unlabeled data are used for training and thus, model calibration is important in mitigating confirmation bias. Yet, many SOTA methods are optimized for model performance, with little focus directed to improve model calibration. In this work, we empirically demonstrate that model calibration is strongly correlated with model performance and propose to improve calibration via approximate Bayesian techniques. We introduce a family of new SSL models that optimizes for calibration and demonstrate their effectiveness across standard vision benchmarks of CIFAR-10, CIFAR-100 and ImageNet, giving up to 15.9% improvement in test accuracy. Furthermore, we also demonstrate their effectiveness in additional realistic and challenging problems, such as class-imbalanced datasets and in photonics science.

cs.LG↗

arXiv2022

Quantum Electrodynamical Metamaterials

Recent experiments have revealed ultrastrong coupling between light and matter as a promising avenue for modifying material properties, such as electrical transport, chemical reaction rates, and even superconductivity. Here, we explore (ultra)strong coupling as a means for manipulating the optical response of metamaterials based on ensembles of constituent units individually in the ultrastrong coupling regime. We develop a framework based on linear response for quantum electrodynamical systems to study how light-matter coupling affects the optical response. We begin by applying this framework to find the optical response of a two-level emitter coupled to a single cavity mode, which could be seen as a "meta-atom" of a metamaterial built from repeated units of this system. We find optical behaviors ranging from that of a simple two-level system (Lorentz-oscillator) to effectively transparent, as the coupling goes from the weak to deep strong coupling regimes. We explore a one-dimensional chain of these meta-atoms, demonstrating the tunability of its optical behavior. Our scheme may ultimately provide a framework for designing new metamaterials with low-loss, highly-confined modes, as well as tunable (single-photon) nonlinearities.

physics.optics↗

arXiv2022

Complete condensation of photon noise in nonlinear dissipative systems

Fock states are the most fundamental quantum states of bosonic fields, forming an important basis for understanding their quantum dynamics. As energy and number eigenstates, they have an exactly defined number of quanta, and most faithfully express the particle nature of fields. These properties make them attractive for many applications in metrology, and quantum simulation and information processing. Yet, Fock states are notoriously difficult to generate. The problem is especially acute in optics, where it is difficult to deterministically produce Fock states with more than a single photon, let alone at macroscopic scales. This is in part due to a dearth of mechanisms to produce large Fock states, as well as the deleterious effects of linear dissipation. Here, we introduce a new effect in the physics of nonlinear bosons, arising from the interplay of dissipation and Kerr nonlinearity. In this effect, a nonlinear resonance is dissipationless when it has a particular number of quanta (e.g., photons) inside it, and lossy otherwise. This loss, which results from nonlinear interference, leads to several new quantum statistical effects. For example, it leads to spontaneous condensation of intensity noise, which may enable generation of large Fock and extremely photon-number-squeezed states of light. We also show how this effect has implications for new classes of optoelectronic devices such as lasers, which can stabilize extremely low-noise states in an equilibrium between gain and the nonlinear loss that we introduce. Throughout the text, we present examples of systems that may realize these effects. In one, we show how the nonlinear dissipation could lead to optical Fock states of $n=1000$, while in another, we show how conventional laser architectures could be used to generate macroscopic light ($>10^{12}$ photons) with nearly 95% less noise than the standard quantum limit.

quant-ph↗

arXiv2021

Surrogate- and invariance-boosted contrastive learning for data-scarce applications in science

Deep learning techniques have been increasingly applied to the natural sciences, e.g., for property prediction and optimization or material discovery. A fundamental ingredient of such approaches is the vast quantity of labelled data needed to train the model; this poses severe challenges in data-scarce settings where obtaining labels requires substantial computational or labor resources. Here, we introduce surrogate- and invariance-boosted contrastive learning (SIB-CL), a deep learning framework which incorporates three ``inexpensive'' and easily obtainable auxiliary information sources to overcome data scarcity. Specifically, these are: 1)~abundant unlabeled data, 2)~prior knowledge of symmetries or invariances and 3)~surrogate data obtained at near-zero cost. We demonstrate SIB-CL's effectiveness and generality on various scientific problems, e.g., predicting the density-of-states of 2D photonic crystals and solving the 3D time-independent Schrodinger equation. SIB-CL consistently results in orders of magnitude reduction in the number of labels needed to achieve the same network accuracies.

