arXiv subjects

Liang Luo

Publications and source records attributed to Liang Luo.

71 records · Page 4Linked to original sources

arXiv2022

Stochastic threshold in cell size control

Classic models of cell size control consider cells divide while reaching a threshold, e.g. size, age, or size extension. The molecular basis of the threshold involves multiple layers of regulation as well as gene noises. In this work, we study cell cycle as first-passage problem with stochastic threshold and discover such stochasticity affects the inter-division statistics, which bewilders the criteria to distinguish the types of size control models. The analytic results show the autocorrelation in the threshold can drive a sizer model to the adder-like and even timer-like inter-division statistics, which is supported by simulations. Following the picture that the autocorrelation in the threshold can propagate to the inter-division statistics, we further show that the adder model can be driven to the timer-like one by positive autocorrelated threshold, and even to the sizer-like one when the threshold is negatively autocorrelated. This work highlights the importance to examine gene noise in size control.

cond-mat.stat-mech↗

arXiv2022

DHEN: A Deep and Hierarchical Ensemble Network for Large-Scale Click-Through Rate Prediction

Learning feature interactions is important to the model performance of online advertising services. As a result, extensive efforts have been devoted to designing effective architectures to learn feature interactions. However, we observe that the practical performance of those designs can vary from dataset to dataset, even when the order of interactions claimed to be captured is the same. That indicates different designs may have different advantages and the interactions captured by them have non-overlapping information. Motivated by this observation, we propose DHEN - a deep and hierarchical ensemble architecture that can leverage strengths of heterogeneous interaction modules and learn a hierarchy of the interactions under different orders. To overcome the challenge brought by DHEN's deeper and multi-layer structure in training, we propose a novel co-designed training system that can further improve the training efficiency of DHEN. Experiments of DHEN on large-scale dataset from CTR prediction tasks attained 0.27\% improvement on the Normalized Entropy (NE) of prediction and 1.2x better training throughput than state-of-the-art baseline, demonstrating their effectiveness in practice.

cs.IR↗

arXiv2021

Characterizing and Taming Resolution in Convolutional Neural Networks

Image resolution has a significant effect on the accuracy and computational, storage, and bandwidth costs of computer vision model inference. These costs are exacerbated when scaling out models to large inference serving systems and make image resolution an attractive target for optimization. However, the choice of resolution inherently introduces additional tightly coupled choices, such as image crop size, image detail, and compute kernel implementation that impact computational, storage, and bandwidth costs. Further complicating this setting, the optimal choices from the perspective of these metrics are highly dependent on the dataset and problem scenario. We characterize this tradeoff space, quantitatively studying the accuracy and efficiency tradeoff via systematic and automated tuning of image resolution, image quality and convolutional neural network operators. With the insights from this study, we propose a dynamic resolution mechanism that removes the need to statically choose a resolution ahead of time.

cs.CV↗

arXiv2021

Local atomic configuration control of superconductivity in the undoped pnictide parent compound BaFe2As2

Emergent superconductivity is strongly correlated with the symmetry of local atomic configuration in the parent compounds of iron-based superconductors. While chemical doping or hydrostatic pressure can change the local geometry, these conventional approaches do not provide a clear pathway in tuning the detailed atomic arrangement predictably, due to the parent compounds complicated structural deformation in the presence of the tetragonal-to-orthorhombic phase transition. Here, we demonstrate a systematic approach to manipulate the local structural configurations in BaFe2As2 epitaxial thin films by controlling two independent structural factors orthorhombicity (in-plane anisotropy) and tetragonality (out-of-plane/in-plane balance) from lattice parameters. We tune superconductivity without chemical doping utilizing both structural factors separately, controlling local tetrahedral coordination in designed thin film heterostructures with substrate clamping and bi-axial strain. We further show that this allows quantitative control of both the structural phase transition, associated magnetism, and superconductivity in the parent material BaFe2As2. This approach will advance the development of tunable thin film superconductors in reduced dimension.

cond-mat.supr-con↗

arXiv2021

Master equation approach to the stochastic accumulation dynamics of bacterial cell cycle

