Searcharxiv⌕ Search

arXiv subjects

Tian Li

Publications and source records attributed to Tian Li.

At least 127 records · Page 7Linked to original sources
arXiv2019

LEAF: A Benchmark for Federated Settings

Modern federated networks, such as those comprised of wearable devices, mobile phones, or autonomous vehicles, generate massive amounts of data each day. This wealth of data can help to learn models that can improve the user experience on each device. However, the scale and heterogeneity of federated data presents new challenges in research areas such as federated learning, meta-learning, and multi-task learning. As the machine learning community begins to tackle these challenges, we are at a critical time to ensure that developments made in these areas are grounded with realistic benchmarks. To this end, we propose LEAF, a modular benchmarking framework for learning in federated settings. LEAF includes a suite of open-source federated datasets, a rigorous evaluation framework, and a set of reference implementations, all geared towards capturing the obstacles and intricacies of practical federated environments.

cs.LG↗
arXiv2019

Privacy for Free: Communication-Efficient Learning with Differential Privacy Using Sketches

Communication and privacy are two critical concerns in distributed learning. Many existing works treat these concerns separately. In this work, we argue that a natural connection exists between methods for communication reduction and privacy preservation in the context of distributed machine learning. In particular, we prove that Count Sketch, a simple method for data stream summarization, has inherent differential privacy properties. Using these derived privacy guarantees, we propose a novel sketch-based framework (DiffSketch) for distributed learning, where we compress the transmitted messages via sketches to simultaneously achieve communication efficiency and provable privacy benefits. Our evaluation demonstrates that DiffSketch can provide strong differential privacy guarantees (e.g., $\varepsilon$= 1) and reduce communication by 20-50x with only marginal decreases in accuracy. Compared to baselines that treat privacy and communication separately, DiffSketch improves absolute test accuracy by 5%-50% while offering the same privacy guarantees and communication compression.

cs.LG↗
arXiv2019

Enhancing the Privacy of Federated Learning with Sketching

In response to growing concerns about user privacy, federated learning has emerged as a promising tool to train statistical models over networks of devices while keeping data localized. Federated learning methods run training tasks directly on user devices and do not share the raw user data with third parties. However, current methods still share model updates, which may contain private information (e.g., one's weight and height), during the training process. Existing efforts that aim to improve the privacy of federated learning make compromises in one or more of the following key areas: performance (particularly communication cost), accuracy, or privacy. To better optimize these trade-offs, we propose that \textit{sketching algorithms} have a unique advantage in that they can provide both privacy and performance benefits while maintaining accuracy. We evaluate the feasibility of sketching-based federated learning with a prototype on three representative learning models. Our initial findings show that it is possible to provide strong privacy guarantees for federated learning without sacrificing performance or accuracy. Our work highlights that there exists a fundamental connection between privacy and communication in distributed settings, and suggests important open problems surrounding the theoretical understanding, methodology, and system design of practical, private federated learning.

cs.LG↗
arXiv2019

Federated Learning: Challenges, Methods, and Future Directions

Federated learning involves training statistical models over remote devices or siloed data centers, such as mobile phones or hospitals, while keeping data localized. Training in heterogeneous and potentially massive networks introduces novel challenges that require a fundamental departure from standard approaches for large-scale machine learning, distributed optimization, and privacy-preserving data analysis. In this article, we discuss the unique characteristics and challenges of federated learning, provide a broad overview of current approaches, and outline several directions of future work that are relevant to a wide range of research communities.

cs.LG↗
arXiv2019

Beyond sub-Rayleigh imaging via high order correlation of speckle illumination

Second order intensity correlations of speckle illumination are extensively used in imaging applications that require going beyond the Rayleigh limit. The theoretical analysis shows that significantly improved imaging can be extracted from the study of increasingly higher order intensity cumulants. We provide experimental evidence by demonstrating resolution beyond what is achievable by second order correlations. We present results up to 20th order. We also show an increased visibility of cumulant correlations compared to moment correlations. Our findings clearly suggest the benefits of using higher order intensity cumulants in other disciplines like astronomy and biology.

physics.optics↗
arXiv2018

Magnetotransport in phase-separated (Ga,Fe)N with $γ$'-Ga$_y$Fe$_{4-y}$N nanocrystals

The magnetotransport in phase-separated (Ga,Fe)N containing $γ$'-Ga$_y$Fe$_{4-y}$N (0\,$<$\,y\,$<$1) nanocrystals (NCs) is studied in the temperature range between 2\,K and 300\,K. The evolution of the resistivity and of the magnetoresistance (MR) as a function of temperature points at two conduction mechanisms: namely a conventional Arrhenius-type one down to 50\,K, and Mott variable range hopping at lower temperatures, where the spin-polarized current is transported between NCs in a regime in which phonon-scattering effects are not dominant. Below 25\,K, the MR shows a hysteretic contribution at magnetic fields $<$1\,T and proportional to the coercive field. Anisotropic magnetoresistance with values one order of magnitude greater than those previously reported for $γ$'-Fe$_4$N thin films over the whole considered temperature range, confirms that the observed MR in these layers is determined by the embedded nanocrystals.