cs.LG↗

arXiv2019

Heuristic Recurrent Algorithms for Photonic Ising Machines

The inability of conventional electronic architectures to efficiently solve large combinatorial problems motivates the development of novel computational hardware. There has been much effort recently toward developing novel, application-specific hardware, across many different fields of engineering, such as integrated circuits, memristors, and photonics. However, unleashing the true potential of such novel architectures requires the development of featured algorithms which optimally exploit their fundamental properties. We here present the Photonic Recurrent Ising Sampler (PRIS), a heuristic method tailored for parallel architectures that allows for fast and efficient sampling from distributions of combinatorially hard Ising problems. Since the PRIS relies essentially on vector-to-fixed matrix multiplications, we suggest the implementation of the PRIS in photonic parallel networks, which realize these operations at an unprecedented speed. The PRIS provides sample solutions to the ground state of arbitrary Ising models, by converging in probability to their associated Gibbs distribution. By running the PRIS at various noise levels, we probe the critical behavior of universality classes and their critical exponents. In addition to the attractive features of photonic networks, the PRIS relies on intrinsic dynamic noise and eigenvalue dropout to find ground states more efficiently. Our work suggests speedups in heuristic methods via photonic implementations of the PRIS. We also hint at a broader class of (meta)heuristic algorithms derived from the PRIS, such as combined simulated annealing on the noise and eigenvalue dropout levels. Our algorithm can also be implemented in a competitive manner on fast parallel electronic hardware, such as FPGAs and ASICs.

physics.app-ph↗

arXiv2019

Monochromatic X-ray source based on scattering from a magnetic nanoundulator

We present a novel design for an ultra-compact, passive light source capable of generating ultraviolet and X-ray radiation, based on the interaction of free electrons with the magnetic near-field of a ferromagnet. Our design is motivated by recent advances in the fabrication of nanostructures, which allow the confinement of large magnetic fields at the surface of ferromagnetic nanogratings. Using ab initio simulations and a complementary analytical theory, we show that highly directional, tunable, monochromatic radiation at high frequencies could be produced from relatively low-energy electrons within a tabletop design. The output frequency is tunable in the extreme ultraviolet to hard X-ray range via electron kinetic energies from 1 keV-5 MeV and nanograting periods from 1 μm-5 nm. Our design reduces the scale, cost, and complexity of current free-electron-driven light schemes, bypassing the need for lengthy acceleration stages in conventional synchrotrons and free-electron lasers and driving lasers in other compact designs. Our design could help realize the next generation of tabletop or on-chip X-ray sources.

physics.optics↗

arXiv2019

Migrating Knowledge between Physical Scenarios based on Artificial Neural Networks

Deep learning is known to be data-hungry, which hinders its application in many areas of science when datasets are small. Here, we propose to use transfer learning methods to migrate knowledge between different physical scenarios and significantly improve the prediction accuracy of artificial neural networks trained on a small dataset. This method can help reduce the demand for expensive data by making use of additional inexpensive data. First, we demonstrate that in predicting the transmission from multilayer photonic film, the relative error rate is reduced by 46.8% (26.5%) when the source data comes from 10-layer (8-layer) films and the target data comes from 8-layer (10-layer) films. Second, we show that the relative error rate is decreased by 22% when knowledge is transferred between two very different physical scenarios: transmission from multilayer films and scattering from multilayer nanoparticles. Finally, we propose a multi-task learning method to improve the performance of different physical scenarios simultaneously in which each task only has a small dataset.

cs.CV↗

arXiv2018

WaveletNet: Logarithmic Scale Efficient Convolutional Neural Networks for Edge Devices

We present a logarithmic-scale efficient convolutional neural network architecture for edge devices, named WaveletNet. Our model is based on the well-known depthwise convolution, and on two new layers, which we introduce in this work: a wavelet convolution and a depthwise fast wavelet transform. By breaking the symmetry in channel dimensions and applying a fast algorithm, WaveletNet shrinks the complexity of convolutional blocks by an O(logD/D) factor, where D is the number of channels. Experiments on CIFAR-10 and ImageNet classification show superior and comparable performances of WaveletNet compared to state-of-the-art models such as MobileNetV2.

