skip to main content

Search for: All records

Creators/Authors contains: "Xu, Fei"

Note: When clicking on a Digital Object Identifier (DOI) number, you will be taken to an external site maintained by the publisher. Some full text articles may not yet be available without a charge during the embargo (administrative interval).
What is a DOI Number?

Some links on this page may take you to non-federal websites. Their policies may differ from this site.

  1. Abstract Strong gravitational lensing of gravitational wave sources offers a novel probe of both the lens galaxy and the binary source population. In particular, the strong lensing event rate and the time-delay distribution of multiply imaged gravitational-wave binary coalescence events can be used to constrain the mass distribution of the lenses as well as the intrinsic properties of the source population. We calculate the strong lensing event rate for a range of second- (2G) and third-generation (3G) detectors, including Advanced LIGO/Virgo, A+, Einstein Telescope (ET), and Cosmic Explorer (CE). For 3G detectors, we find that ∼0.1% of observed events are expected to be strongly lensed. We predict detections of ∼1 lensing pair per year with A+, and ∼50 pairs per year with ET/CE. These rates are highly sensitive to the characteristic galaxy velocity dispersion, σ * , implying that observations of the rates will be a sensitive probe of lens properties. We explore using the time-delay distribution between multiply imaged gravitational-wave sources to constrain properties of the lenses. We find that 3G detectors would constrain σ * to ∼21% after 5 yr. Finally, we show that the presence or absence of strong lensing within the detected population provides useful insightsmore »into the source redshift and mass distribution out to redshifts beyond the peak of the star formation rate, which can be used to constrain formation channels and their relation to the star formation rate and delay-time distributions for these systems.« less
    Free, publicly-accessible full text available April 1, 2023
  2. Abstract. End-member mixing analysis (EMMA) is a method of interpreting stream water chemistry variations and is widely used for chemical hydrograph separation. It is based on the assumption that stream water is a conservative mixture of varying contributions from well-characterized source solutions (end-members). These end-members are typically identified by collecting samples of potential end-member source waters from within the watershed and comparing these to the observations. Here we introduce a complementary data-driven method (convex hull end-member mixing analysis – CHEMMA) to infer the end-member compositions and their associated uncertainties from the stream water observations alone. The method involves two steps. The first uses convex hull nonnegative matrix factorization (CH-NMF) to infer possible end-member compositions by searching for a simplex that optimally encloses the stream water observations. The second step uses constrained K-means clustering (COP-KMEANS) to classify the results from repeated applications of CH-NMF and analyzes the uncertainty associated with the algorithm. In an example application utilizing the 1986 to 1988 Panola Mountain Research Watershed dataset, CHEMMA is able to robustly reproduce the three field-measured end-members found in previous research using only the stream water chemical observations. CHEMMA also suggests that a fourth and a fifth end-member can be (less robustly) identified. We examine uncertainties inmore »end-member identification arising from non-uniqueness, which is related to the data structure, of the CH-NMF solutions, and from the number of samples using both real and synthetic data. The results suggest that the mixing space can be identified robustly when the dataset includes samples that contain extremely small contributions of one end-member, i.e., samples containing extremely large contributions from one end-member are not necessary but do reduce uncertainty about the end-member composition.« less
  3. Free, publicly-accessible full text available May 6, 2023
  4. This research work explores different machine learning techniques for recognizing the existence of rapport between two people engaged in a conversation, based on their facial expressions. First using artificially generated pairs of correlated data signals, a coupled gated recurrent unit (cGRU) neural network is developed to measure the extent of similarity between the temporal evolution of pairs of time-series signals. By pre-selecting their covariance values (between 0.1 and 1.0), pairs of coupled sequences are generated. Using the developed cGRU architecture, this covariance between the signals is successfully recovered. Using this and various other coupled architectures, tests for rapport (measured by the extent of mirroring and mimicking of behaviors) are conducted on real-life datasets. On fifty-nine (N = 59) pairs of interactants in an interview setting, a transformer based coupled architecture performs the best in determining the existence of rapport. To test for generalization, the models were applied on never-been-seen data collected 14 years prior, also to predict the existence of rapport. The coupled transformer model again performed the best for this transfer learning task, determining which pairs of interactants had rapport and which did not. The experiments and results demonstrate the advantages of coupled architectures for predicting an interactional processmore »such as rapport, even in the presence of limited data.« less