cond-mat.mtrl-sci↗
arXiv2017

Improved measurement of two-mode quantum correlations using a phase-sensitive amplifier

We demonstrate the ability of a phase-sensitive amplifier (PSA) to pre-amplify a selected quadrature of one mode of a two-mode squeezed state in order to improve the measurement of two-mode quantum correlations that exist before degradation due to optical and detection losses. We use four-wave mixing (4WM) in $^{85}$Rb vapor to generate bright beams in a two-mode squeezed state. One of these two modes then passes through a second 4WM interaction in a PSA configuration to noiselessly pre-amplify the desired quadrature of the mode before loss is intentionally introduced. We demonstrate an enhancement in the measured degree of intensity correlation and intensity-difference squeezing between the two modes.

physics.optics↗
arXiv2017

Ease.ml: Towards Multi-tenant Resource Sharing for Machine Learning Workloads

We present ease.ml, a declarative machine learning service platform we built to support more than ten research groups outside the computer science departments at ETH Zurich for their machine learning needs. With ease.ml, a user defines the high-level schema of a machine learning application and submits the task via a Web interface. The system automatically deals with the rest, such as model selection and data movement. In this paper, we describe the ease.ml architecture and focus on a novel technical problem introduced by ease.ml regarding resource allocation. We ask, as a "service provider" that manages a shared cluster of machines among all our users running machine learning workloads, what is the resource allocation strategy that maximizes the global satisfaction of all our users? Resource allocation is a critical yet subtle issue in this multi-tenant scenario, as we have to balance between efficiency and fairness. We first formalize the problem that we call multi-tenant model selection, aiming for minimizing the total regret of all users running automatic model selection tasks. We then develop a novel algorithm that combines multi-armed bandits with Bayesian optimization and prove a regret bound under the multi-tenant setting. Finally, we report our evaluation of ease.ml on synthetic data and on one service we are providing to our users, namely, image classification with deep neural networks. Our experimental evaluation results show that our proposed solution can be up to 9.8x faster in achieving the same global quality for all users as the two popular heuristics used by our users before ease.ml.

cs.DB↗
arXiv2017

Origin of threshold current density for asymmetric magnetoresistance in Pt/Py bilayers

An asymmetric magnetoresistance (MR) is investigated in Py/Pt bilayers. The asymmetric MR linearly increases with current density up to a threshold, and increases more rapidly above the threshold. To reveal the origin of threshold behavior, we investigate the magnetic field dependence of the asymmetric MR. It is found that the magnetic field strongly suppresses the asymmetric MR only above the threshold current density. Micromagnetic simulation reveals that the reduction of magnetization due to the spin-torque oscillation can be the origin of the threshold behavior of asymmetric MR.

cond-mat.mtrl-sci↗
arXiv2016

Stretching magnetism with an electric field in a nitride semiconductor

By direct magnetization measurements, performed employing a new detection scheme, we demonstrate an electrical control of magnetization in wurtzite (Ga,Mn)N. In this dilute magnetic insulator the Fermi energy is pinned by Mn ions in the mid-gap region, and the Mn3+ ions show strong single-ion anisotropy. We establish that (Ga,Mn)N sustains an electric field up to at least 5 MV/cm, indicating that Mn doping turns GaN into a worthwhile semi-insulating material. Under these conditions, the magnetoelectric coupling may be driven by the inverse piezoelectric effect that stretches the elementary cell along the c axis and, thus, affects the magnitude of magnetic anisotropy. We develop a corresponding theory and show that it describes the experimentally determined dependence of magnetization on the electric field quantitatively with no adjustable parameters as a function of the magnetic field and temperature. In this way, our work bridges two research domains developed so far independently: piezoelectricity of wurtzite semiconductors and electrical control of magnetization in hybrid and composite magnetic structures containing piezoelectric components.

cond-mat.mtrl-sci↗
arXiv2016

The effect of input phase modulation to a phase-sensitive optical amplifier

Many optical applications depend on amplitude modulating optical beams using devices such as acousto-optical modulators (AOMs) or optical choppers. Methods to add amplitude modulation (AM) often inadvertently impart phase modulation (PM) onto the light as well. While this PM is of no consequence to many phase-insensitive applications, phase-sensitive processes can be affected. Here we study the effects of input phase and amplitude modulation on the output of a quantum-noise limited phase-sensitive optical amplifier (PSA) realized in hot $^{85}$Rb vapor. We investigate the dependence of PM on AOM alignment and demonstrate a novel approach to quantifying PM by using the PSA as a diagnostic tool. We then use this method to measure the alignment-dependent PM of an optical chopper which arises due to diffraction effects as the chopper blade passes through the optical beam.

physics.optics↗
arXiv2015

Incorporation of Mn in Al$_{x}$Ga$_{1-x}$N probed by x-ray absorption and emission spectroscopy, high-resolution microscopy, x-ray diffraction and first-principles calculations

