Lazily Aggregated Quantized Gradient Innovation for Communication-Efficient Federated Learning.

This paper focuses on communication-efficient federated learning problem, and develops a novel distributed quantized gradient approach, which is characterized by adaptive communications of the quantized gradients. Specifically, the federated learning builds upon the server-worker infrastructure, where the workers calculate local gradients and upload them to the server; then the server obtain the global gradient by aggregating all the local gradients and utilizes it to update the model parameter. The key idea to save communications from the worker to the server is to quantize gradients as well as skip less informative quantized gradient communications by reusing previous gradients. Quantizing and skipping result in 'lazy' worker-server communications, which justifies the term Lazily Aggregated Quantized (LAQ) gradient. Theoretically, the LAQ algorithm achieves the same linear convergence as the gradient descent in the strongly convex case, while effecting major savings in the communication in terms of transmitted bits and communication rounds. Empirically, extensive experiments using realistic data corroborate a significant communication reduction compared with state-of-the-art gradient- and stochastic gradient-based algorithms.

Fine-Grained Unsupervised Temporal Action Segmentation and Distributed Representation for Skeleton-Based Human Motion Analysis.

Understanding the fine-grained temporal structure of human actions and its semantic interpretation is beneficial to many real-world tasks, such as sports movements, rehabilitation exercises, and daily-life activities analysis. Current action segmentation methods mainly rely on deep neural networks to derive feature embedding of actions from motion data, while works on analyzing human actions in fine-granularity are still lacking due to the lack of clear and generic definitions of subactions and related datasets. On the other hand, the motion representations obtained in current action segmentation methods lack semantic or mathematical interpretability that can be used to evaluate action/subaction similarity in quantitative motion analysis. Toward the goal of fine-grained, interpretable, scalable, and efficient action segmentation, we propose a novel unsupervised action segmentation and distributed representation framework based on intuitive motion primitives defined on pose data. Metrics for comprehensive evaluation of the unsupervised fine-grained action segmentation task performance are proposed, and both public and self-constructed datasets are adopted in the experiments. The results show that the proposed method has good performance and generality across different subjects, datasets, and application scenarios.

Training Data Selection and Optimal Sensor Placement for Deep-Learning-Based Sparse Inertial Sensor Human Posture Reconstruction.

Although commercial motion-capture systems have been widely used in various applications, the complex setup limits their application scenarios for ordinary consumers. To overcome the drawbacks of wearability, human posture reconstruction based on a few wearable sensors have been actively studied in recent years. In this paper, we propose a deep-learning-based sparse inertial sensor human posture reconstruction method. This method uses bidirectional recurrent neural network (Bi-RNN) to build an a priori model from a large motion dataset to build human motion, thereby the low-dimensional motion measurements are mapped to whole-body posture. To improve the motion reconstruction performance for specific application scenarios, two fundamental problems in the model construction are investigated: training data selection and sparse sensor placement. The problem of deep-learning training data selection is to select independent and identically distributed (IID) data for a certain scenario from the accumulated imbalanced motion dataset with sufficient information. We formulate the data selection into an optimization problem to obtain continuous and IID data segments, which comply with a small reference dataset collected from the target scenario. A two-step heuristic algorithm is proposed to solve the data selection problem. On the other hand, the optimal sensor placement problem is studied to exploit most information from partial observation of human movement. A method for evaluating the motion information amount of any group of wearable inertial sensors based on mutual information is proposed, and a greedy searching method is adopted to obtain the approximate optimal sensor placement of a given sensor number, so that the maximum motion information and minimum redundancy is achieved. Finally, the human posture reconstruction performance is evaluated with different training data and sensor placement selection methods, and experimental results show that the proposed method takes advantages in both posture reconstruction accuracy and model training time. In the 6 sensors configuration, the posture reconstruction errors of our model for walking, running, and playing basketball are 7.25°, 8.84°, and 14.13°, respectively.

A K-Medoids based Point-Process Modeling on Neural Spike Transformation using Binless Kernel.

A neural prosthesis is designed to compensate for cognitive functional losses by modeling the information transmission among cortical areas. Existing methods generally build a generalized linear model to approximate the nonlinear transformation among two areas, and use the temporal information of the neural spike with low efficiency. It is essential to efficiently model the nonlinearity embedded in spike generation and transmission for the real-time. This paper proposes a nonlinear point-process model to describe spike-in and spike-out transformation using the theory of reproducing kernel Hilbert space (RKHS) and the binless kernel on spike trains. The binless kernel efficiently maps exact spike timing information to the RKHS to describe nonlinear transformations with global minimum regardless of the weight initialization. A streaming K-medoids algorithm is introduced to select typical and important features in this nonlinear binless kernel for further modeling. We test our model on the nonlinearly generated synthetic neural spike trains, and compare with the existing spike transformation methods, such as Volterra model and staged point-process model. The results show that our model has higher goodness-of-fit evaluated by Kolmogorov-Smirnov test and less variance on the prediction results, which indicates the potential better modeling approach for neural prosthesis application.