cs.LG↗

arXiv2018

Extreme enhancement of spin relaxation mediated by surface magnon polaritons

Polaritons in metals, semimetals, semiconductors, and polar insulators, with their extreme confinement of electromagnetic energy, provide many promising opportunities for enhancing typically weak light-matter interactions such as multipolar radiation, multiphoton spontaneous emission, Raman scattering, and material nonlinearities. These highly confined polaritons are quasi-electrostatic in nature, with most of their energy residing in the electric field. As a result, these "electric" polaritons are far from optimized for enhancing emission of a magnetic nature, such as spin relaxation, which is typically many orders of magnitude slower than corresponding electric decays. Here, we propose using surface magnon polaritons in negative magnetic permeability materials such as MnF$_2$ and FeF$_2$ to strongly enhance spin-relaxation in nearby emitters in the THz spectral range. We find that these magnetic polaritons in 100 nm thin-films can be confined to lengths over 10,000 times smaller than the wavelength of a photon at the same frequency, allowing for a surprising twelve orders of magnitude enhancement in magnetic dipole transitions. This takes THz spin-flip transitions, which normally occur at timescales on the order of a year, and forces them to occur at sub-ms timescales. Our results suggest an interesting platform for polaritonics at THz frequencies, and more broadly, a new way to use polaritons to control light-matter interactions.

physics.optics↗

arXiv2018

On-Chip Optical Convolutional Neural Networks

Convolutional Neural Networks (CNNs) are a class of Artificial Neural Networks(ANNs) that employ the method of convolving input images with filter-kernels for object recognition and classification purposes. In this paper, we propose a photonics circuit architecture which could consume a fraction of energy per inference compared with state of the art electronics.

cs.ET↗

arXiv2018

Controlling Cherenkov angles with resonance transition radiation

Cherenkov radiation provides a valuable way to identify high energy particles in a wide momentum range, through the relation between the particle velocity and the Cherenkov angle. However, since the Cherenkov angle depends only on material's permittivity, the material unavoidably sets a fundamental limit to the momentum coverage and sensitivity of Cherenkov detectors. For example, Ring Imaging Cherenkov detectors must employ materials transparent to the frequency of interest as well as possessing permittivities close to unity to identify particles in the multi GeV range, and thus are often limited to large gas chambers. It would be extremely important albeit challenging to lift this fundamental limit and control Cherenkov angles as preferred. Here we propose a new mechanism that uses constructive interference of resonance transition radiation from photonic crystals to generate both forward and backward Cherenkov radiation. This mechanism can control Cherenkov angles in a flexible way with high sensitivity to any desired range of velocities. Photonic crystals thus overcome the severe material limit for Cherenkov detectors, enabling the use of transparent materials with arbitrary values of permittivity, and provide a promising option suited for identification of particles at high energy with enhanced sensitivity.

physics.app-ph↗

arXiv2017

Shaping Long-lived Electron Wavepackets for Customizable Optical Spectra

Electrons in atoms and molecules are versatile physical systems covering a vast range of light-matter interactions, enabling the physics of Rydberg states, photon-photon bound states, simulation of condensed matter Hamiltonians, and quantum sources of light. A limitation on the versatility of such electronic systems for optical interactions would appear to arise from the discrete nature of the electronic transitions and from the limited ionization energy, constraining the energy scale through which the rich physics of bound electrons can be accessed. In this work, we propose the concept of shaping spatially confined electron wavepackets as superpositions of extended states in the ionization continuum. These wavepackets enable customizable optical emission spectra transitions in the eV-keV range. We find that the specific shaping lengthens the diffraction lifetime of the wavepackets in exchange for increasing their spatial spreads. Finally, we study the spontaneous radiative transitions of these excited states, examining the influence of the wavepacket shape and size on the radiation rate of the excited state. We observe that the rate of radiative capture is primarily limited by the diffraction lifetime of the wavepacket. The approach proposed in this work could have applications towards developing designer emitters at tunable energy and length scales, potentially bringing the physics of Rydberg states to new energy scales and extending the range and versatility of emitters that can be developed using shaped electrons.

quant-ph↗

arXiv2017

Semi-Transparent Solar Cell enabled by Frequency Selective Light Trapping

We propose a frequency selective light trapping scheme that enables the creation of more visually-transparent and yet simultaneously more efficient semitransparent solar cells. A nanoparticle scattering layer and photonic stack back reflector create a selective trapping effect by total internal reflection within a medium, increasing absorption of IR light. We propose a strong frequency selective scattering layer using spherical TiO2 nanoparticles with radius of 255 nm and area density of 1.1% in a medium with index of refraction of 1.5. Using detailed numerical simulations for this configuration, we find that it is possible to create a semitransparent silicon solar cell that has a Shockley Queisser efficiency of 12.0%\pm0.4% with a visible transparency of 60.2%\pm1.3%, 13.3%\pm1.3 more visibly-transparent than a bare silicon cell at the same efficiency.