Synchrotron radiation x-ray absorption and emission spectroscopy techniques, complemented by high-resolution transmission electron microscopy methods and density functional theory calculations are employed to investigate the effect of Mn in Al$_{x}$Ga$_{1-x}$N:Mn samples with an Al content up to 100%. The atomic and electronic structure of Mn is established together with its local environment and valence state. A dilute alloy without precipitation is obtained for Al$_{x}$Ga$_{1-x}$N:Mn with Al concentrations up to 82%, and the surfactant role of Mn in the epitaxial process is confirmed.

cond-mat.mtrl-sci↗
arXiv2015

Controlling a three dimensional electron slab of graded Al$_{x}$Ga$_{1-x}$N

Polarization induced degenerate $n$-type doping with electron concentrations up to $\sim$10$^{20}$\,cm$^{-3}$ is achieved in graded Al$_{x}$Ga$_{1-x}$N layers ($x$: 0\%$\rightarrow$37\%) grown on unintentionally doped and on $n$-doped GaN:Si buffer/reservoir layers by metal organic vapor phase epitaxy. High resolution x-ray diffraction, transmission electron microscopy and electron energy loss spectroscopy confirm the gradient in the composition of the Al$_{x}$Ga$_{1-x}$N layers, while magnetotransport studies reveal the formation of a three dimensional electron slab, whose conductivity can be adjusted through the GaN(:Si) buffer/reservoir.

cond-mat.mes-hall↗
arXiv2015

Non-Resonant Below-Bandgap Two-Photon Absorption in Quantum Dot Solar Cells

We are the first to show with clear experimental results that photons that have energy lower than the transition energy between quantum dots states and valance band can still contribute greatly to the photocurrent via both one-photon absorption process (1PA) and two-photon absorption process (2PA). The tailing states function as both the energy states for low energy photon absorption and the photocarriers extraction pathway. One of the biggest advantages of our method is that it can clearly differentiate the photocurrent due to 1PA process and 2PA process. Both 1PA and 2PA photocurrent generation efficiency in an InAs/GaAs QD device with photon excitation at 1550 nm have been quantitatively evaluated.

cond-mat.mes-hall↗
arXiv2014

Planar array of self-assembled Ga$_{x}$Fe$_{4-x}$N nanocrystals in GaN: Magnetic anisotropy determined via ferromagnetic resonance

The magnetic anisotropy of a planar array of Ga$_{x}$Fe$_{4-x}$N nanocrystals (NCs) embedded in a GaN host is studied by ferromagnetic resonance. X-ray diffraction and transmission electron microscopy are employed to determine the phase and distribution of the nanocrystals. The magnetic anisotropy is found to be primarily uniaxial with the hard axis normal to the NCs plane and to have a comparably weak in-plane hexagonal symmetry. The origin of the magnetic anisotropy is discussed taking into consideration the morphology of the nanocrystals, the epitaxial relations, strain effects and magnetic coupling between the NCs.

cond-mat.mtrl-sci↗
arXiv2014

Quantum mutual information of an entangled state propagating through a fast-light medium

Although it is widely accepted that classical information cannot travel faster than the speed of light in vacuum, the behavior of quantum correlations and quantum information propagating through actively-pumped fast-light media has not been studied in detail. To investigate this behavior, we send one half of an entangled state of light through a gain-assisted fast-light medium and detect the remaining quantum correlations. We show that the quantum correlations can be advanced by a small fraction of the correlation time while the entanglement is preserved even in the presence of noise added by phase-insensitive gain. Additionally, although we observe an advance of the peak of the quantum mutual information between the modes, we find that the degradation of the mutual information due to the added noise appears to prevent an advancement of the leading edge. In contrast, we show that both the leading and trailing edges of the mutual information in a slow-light system can be significantly delayed.

quant-ph↗
arXiv2013

Method for traveling-wave deceleration of buffer-gas beams of CH

Cryogenic buffer-gas beams are a promising method for producing bright sources of cold molecular radicals for cold collision and chemical reaction experiments. In order to use these beams in studies of reactions with controlled collision energies, or in trapping experiments, one needs a method of controlling the forward velocity of the beam. A Stark decelerator can be an effective tool for controlling the mean speed of molecules produced by supersonic jets, but efficient deceleration of buffer-gas beams presents new challenges due to longer pulse lengths. Traveling-wave decelerators are uniquely suited to meet these challenges because of their ability to confine molecules in three dimensions during deceleration and the versatility afforded by the analog control of the electrodes. We have created ground state CH($X^2Π$) radicals in a cryogenic buffer-gas cell with the potential to produce a cold molecular beam of $10^{11}$ mol./pulse. We present a general protocol for Stark deceleration of beams with large position and velocity spreads for use with a traveling-wave decelerator. Our method involves confining molecules transversely with a hexapole for an optimized distance before deceleration. This rotates the phase-space distribution of the molecular packet so that the packet is matched to the time varying phase-space acceptance of the decelerator. We demonstrate with simulations that this method can decelerate a significant fraction of the molecules in successive wells of a traveling-wave decelerator to produce energy-tuned beams for cold and controlled molecule experiments.

physics.chem-ph↗