A5077972639- Photonic and Optical Devices
- Advanced Fiber Laser Technologies
- Statistical Methods and Inference
- Advanced Photonic Communication Systems
- GaN-based semiconductor devices and materials
- Semiconductor Lasers and Optical Devices
- Machine Learning and Algorithms
- Photorefractive and Nonlinear Optics
- Distributed Sensor Networks and Detection Algorithms
- Markov Chains and Monte Carlo Methods
- ZnO doping and properties
- Ga2O3 and related materials
- Optical Network Technologies
- Bayesian Methods and Mixture Models
- Bayesian Modeling and Causal Inference
- Advanced Bandit Algorithms Research
- Advanced Optical Sensing Technologies
- Semiconductor Quantum Structures and Devices
- Statistical Methods and Bayesian Inference
- Wireless Communication Security Techniques
- Optical Coatings and Gratings
- Advanced Fiber Optic Sensors
- Metal and Thin Film Mechanics
- Optical Coherence Tomography Applications
- Plasmonic and Surface Plasmon Research
Tsinghua University
2016-2025
Courant Institute of Mathematical Sciences
2024
New York University
2024
Shijiazhuang Tiedao University
2008-2023
Massachusetts Institute of Technology
2023
Institute of Physics
2023
National Engineering Research Center for Information Technology in Agriculture
2021-2022
Stanford University
2015-2021
Simons Foundation
2021
Kyoto University
2019
We propose a general methodology for the construction and analysis of essentially minimax estimators wide class functionals finite dimensional parameters, elaborate on case discrete distributions, where support size S is unknown may be comparable with or even much larger than number observations n. treat respective regions functional nonsmooth smooth separately. In regime, we apply an unbiased estimator best polynomial approximation whereas, in bias-corrected version maximum likelihood...
Location-aware networks are of great importance and interest in both civil military applications. This paper determines the localization accuracy an agent, which is equipped with antenna array localizes itself using wireless measurements anchor nodes, a far-field environment. In view Cram\'er-Rao bound, we first derive information for static scenarios demonstrate that such weighed sum Fisher matrices from each anchor-antenna measurement pair. Each matrix can be further decomposed into two...
Development of planar-integrated microresonators with high quality factors (Q's) is crucial for nonlinear photonics in a robust chip. Compared silicon and nitride, aluminum nitride (AlN) features intrinsic quadratic cubic susceptibilities as well an enormous band gap (∼6.2 eV), making it ideal optical interactions. However, sputtered polycrystalline AlN susceptible to scattering defect-related absorption losses, thereby inducing limited Q-factors. Here, we demonstrate single-crystalline...
Abstract Gallium nitride (GaN) as a wide bandgap material is widely used in solid‐state lighting. Thanks to its high nonlinearity and refractive index contrast, GaN‐on‐insulator (GaNOI) also promising platform for nonlinear optical applications. Despite intriguing proprieties, applications of GaN are rarely studied owing the relatively loss waveguides (typically ≈2 dB cm −1 ). In this paper, GaNOI microresonators with intrinsic quality factor over 2.5 million reported, corresponding an 0.17...
Novel back-illuminated modified uni-traveling-carrier photodiodes (MUTC-PDs) with wide bandwidth and high saturation power are demonstrated. The effect of cliff layer doping on the electric field distribution is investigated to achieve fast carrier transport. MUTC-PDs miniaturized device diameter low contact resistance fabricated improve RC-limited bandwidth. Meanwhile, inductive peaking implemented further extend PDs 3-µm 3.6-µm-diameter exhibit a ultrawide 230 GHz 200 GHz, together -4.94...
We consider the problem of estimating functionals discrete distributions, and focus on tight nonasymptotic analysis worst case squared error risk widely used estimators. apply concentration inequalities to analyze random fluctuation these estimators around their expectations, theory approximation using positive linear operators deviation expectations from true functional, namely \emph{bias}. characterize incurred by Maximum Likelihood Estimator (MLE) in Shannon entropy $H(P) = \sum_{i 1}^S...
Free-space optical (FSO) communications can achieve high capacity with huge unlicensed spectrum and low operational costs. The corresponding performance analysis of FSO systems over turbulence channels is very limited, particularly when using multiple apertures at both transmitter receiver sides. This paper aims to provide the ergodic characterization multiple-input-multiple-output (MIMO) atmospheric turbulence-induced fading channels. fluctuations irradiance distorted by conditions usually...
We consider the problem of minimax estimation entropy a density over Lipschitz balls. Dropping usual assumption that is bounded away from zero, we obtain rates $(n\ln n)^{-s/(s+d)}+n^{-1/2}$ for $0<s\leq 2$ densities supported on $[0,1]^{d}$, where $s$ smoothness parameter and $n$ number independent samples. generalize results to with unbounded support: given an Orlicz functions $\Psi $ rapid growth (such as subexponential sub-Gaussian classes), $-Orlicz norm increase n)^{-s/(s+d)}(\Psi...