The mechanism of bacterial cell size control has been a mystery for decades, which involves the well-coordinated growth and division in the cell cycle. The revolutionary modern techniques of microfluidics and the advanced live imaging analysis techniques allow long term observations and high-throughput analysis of bacterial growth on single cell level, promoting a new wave of quantitative investigations on this puzzle. Taking the opportunity, this theoretical study aims to clarify the stochastic nature of bacterial cell size control under the assumption of the accumulation mechanism, which is favoured by recent experiments on species of bacteria. Via the master equation approach with properly chosen boundary conditions, the distributions concerned in cell size control are estimated and are confirmed by experiments. In this analysis, the inter-generation Green's function is analytically evaluated as the key to bridge two kinds of statistics used in batch-culture and mother machine experiments. This framework allows us to quantify the noise level in growth and accumulation according to experimental data. As a consequence of non-Gaussian noises of the added sizes, the non-equilibrium nature of bacterial cell size homeostasis is predicted, of which the biological meaning requires further investigation.

cond-mat.stat-mech↗

arXiv2021

Cloud Collectives: Towards Cloud-aware Collectives forML Workloads with Rank Reordering

ML workloads are becoming increasingly popular in the cloud. Good cloud training performance is contingent on efficient parameter exchange among VMs. We find that Collectives, the widely used distributed communication algorithms, cannot perform optimally out of the box due to the hierarchical topology of datacenter networks and multi-tenancy nature of the cloudenvironment.In this paper, we present Cloud Collectives , a prototype that accelerates collectives by reordering theranks of participating VMs such that the communication pattern dictated by the selected collectives operation best exploits the locality in the network.Collectives is non-intrusive, requires no code changes nor rebuild of an existing application, and runs without support from cloud providers. Our preliminary application of Cloud Collectives on allreduce operations in public clouds results in a speedup of up to 3.7x in multiple microbenchmarks and 1.3x in real-world workloads of distributed training of deep neural networks and gradient boosted decision trees using state-of-the-art frameworks.

cs.DC↗

arXiv2021

Accelerating SpMM Kernel with Cache-First Edge Sampling for Graph Neural Networks

Graph neural networks (GNNs), an emerging deep learning model class, can extract meaningful representations from highly expressive graph-structured data and are therefore gaining popularity for wider ranges of applications. However, current GNNs suffer from the poor performance of their sparse-dense matrix multiplication (SpMM) operator, even when using powerful GPUs. Our analysis shows that 95% of the inference time could be spent on SpMM when running popular GNN models on NVIDIA's advanced V100 GPU. Such SpMM performance bottleneck hinders GNNs' applicability to large-scale problems or the development of more sophisticated GNN models. To address this inference time bottleneck, we introduce ES-SpMM, a cache-first edge sampling mechanism and codesigned SpMM kernel. ES-SpMM uses edge sampling to downsize the graph to fit into GPU's shared memory. It thus reduces the computation cost and improves SpMM's cache locality. To evaluate ES-SpMM's performance, we integrated it with a popular GNN framework, DGL, and tested it using representative GNN models and datasets. Our results show that ES-SpMM outperforms the highly optimized cuSPARSE SpMM kernel by up to 4.35x with no accuracy loss and by 45.3x with less than a 1% accuracy loss.

cs.LG↗

arXiv2020

Wavelength conversion through plasmon-coupled surface states

Surface states generally degrade semiconductor device performance by raising the charge injection barrier height, introducing localized trap states, inducing surface leakage current, and altering the electric potential. Therefore, there has been an endless effort to use various surface passivation treatments to suppress the undesirable impacts of the surface states. We show that the giant built-in electric field created by the surface states can be harnessed to enable passive wavelength conversion without utilizing any nonlinear optical phenomena. Photo-excited surface plasmons are coupled to the surface states to generate an electron gas, which is routed to a nanoantenna array through the giant electric field created by the surface states. The induced current on the nanoantennas, which contains mixing product of different optical frequency components, generates radiation at the beat frequencies of the incident photons. We utilize the unprecedented functionalities of plasmon-coupled surface states to demonstrate passive wavelength conversion of nanojoule optical pulses at a 1550 nm center wavelength to terahertz regime with record-high efficiencies that exceed nonlinear optical methods by 4-orders of magnitude. The presented scheme can be used for optical wavelength conversion to different parts of the electromagnetic spectrum ranging from microwave to infrared regimes by using appropriate optical beat frequencies.