  5. The salient pay-per-use nature of serverless computing has driven its continuous penetration as an alternative computing paradigm for various workloads. Yet, challenges arise and remain open when shifting machine learning workloads to the serverless environment. Specifically, the restriction on the deployment size over serverless platforms combining with the complexity of neural network models makes it difficult to deploy large models in a single serverless function. In this paper, we aim to fully exploit the advantages of the serverless computing paradigm for machine learning workloads targeting at mitigating management and overall cost while meeting the response-time Service Level Objective (SLO). We design and implement AMPS-Inf, an autonomous framework customized for model inferencing in serverless computing. Driven by the cost-efficiency and timely-response, our proposed AMPS-Inf automatically generates the optimal execution and resource provisioning plans for inference workloads. The core of AMPS-Inf relies on the formulation and solution of a Mixed-Integer Quadratic Programming problem for model partitioning and resource provisioning with the objective of minimizing cost without violating response time SLO. We deploy AMPS-Inf on the AWS Lambda platform, evaluate with the state-of-the-art pre-trained models in Keras including ResNet50, Inception-V3 and Xception, and compare with Amazon SageMaker and three baselines. Experimental results demonstratemore »that AMPSInf achieves up to 98% cost saving without degrading response time performance.« less
  6. Optimizing performance for Distributed Deep Neural Network (DDNN) training has recently become increasingly compelling, as the DNN model gets complex and the training dataset grows large. While existing works on communication scheduling mostly focus on overlapping the computation and communication to improve DDNN training performance, the GPU and network resources are still under-utilized in DDNN training clusters. To tackle this issue, in this paper, we design and implement a predictable communication scheduling strategy named Prophet to schedule the gradient transfer in an adequate order, with the aim of maximizing the GPU and network resource utilization. Leveraging our observed stepwise pattern of gradient transfer start time, Prophet first uses the monitored network bandwidth and the profiled time interval among gradients to predict the appropriate number of gradients that can be grouped into blocks. Then, these gradient blocks can be transferred one by one to guarantee high utilization of GPU and network resources while ensuring the priority of gradient transfer (i.e., low-priority gradients cannot preempt high-priority gradients in the network transfer). Prophet can make the forward propagation start as early as possible so as to greedily reduce the waiting (idle) time of GPU resources during the DDNN training process. Prototype experiments withmore »representative DNN models trained on Amazon EC2 demonstrate that Prophet can improve the DDNN training performance by up to 40% compared with the state-of-theart priority-based communication scheduling strategies, yet with negligible runtime performance overhead.« less
  7. Del Bimbo, Alberto ; Cucchiara, Rita ; Sclaroff, Stan ; Farinella, Giovanni M ; Mei, Tao ; Bertini, Marco ; Escalante, Hugo J ; Vezzani, Roberto. (Ed.)
    The volume of online lecture videos is growing at a frenetic pace. This has led to an increased focus on methods for automated lecture video analysis to make these resources more accessible. These methods consider multiple information channels including the actions of the lecture speaker. In this work, we analyze two methods that use spatio-temporal features of the speaker skeleton for action classification in lecture videos. The first method is the AM Pose model which is based on Random Forests with motion-based features. The second is a state-of-the-art action classifier based on a two-stream adaptive graph convolutional network (2S-AGCN) that uses features of both joints and bones of the speaker skeleton. Each video is divided into fixed-length temporal segments. Then, the speaker skeleton is estimated on every frame in order to build a representation for each segment for further classification. Our experiments used the AccessMath dataset and a novel extension which will be publicly released. We compared four state-of-the-art pose estimators: OpenPose, Deep High Resolution, AlphaPose and Detectron2. We found that AlphaPose is the most robust to the encoding noise found in online videos. We also observed that 2S-AGCN outperforms the AM Pose model by using the right domain adaptations.
  8. Abstract The large size and complexity of most fern genomes have hampered efforts to elucidate fundamental aspects of fern biology and land plant evolution through genome-enabled research. Here we present a chromosomal genome assembly and associated methylome, transcriptome and metabolome analyses for the model fern species Ceratopteris richardii . The assembly reveals a history of remarkably dynamic genome evolution including rapid changes in genome content and structure following the most recent whole-genome duplication approximately 60 million years ago. These changes include massive gene loss, rampant tandem duplications and multiple horizontal gene transfers from bacteria, contributing to the diversification of defence-related gene families. The insertion of transposable elements into introns has led to the large size of the Ceratopteris genome and to exceptionally long genes relative to other plants. Gene family analyses indicate that genes directing seed development were co-opted from those controlling the development of fern sporangia, providing insights into seed plant evolution. Our findings and annotated genome assembly extend the utility of Ceratopteris as a model for investigating and teaching plant biology.
    Free, publicly-accessible full text available September 1, 2023