Feedforward Way Research Articles

Temporal-Enhanced Convolutional Network for Person Re-Identification

We propose a new neural network called Temporal-enhanced Convolutional Network (T-CN) for video-based person reidentification. For each video sequence of a person, a spatial convolutional subnet is first applied to each frame for representing appearance information, and then a temporal convolutional subnet links small ranges of continuous frames to extract local motion information. Such spatial and temporal convolutions together construct our T-CN based representation. Finally, a recurrent network is utilized to further explore global dynamics, followed by temporal pooling to generate an overall feature vector for the whole sequence. In the training stage, a Siamese network architecture is adopted to jointly optimize all the components with losses covering both identification and verification. In the testing stage, our network generates an overall discriminative feature representation for each input video sequence (whose length may vary a lot) in a feed-forward way, and even a simple Euclidean distance based matching can generate good re-identification results. Experiments on the most widely used benchmark datasets demonstrate the superiority of our proposal, in comparison with the state-of-the-art.

Active Disturbance Rejection Adaptive Control of Hydraulic Servo Systems

This paper presents an active disturbance rejection adaptive control scheme via full state feedback for motion control of hydraulic servo systems subjected to both parametric uncertainties and uncertain nonlinearities. The proposed controller is derived by effectively integrating adaptive control with extended state observer via backstepping method. The adaptive law is synthesized to handle parametric uncertainties and the remaining uncertainties are estimated by the extended state observer and then compensated in a feedforward way. The unique features of the proposed controller are that not only the matched uncertainties but also unmatched uncertainties are estimated by constructing two extended state observers, and the parameter adaptation law is driven by both tracking errors and state estimation errors. Since the majority of parametric uncertainties can be reduced by the parameter adaptation, the task of the extended state observer is much alleviated. Consequently, high-gain feedback is avoided and improved tracking performance can be expected. The proposed controller theoretically achieves an asymptotic tracking performance in the presence of parametric uncertainties and constant disturbances. In addition, prescribed transient tracking performance and final tracking accuracy can also be guaranteed when existing time-variant uncertain nonlinearities. Comparative experimental results are obtained to verify the high tracking performance nature of the proposed control strategy.

Synthesis of Predictive Equivalent Consumption Minimization Strategy for Hybrid Electric Vehicles Based on Closed-Form Solution of Optimal Equivalence Factor

Previously, an equivalent consumption minimization strategy (ECMS) was developed that provides near-optimal performance of hybrid vehicles based on an adaptation of equivalence factor from state of charge feedback. However, under real-world driving conditions with uncertainties, such as hilly roads, ECMS requires a predictive scheme utilizing future driving information in order to prevent a loss of optimality. In this paper, we synthesize predictive ECMS in a feedforward way to adjust the equivalence factor based on its theoretical connection with future driving statistics, in a systematic manner. First, a useful noncausal adaptation strategy is extracted from dynamic programming results. Then, the inverse problem is formulated and solved to derive an explicit representation of the constant optimal equivalence factor with justified assumptions. Finally, a causal, predictive adaptation strategy using this closed-form solution is synthesized to mimic the noncausal one, and its effectiveness is evaluated for fuel cell hybrid electric vehicles. Results show that if the predicted statistical information reflects well the future driving conditions, the proposed strategy accurately estimates the constant optimal equivalence factor, including the jump behavior, thereby yielding less than 1.5% loss of fuel optimality. Moreover, this approach is extendible to other configurations.

Active disturbance rejection adaptive control of uncertain nonlinear systems: theory and application

This paper proposes an active disturbance rejection adaptive controller for tracking control of a class of uncertain nonlinear systems with consideration of both parametric uncertainties and uncertain nonlinearities by effectively integrating adaptive control with extended state observer via backstepping method. Parametric uncertainties are handled by the synthesized adaptive law and the remaining uncertainties are estimated by extended state observer and then compensated in a feedforward way. Moreover, both matched uncertainties and unmatched uncertainties can be estimated by constructing an extended state observer for each channel of the considered nonlinear plant. Since parametric uncertainties can be reduced by parameter adaptation, the learning burden of extended state observer is much reduced. Consequently, high-gain feedback is avoided and improved tracking performance can be expected. The proposed controller theoretically guarantees a prescribed transient tracking performance and final tracking accuracy in general while achieving asymptotic tracking when the uncertain nonlinearities are not time-variant. The motion control of a motor-driven robot manipulator is investigated as an application example with some suitable modifications and improvements, and comparative simulation results are obtained to verify the high tracking performance nature of the proposed control strategy.

MyoD reprogramming requires Six1 and Six4 homeoproteins: genome-wide cis-regulatory module analysis.