physics.optics↗

arXiv2020

Ergodicity recovery of random walk in heterogeneous disordered media

Significant and persistent trajectory-to-trajectory variance are commonly observed in the particle tracking experiments, which have become a major challenge for the experiment data analysis. In this theoretical paper, we investigate the ergodicity recovery behavior, which helps to clarify the origin and the convergence of trajectory-to-trajectory fluctuation in various heterogeneous disordered media. The concepts of self-averaging and ergodicity are revisited in the context of trajectory analysis. The slow ergodicity recovery and the non-Gaussian diffusion in the annealed disordered media are shown as the consequences of the central limit theorem in different situations. The strange ergodicity recovery behavior is reported in the quenched disordered case, which arises from a localization mechanism. The first-passage approach is introduced to the ergodicity analysis for this case, of which the central limit theorem can be employed and the ergodicity is recovered in the length scale of diffusivity correlation.

cond-mat.stat-mech↗

arXiv2020

Quenched trap model on the extreme landscape: the rise of sub-diffusion and non-Gaussian diffusion

Non-Gaussian diffusion has been intensively studied in recent years, which reflects the dynamic heterogeneity in the disordered media. The recent study on the non-Gaussian diffusion in a static disordered landscape suggests novel phenomena due to the quenched disorder. In this work, we further investigate the random walk in this landscape under various effective temperature $μ$, which continuously modulates the dynamic heterogeneity. We show in the long time limit, the trap dynamics on the landscape is equivalent to the quenched trap model, in which sub-diffusion appears for $μ<1$. The non-Gaussian distribution of displacement has been analytically estimated for short $t$, of which the stretched exponential tail is expected for $μ\neq1$. Due to the localization in the ensemble of trajectory segments, an additional peak arises in $P(x,t)$ around $x=0$ even for $μ>1$. Evolving in different time scales, the peak and the tail of $P(x,t)$ are well split for a wide range of $t$. This theoretical study reveals the connections among the sub-diffusion, non-Gaussian diffusion, and the dynamic heterogeneity in the static disordered medium. It also offers an insight on how the cell would benefit from the quasi-static disordered structures.

cond-mat.stat-mech↗

arXiv2020

Parameter Box: High Performance Parameter Servers for Efficient Distributed Deep Neural Network Training

Most work in the deep learning systems community has focused on faster inference, but arriving at a trained model requires lengthy experiments. Accelerating training lets developers iterate faster and come up with better models. DNN training is often seen as a compute-bound problem, best done in a single large compute node with many GPUs. As DNNs get bigger, training requires going distributed. Distributed deep neural network (DDNN) training constitutes an important workload on the cloud. Larger DNN models and faster compute engines shift the training performance bottleneck from computation to communication. Our experiments show existing DNN training frameworks do not scale in a typical cloud environment due to insufficient bandwidth and inefficient parameter server software stacks.We propose PBox, a balanced, scalable central PS hardware that balances compute and communication resources, and PHub, a high performance parameter server (PS) software design that provides an optimized network stack and a streamlined gradient processing pipeline to benefit common PS setups to utilize PBox. We show that in a typical cloud environment, PBox can achieve up to 3.8x speedup over state-of-the-art designs when training ImageNet. We discuss future directions of integrating PBox with programmable switches for in-network aggregation during training, leveraging the datacenter network topology to reduce bandwidth usage and localize data movement.

cs.DC↗

arXiv2020

Parameter Hub: a Rack-Scale Parameter Server for Distributed Deep Neural Network Training

Distributed deep neural network (DDNN) training constitutes an increasingly important workload that frequently runs in the cloud. Larger DNN models and faster compute engines are shifting DDNN training bottlenecks from computation to communication. This paper characterizes DDNN training to precisely pinpoint these bottlenecks. We found that timely training requires high performance parameter servers (PSs) with optimized network stacks and gradient processing pipelines, as well as server and network hardware with balanced computation and communication resources. We therefore propose PHub, a high performance multi-tenant, rack-scale PS design. PHub co-designs the PS software and hardware to accelerate rack-level and hierarchical cross-rack parameter exchange, with an API compatible with many DDNN training frameworks. PHub provides a performance improvement of up to 2.7x compared to state-of-the-art distributed training techniques for cloud-based ImageNet workloads, with 25% better throughput per dollar.