A novel thin-film lithium niobate (TFLN) electro-optic modulator is proposed and demonstrated. LiNbO3-silica hybrid waveguide adopted to maintain low optical loss for an electrode spacing as narrow 3 µm, resulting in a half-wave-voltage length product of only 1.7 V·cm. Capacitively loaded traveling-wave electrodes are employed reduce the microwave loss, while quartz substrate used place silicon achieve velocity matching. The fabricated TFLN with 5-mm-long modulation region exhibits half-wave...
Face sketch synthesis with a photo is challenging due to that the psychological mechanism of generation difficult be expressed precisely by rules. Current learning-based methods concentrate on learning rules optimizing cost functions low-level image features. In this paper, new face method presented, which inspired recent advances in sparse signal representation and neuroscience human brain probably perceives images using high-level features are sparse. Sparse representations desired...
Semiconductor devices capable of generating a vortex beam with specific orbital angular momentum (OAM) order are highly attractive for applications ranging from nanoparticle manipulation, imaging and microscopy to fiber quantum communications. In this work, an electrically pumped integrated OAM emitter operating at telecom wavelengths is fabricated by monolithically integrating optical distributed feedback laser on the same InGaAsP/InP epitaxial wafer. A single-step dry-etching process...
We consider parameter estimation in distributed networks, where each sensor the network observes an independent sample from underlying distribution and has $k$ bits to communicate its a centralized processor which computes estimate of desired parameter. develop lower bounds for minimax risk estimating large class losses distributions. Our results show that under mild regularity conditions, communication constraint reduces effective size by factor $d$ when is small, dimension estimated...
We propose a framework to analyze and quantify the bias in adaptive data analysis. It generalizes that proposed by Russo Zou'15, applying measurements whose moment generating function exists, with finite p-norm, general Orlicz spaces. introduce new class of dependence measures which retain key properties mutual information while more effectively quantifying exploration for heavy tailed distributions. provide examples cases where our bounds are nearly tight situations original Zou'15 does not apply.
We consider the problem of discrete distribution estimation under l <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sub> loss. provide tight upper and lower bounds on maximum risk empirical (the likelihood estimator), minimax in regimes where support size S may grow with number observations n. show that among distributions bounded entropy H, asymptotic for is 2H/ln n, while H/ ln Moreover, we a hard-thresholding estimator oblivious to unknown...
We consider the problem of estimating $L_1$ distance between two discrete probability measures $P$ and $Q$ from empirical data in a nonasymptotic large alphabet setting. When is known one obtains $n$ samples $P$, we show that for every $Q$, minimax rate-optimal estimator with achieves performance comparable to maximum likelihood (MLE) $n\ln n$ samples. both are unknown, construct estimators whose worst case essentially being uniform, implying uniform most difficult case. The \emph{effective...
Novel back-illuminated modified uni-traveling- carrier photodiodes (MUTC-PDs) are reported to demonstrate wide bandwidth and high output power performance at D-band (110–170 GHz) regime. A comprehensive design model is employed predict tune the frequency response by incorporating coplanar waveguides (CPWs) with different inductive peaking effects. As a demonstration, 4.5-μm-diameter two types of CPWs fabricated exhibit profiles. Both PDs 3-dB over 150 GHz measured responses in excellent...
Novel evanescently coupled waveguide modified uni-traveling carrier photodiodes (MUTC-PDs) employing a thick multi-layer coupling are reported. To improve the optical-to-electrical (O/E) conversion efficiency, with gradually increased refractive index from bottom layer to absorption is utilized. The profile facilitates upward transmission of incident light into region, thereby enhancing evanescent efficiency. Meanwhile, waveguide, total thickness 1.75 µm, expands mode field diameter,...
In “Optimal No-Regret Learning in Repeated First-Price Auctions,” Y. Han, W. Tsachy, and Z. Zhou study online learning repeated first-price auctions where a bidder, only observing the winning bid at end of each auction, learns to adaptively maximize her cumulative payoff. To achieve this goal, bidder faces censored feedback: If she wins bid, then is not able observe highest other bidders, which we assume i.i.d. drawn from an unknown distribution. paper, they develop first algorithm that...
Optical neural networks are at the forefront of computational innovation, utilizing photons as primary carriers information and employing optical components for computation. However, fundamental nonlinear device in is barely satisfied because its high energy threshold poor reconfigurability. This paper proposes demonstrates an sigmoid-type computation mode Vertical-Cavity Surface-Emitting Lasers (VCSELs) biased beneath threshold. The programmable by simply adjusting injection current....
In this paper, we study the multi-armed bandit problem in batched setting where employed policy must split data into a small number of batches. While minimax regret for two-armed stochastic bandits has been completely characterized \cite{perchet2016batched}, effect arms on case is still open. Moreover, question whether adaptively chosen batch sizes will help to reduce also remains underexplored. propose BaSE (batched successive elimination) achieve rate-optimal regrets (within logarithmic...