Tracking the Time Varying Neural Tuning via Adam on Point Process Observations.

Brain machine interfaces(BMIs) translate the neural activity into the control of movement by understanding how the neural activity responds to the movement intension. However, the neural tuning property, where the modulation depth and preferred direction describe how a neuron responses to stimuli, is time varying gradually and abruptly during the interaction with environment. There has been some research to address such an issue considering either one of the cases, but never address them in a general framework. We propose a novel optimization algorithm based on the point process observations to capture these two changes at the same time. At each time index, the tuning parameter is updated stochastically according to the gradient based Adam searching method, which maximizes the likelihood of point process. Our algorithm is compared with the Adaptive Point Process Estimation (APPE), where the abrupt change is addressed by sampling all the possibilities globally, on synthetic neural data. The results show that our algorithm leads to a better prediction of tuning parameters as well as kinematics over 16.8% and 20% respectively.

Selective capture and rapid identification of Panax notoginseng metabolites in rat faeces by the integration of magnetic molecularly imprinted polymers and high-performance liquid chromatography coupled with orbitrap mass spectrometry.

In the present work, an advanced pretreatment method magnetic molecular imprinted polymers-dispersive solid phase extraction (MMIPs-DSPE) combined with the high sensitivity LTQ-Orbitrap mass spectrometry was applied to the complicated metabolites analysis of Traditional Chinese Medicines (TCMs) in complex matrices. The ginsenoside Rb1 magnetic molecular imprinted polymers (Rb1-MMIPs) were successfully synthesized for specific recognition and selective enrichment of Panax notoginseng saponin metabolites in rat faeces. The polymers were prepared by using Fe3O4@SiO2 as the supporting material, APTES as the functional monomer and TEOS as the cross-linker. The Rb1-MMIPs showed quick separation (10.8 emu/g), large adsorption capacity (636µmol/g), high selectivity and fast binding kinetics (25min). Dispersion solid-phase extraction using Rb1-MMIPs (Rb1-MMIPs-DSPE) integrated with LTQ-Orbitrap MS was applied to fish out and identify saponin metabolites from rat faeces, and totally 58 related compounds were detected within 20min, including 26 PPD-group and 32 PPT-group notoginsenoside metabolites. Parallel tests showed that Rb1-MMIPs-DSPE obtained the lowest matrix effects of 0.98-14.84% and captured the largest number of structural analogues compared with traditional pretreatment methods organic solvent extraction (OSE) and solid phase extraction (SPE).

Selective separation and determination of the synthetic colorants in beverages by magnetic solid-phase dispersion extraction based on a Fe3 O4 /reduced graphene oxide nanocomposite followed by high-performance liquid chromatography with diode array detection.

A facile adsorbent, a nanocomposite of Fe3 O4 and reduced graphene oxide, was fabricated for the selective separation and enrichment of synthetic aromatic azo colorants by magnetic solid-phase dispersion extraction. The nanocomposite was synthesized in a one-step reduction reaction and characterized by atomic force microscopy, scanning electron microscopy, Fourier transform infrared spectroscopy, Raman spectroscopy, X-ray diffraction and Brunauer-Emmett-Teller analysis. The colorants in beverages were quickly adsorbed onto the surface of the nanocomposite with strong π-π interactions between colorants and reduced graphene oxide, and separated with the assistance of an external magnetic field. Moreover, the four colorants in beverages were detected at different wavelengths by high performance liquid chromatography with diode array detection. A linear dependence of peak area was obtained over 0.05-10 µg/mL with the limits of detection of 10.02, 11.90, 10.41, 15.91 ng/mL for tartrazine, allure red, amaranth, and new coccine, respectively (signal to noise = 3). The recoveries for the spiked colorants were in the range of 88.95-95.89% with the relative standard deviation less than 2.66%. The results indicated that the nanocomposite of Fe3 O4 and reduced graphene oxide could be used as an excellent selective adsorbent for aromatic compounds and has potential applications in sample pretreatment.