physics.app-ph↗

arXiv2017

Rotational Unit of Memory

The concepts of unitary evolution matrices and associative memory have boosted the field of Recurrent Neural Networks (RNN) to state-of-the-art performance in a variety of sequential tasks. However, RNN still have a limited capacity to manipulate long-term memory. To bypass this weakness the most successful applications of RNN use external techniques such as attention mechanisms. In this paper we propose a novel RNN model that unifies the state-of-the-art approaches: Rotational Unit of Memory (RUM). The core of RUM is its rotational operation, which is, naturally, a unitary matrix, providing architectures with the power to learn long-term dependencies by overcoming the vanishing and exploding gradients problem. Moreover, the rotational unit also serves as associative memory. We evaluate our model on synthetic memorization, question answering and language modeling tasks. RUM learns the Copying Memory task completely and improves the state-of-the-art result in the Recall task. RUM's performance in the bAbI Question Answering task is comparable to that of models with attention mechanism. We also improve the state-of-the-art result to 1.189 bits-per-character (BPC) loss in the Character Level Penn Treebank (PTB) task, which is to signify the applications of RUM to real-world sequential data. The universality of our construction, at the core of RNN, establishes RUM as a promising approach to language modeling, speech recognition and machine translation.

cs.LG↗

arXiv2017

Nanophotonic Particle Simulation and Inverse Design Using Artificial Neural Networks

We propose a method to use artificial neural networks to approximate light scattering by multilayer nanoparticles. We find the network needs to be trained on only a small sampling of the data in order to approximate the simulation to high precision. Once the neural network is trained, it can simulate such optical processes orders of magnitude faster than conventional simulations. Furthermore, the trained neural network can be used solve nanophotonic inverse design problems by using back- propogation - where the gradient is analytical, not numerical.

physics.comp-ph↗

arXiv2017

Observation of Bulk Fermi Arc and Polarization Half Charge from Paired Exceptional Points

The ideas of topology have found tremendous success in Hermitian physical systems, but even richer properties exist in the more general non-Hermitian framework. Here, we theoretically propose and experimentally demonstrate a new topologically-protected bulk Fermi arc which---unlike the well-known surface Fermi arcs arising from Weyl points in Hermitian systems---develops from non-Hermitian radiative losses in photonic crystal slabs. Moreover, we discover half-integer topological charges in the polarization of far-field radiation around the Fermi arc. We show that both phenomena are direct consequences of the non-Hermitian topological properties of exceptional points, where resonances coincide in their frequencies and linewidths. Our work connects the fields of topological photonics, non-Hermitian physics and singular optics, and paves the way for future exploration of non-Hermitian topological systems.

physics.optics↗

arXiv2017

Limits to the Optical Response of Graphene and 2D Materials

2D materials provide a platform for strong light--matter interactions, creating wide-ranging design opportunities via new-material discoveries and new methods for geometrical structuring. We derive general upper bounds to the strength of such light--matter interactions, given only the optical conductivity of the material, including spatial nonlocality, and otherwise independent of shape and configuration. Our material figure of merit shows that highly doped graphene is an optimal material at infrared frequencies, whereas single-atomic-layer silver is optimal in the visible. For quantities ranging from absorption and scattering to near-field spontaneous-emission enhancements and radiative heat transfer, we consider canonical geometrical structures and show that in certain cases the bounds can be approached, while in others there may be significant opportunity for design improvement. The bounds can encourage systematic improvements in the design of ultrathin broadband absorbers, 2D antennas, and near-field energy harvesters.

physics.optics↗

arXiv2017

Ultra-light Å-scale Optimal Optical Reflectors

High-reflectance in many state-of-the-art optical devices is achieved with noble metals. However, metals are limited by losses, and for certain applications, by their high mass density. Using a combination of ab initio and optical transfer matrix calculations, we evaluate the behavior of graphene-based Å-scale metamaterials and find that they could act as nearly-perfect reflectors in the mid-long wave infrared (IR) range. The low density of states for electron-phonon scattering and interband excitations leads to unprecedented optical properties for graphene heterostructures, especially alternating atomic layers of graphene and hexagonal boron nitride, at wavelengths greater than $10~μ$m. At these wavelengths, these materials exhibit reflectivities exceeding 99.7% at a fraction of the weight of noble metals, as well as plasmonic mode confinement and quality factors that are greater by an order of magnitude compared to noble metals. These findings hold promise for ultra-compact optical components and waveguides for mid-IR applications. Moreover, unlike metals, the photonic properties of these heterostructures could be actively tuned via chemical and/or electrostatic doping, providing exciting possibilities for tunable devices.

cond-mat.mes-hall↗