Myogenic regulatory factors of the MyoD family have the ability to reprogram differentiated cells toward a myogenic fate. In this study, we demonstrate that Six1 or Six4 are required for the reprogramming by MyoD of mouse embryonic fibroblasts (MEFs). Using microarray experiments, we found 761 genes under the control of both Six and MyoD. Using MyoD ChIPseq data and a genome-wide search for Six1/4 MEF3 binding sites, we found significant co-localization of binding sites for MyoD and Six proteins on over a thousand mouse genomic DNA regions. The combination of both datasets yielded 82 genes which are synergistically activated by Six and MyoD, with 96 associated MyoD+MEF3 putative cis-regulatory modules (CRMs). Fourteen out of 19 of the CRMs that we tested demonstrated in Luciferase assays a synergistic action also observed for their cognate gene. We searched putative binding sites on these CRMs using available databases and de novo search of conserved motifs and demonstrated that the Six/MyoD synergistic activation takes place in a feedforward way. It involves the recruitment of these two families of transcription factors to their targets, together with partner transcription factors, encoded by genes that are themselves activated by Six and MyoD, including Mef2, Pbx-Meis and EBF.

Detent Force Compensation for PMLSM Systems Based on Structural Design and Control Method Combination

As direct-drive actuators, permanent-magnet linear synchronous motors (PMLSMs) are widely used in high velocity and high precision applications. The detent force, however, can deteriorate the performance and even excite the mechanical resonance. This paper focuses on a novel detent force compensation scheme for PMLSM systems through a combination of structural design and control method. First, due to the bandwidth constraint of the control system, eliminating high frequency ripples is unfeasible; skewed permanent magnets (PMs) considering an optimal skewing length are designed to suppress high order harmonic components. Second, based on the model of PMLSM with skewed PMs, a linearization observer is derived and applied independently to the velocity controller for further diminishing low-order harmonic components. To facilitate implementation in the digital control system, a discretization method taking account of estimated errors is designed. Through the online calculation, the estimated detent force is injected to the control system in a feedforward way. To tune the proposed scheme properly, the convergence of the algorithm is analyzed by utilizing Lyapunov stability theory. Simulation studies are performed to prove the effectiveness of the proposed method, and experiments are provided to confirm the theoretical analysis and simulation results.

Local connections of excitatory neurons in motor-associated cortical areas of the rat

In spite of recent progress in brain sciences, the local circuit of the cerebral neocortex, including motor areas, still remains elusive. Morphological works on excitatory cortical circuitry from thalamocortical (TC) afferents to corticospinal neurons (CSNs) in motor-associated areas are reviewed here. First, TC axons of motor thalamic nuclei have been re-examined by the single-neuron labeling method. There are middle layer (ML)-targeting and layer (L) 1-preferring TC axon types in motor-associated areas, being analogous to core and matrix types, respectively, of Jones (1998) in sensory areas. However, the arborization of core-like motor TC axons spreads widely and disregards the columnar structure that is the basis of information processing in sensory areas, suggesting that motor areas adopt a different information-processing framework such as area-wide laminar organization. Second, L5 CSNs receive local excitatory inputs not only from L2/3 pyramidal neurons but also from ML spiny neurons, the latter directly processing cerebellar information of core-like TC neurons (TCNs). In contrast, basal ganglia information is targeted to apical dendrites of L2/3 and L5 pyramidal neurons through matrix TCNs. Third, L6 corticothalamic neurons (CTNs) are most densely innervated by ML spiny neurons located just above CTNs. Since CTNs receive only weak connections from L2/3 and L5 pyramidal neurons, the TC recurrent circuit composed of TCNs, ML spiny neurons and CTNs appears relatively independent of the results of processing in L2/3 and L5. It is proposed that two circuits sharing the same TC projection and ML neurons are embedded in the neocortex: one includes L2/3 and L5 neurons, processes afferent information in a feedforward way and sends the processed information to other cortical areas and subcortical regions; and the other circuit participates in a dynamical system of the TC recurrent circuit and may serve as the basis of autonomous activity of the neocortex.

Reentrant Visual Processing Affects Rapid Object Categorization in Natural Scenes

The human visual system is very fast and efficient at extracting information about objects present in complex natural scenes. For example, Thorpe, Fize, and Marlot (1996) showed that participants were able to decide whether a natural scene contained an animal even when the image was presented briefly. Moreover, Rousselet, Fabre-Thorpe, and Thorpe (2002) used the animal versus non-animal categorization task and showed that participants could reliably indicate the side containing an animal when two images were simultaneously flashed in the left and right hemi-fields. Together, these results suggest that a great deal of visual information is processed in a fast and feedforward way so that object categorization in natural scenes can be achieved rapidly. The current study aimed to examine whether recurrent visual processing plays a role in rapid object categorization in natural scenes by using an object substitution masking paradigm. We had participants perform an object categorization task. On each trial, two images were simultaneously presented for 20 ms in the left and right visual fields; participants then had to indicate whether an animal or a vehicle appeared in the target picture, which was surrounded by eight dots. These dots either disappeared simultaneously with the target and distractor pictures or remained on the screen for 200 ms. According to Di Lollo, Enns, and Rensink (2000), object substitution masking occurs when a lateral mask (e.g., small dots) persists beyond the duration of a target, reflecting reentrant processes in vision. Therefore, the trailing dots should not affect categorization performance if rapid object categorization requires no reentrant processing from higher-level visual areas. However, our results showed that the trailing dots impaired participants' categorization performance, suggesting that reentrant visual processing affects rapid object categorization in natural scenes.