Adaptive Neural Control of Nonaffine Systems With Unknown Control Coefficient and Nonsmooth Actuator Nonlinearities.

This brief considers the asymptotic tracking problem for a class of high-order nonaffine nonlinear dynamical systems with nonsmooth actuator nonlinearities. A novel transformation approach is proposed, which is able to systematically transfer the original nonaffine nonlinear system into an equivalent affine one. Then, to deal with the unknown dynamics and unknown control coefficient contained in the affine system, online approximator and Nussbaum gain techniques are utilized in the controller design. It is proven rigorously that asymptotic convergence of the tracking error and ultimate uniform boundedness of all the other signals can be guaranteed by the proposed control method. The control feasibility is further verified by numerical simulations.

Vanillin-molecularly targeted extraction of stir bar based on magnetic field induced self-assembly of multifunctional Fe3O4@Polyaniline nanoparticles for detection of vanilla-flavor enhancers in infant milk powders.

A molecularly imprinted stir bar was constructed based on Fe3O4@Polyaniline nanoparticles with magnetic field-induced self-assembly process. The monomer, methacrylic acid, was pre-assembled into the pre-polymers with vanillin as template by the formation of hydrogen bonds. After that, the magnetic complexes were generated by the hydrogen bonding, the hydrophobic and π-π interaction between the pre-polymers and Fe3O4@Polyaniline. The complexes were adsorbed on the surface of magnetic stir bar under the magnetic induction, and the coating of vanillin-molecularly imprinted polymers was generated by the one-step copolymerization basing on the cross linking of ethylene glycol dimethacrylate. The molecular imprinting stir bar showed superior selectivity and fast binding kinetics for vanillin, and was used for the enrichment of vanilla-flavor enhancers (vanillin, ethyl maltol and methyl vanillin) in infant milk powders. The results measured by HPLC-UV exhibited good linear ranges of 0.01-100, 0.02-100 and 0.03-100µgmL(-1) with the limit of detection of 2.5-10.0ngmL(-1), and the recoveries were 94.7-98.9%, 82.1-96.7% and 84.5-93.2% with RSD<7.2% for the three enhancers, respectively.

Fe3O4@rGO doped molecularly imprinted polymer membrane based on magnetic field directed self-assembly for the determination of amaranth.

Based on magnetic field directed self-assembly (MDSA) of Fe3O4@rGO composites, a novel magnetic molecularly imprinted electrochemical sensor (MIES) was fabricated and developed for the determination of the azo dye amaranth. Fe3O4@rGO composites were obtained by a one-step approach involving the initial intercalating of iron ions between the graphene oxide layers via the electrostatic interaction, followed by the reduction with hydrazine hydrate to deposit Fe3O4 nanoparticles onto the reduced oxide graphene nanosheets. In molecular imprinting, the complex including the function monomer of aniline, the template of amaranth and Fe3O4@rGO was pre-assembled through π-π stacking and hydrogen bonding interactions, and then was self-assembled on the surface of magnetic glassy carbon electrode (MGCE) with the help of magnetic field induction before electropolymerization. The structures and morphologies of Fe3O4@rGO and the doped molecularly imprinted polymers (MIPs) were investigated by Fourier transform infrared spectrometer (FT-IR), Raman spectra and scanning electron microscope (SEM). Besides, the characterization by differential pulse voltammetry (DPV) showed that Fe3O4@rGO composites promoted the electrical conductivity of the imprinted sensors when doped into the MIPs. The adsorption isotherms and adsorption kinetics were employed to evaluate the performances of MIES. The detection of amaranth was achieved via the redox probe K3[Fe(CN)6] by blocking the imprinted cavities, which avoided the interferences of oxidation products and analogs of amaranth. Furthermore, the prepared MIES exhibited good sensitivity, selectivity, reproducibility and efficiency for detecting amaranth in fruit drinks. The average recoveries were 93.15-100.81% with the RSD <3.0%.

Universal neural network control of MIMO uncertain nonlinear systems.

In this brief, a continuous tracking control law is proposed for a class of high-order multi-input-multi-output uncertain nonlinear dynamic systems with external disturbance and unknown varying control direction matrix. The proposed controller consists of high-gain feedback, Nussbaum gain matrix selector, online approximator (OLA) model and a robust term. The OLA model is represented by a two-layer neural network. The continuousness of the control signal is guaranteed to relax the requirement for the actuator bandwidth and avoid the incurred chattering effect. Asymptotic tracking performance is achieved theoretically by standard Lyapunov analysis. The control feasibility is also verified in simulation environment.