cs.DC↗

arXiv2018

Non-Gaussian diffusion in static disordered media

Non-Gaussian diffusion is commonly considered as a result of fluctuating diffusivity, which is correlated in time or in space or both. In this work, we investigate the non-Gaussian diffusion in static disordered media via a quenched trap model, where the diffusivity is spatially correlated. Several unique effects due to quenched disorder are reported. We analytically estimate the diffusion coefficient $D_{\text{dis}}$ and its fluctuation over samples of finite size. We show a mechanism of population splitting in the non-Gaussian diffusion. It results in a sharp peak in the distribution of displacement $P(x,t)$ around $x=0$, that has frequently been observed in experiments. We examine the fidelity of the coarse-grained diffusion map, which is reconstructed from particle trajectories. Finally, we propose a procedure to estimate the correlation length in static disordered environments, where the information stored in the sample-to-sample fluctuation has been utilized.

cond-mat.stat-mech↗

arXiv2015

Sample-dependent first-passage time distribution in a disordered medium

Above two dimensions, diffusion of a particle in a medium with quenched random traps is believed to be well-described by the annealed continuous time random walk (CTRW). We propose an approximate expression for the first-passage-time (FPT) distribution in a given sample that enables detailed comparison of the two problems. For a system of finite size, the number and spatial arrangement of deep traps yield significant sample-to-sample variations in the FPT statistics. Numerical simulations of a quenched trap model with power-law sojourn times confirm the existence of two characteristic time scales and a non-self-averaging FPT distribution, as predicted by theory.

cond-mat.stat-mech↗

arXiv2015

Ultrafast terahertz probes of interacting dark excitons in chirality-specific semiconducting single-walled carbon nanotubes

Ultrafast terahertz spectroscopy accesses the {\em dark} excitonic ground state in resonantly-excited (6,5) SWNTs via internal, direct dipole-allowed transitions between lowest lying dark-bright pair state $\sim$6 meV. An analytical model reproduces the response which enables quantitative analysis of transient densities of dark excitons and {\em e-h} plasma, oscillator strength, transition energy renormalization and dynamics. %excitation-induced renormalization. Non-equilibrium, yet stable, quasi-1D quantum states with dark excitonic correlations rapidly emerge even with increasing off-resonance photoexcitation and experience a unique crossover to complex phase-space filling of %a complex distribution between both dark and bright pair states, different from dense 2D/3D excitons influenced by the thermalization, cooling and ionization to free carriers.

cond-mat.mes-hall↗

arXiv2014

One- and two-dimensional photo-imprinted diffraction gratings for manipulating terahertz waves

Emerging technology based on artificial materials containing metallic structures has raised the prospect for unprecedented control of terahertz waves through components like filters, absorbers and polarizers. The functionality of these devices is static by the very nature of their metallic or polaritonic composition, although some degree of tunability can be achieved by incorporating electrically biased semiconductors. Here, we demonstrate a photonic structure by projecting the optical image of a metal mask onto a thin GaAs substrate using a femtosecond pulsed laser source. We show that the resulting high-contrast pattern of photo- excited carriers can create diffractive elements operating in transmission. With the metal mask replaced by a digital micromirror device, our photo-imprinted photonic structures provide a route to terahertz components with reconfigurable functionality.

physics.optics↗

arXiv2014

Broadband terahertz generation from metamaterials

The terahertz spectral regime, ranging from about 0.1 to 15 THz, is one of the least explored yet most technologically transformative spectral regions. One current challenge is to develop efficient and compact terahertz emitters/detectors with a broadband and gapless spectrum that can be tailored for various pump photon energies. Here we demonstrate efficient single-cycle broadband THz generation, ranging from about 0.1 to 4 THz, from a thin layer of split-ring resonators with few tens of nanometers thickness by pumping at the telecommunications wavelength of 1.5 micrometer (200 THz). The terahertz emission arises from exciting the magnetic-dipole resonance of the split-ring resonators and quickly decreases under off-resonance pumping. This, together with pump polarization dependence and power scaling of the terahertz emission, identifies the role of optically induced nonlinear currents in split-ring resonators. We also reveal a giant sheet nonlinear susceptibility $\sim$10$^{-16}$ m$^2$V$^{-1}$ that far exceeds thin films and bulk non-centrosymmetric materials.

cond-mat.mes-hall↗