Dual-stage HDD head positioning using an H∞ almost disturbance decoupling controller and a tracking differentiator

We adopt in this paper a novel control scheme to achieve fast and accurate head positioning for dual-stage actuated hard disk drive (HDD) servo system with actuator saturation and disturbances. This control scheme consists of a tracking differentiator (TD) to avoid actuator saturation as large as possible and an H ∞ almost disturbance decoupling controller to deal with disturbances and improve the tracking performance. More specifically, the TD aims to provide a smooth reference signal in a feedforward way so as to reduce the system error, and further decrease the control inputs to both the VCM actuator and the micro-actuator such that the saturation problem can be effectively avoided. The H ∞ almost disturbance decoupling controller, when it is applied to control the micro-actuator, is able to almost decouple the disturbance and the controlled output such that satisfactory tracking performance can be achieved. Furthermore, the VCM actuator is controlled by a notch filter in series with a lead compensator so as to stabilize the servo loop. Finally, simulation results in time domain and frequency domain verify the effectiveness of the proposed control scheme.

New insights into action–perception coupling

According to a view that has dominated the field for over a century, the brain programs muscle commands and uses a copy of these commands [efference copy (EC)] to adjust not only resulting motor action but also ongoing perception. This view was helpful in formulating several classical problems of action and perception: (1) the posture-movement problem of how movements away from a stable posture can be made without evoking resistance of posture-stabilizing mechanisms resulting from intrinsic muscle and reflex properties; (2) the problem of kinesthesia or why our sense of limb position is good despite ambiguous positional information delivered by proprioceptive and cutaneous signals; (3) the problem of visual space constancy or why the world is perceived as stable while its retinal image shifts following changes in gaze. On closer inspection, the EC theory actually does not solve these problems in a physiologically feasible way. Here solutions to these problems are proposed based on the advanced formulation of the equilibrium-point hypothesis that suggests that action and perception are accomplished in a common spatial frame of reference selected by the brain from a set of available frames. Experimental data suggest that the brain is also able to translate or/and rotate the selected frame of reference by modifying its major attributes-the origin, metrics and orientation-and thus substantially influence action and perception. Because of this ability, such frames are called physical to distinguish them from symbolic or mathematical frames that are used to describe system behavior without influencing this behavior. Experimental data also imply that once a frame of reference is chosen, its attributes are modified in a feedforward way, thus enabling the brain to act in an anticipatory and predictive manner. This approach is extended to sense of effort, kinesthetic illusions, phantom limb and phantom body phenomena. It also addresses the question of why retinal images of objects are sensed as objects located in the external, physical world, rather than in internal representations of the brain.

Semantic Home Photo Categorization

<para xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> A semantic categorization method for generic home photo is proposed. The main contribution of this paper is to exploit a two-layered classification model incorporating camera metadata with low-level features for multilabel detection. The two-layered support vector machine (SVM) classifiers operate to detect local and global photo semantics in a feed-forward way. The first layer aims to predict likelihood of predefined local photo semantics based on camera metadata and regional low-level visual features. In the second layer, one or more global photo semantics is detected based on the likelihood. To construct classifiers producing <emphasis emphasistype="boldital">a posterior</emphasis> probability, we use a parametric model to fit the output of SVM classifiers to posterior probability. A concept merging process based on a set of semantic-confidence maps is also presented to cope with selecting more likelihood photo semantics on spatially overlapping local regions. Experiment was performed with 3086 photos that come from MPEG-7 visual core experiment two official databases. Results showed that the proposed method would much better capture multiple semantic meanings of home photos, compared to other similar technologies. </para>

Input-driven components of spike-frequency adaptation can be unmasked in vivo.

Spike-frequency adaptation affects the response characteristics of many sensory neurons, and different biophysical processes contribute to this phenomenon. Many cellular mechanisms underlying adaptation are triggered by the spike output of the neuron in a feedback manner (e.g., specific potassium currents that are primarily activated by the spiking activity). In contrast, other components of adaptation may be caused by, in a feedforward way, the sensory or synaptic input, which the neuron receives. Examples include viscoelasticity of mechanoreceptors, transducer adaptation in hair cells, and short-term synaptic depression. For a functional characterization of spike-frequency adaptation, it is essential to understand the dependence of adaptation on the input and output of the neuron. Here, we demonstrate how an input-driven component of adaptation can be uncovered in vivo from recordings of spike trains in an insect auditory receptor neuron, even if the total adaptation is dominated by output-driven components. Our method is based on the identification of different inputs that yield the same output and sudden switches between these inputs. In particular, we determined for different sound frequencies those intensities that are required to yield a predefined steady-state firing rate of the neuron. We then found that switching between these sound frequencies causes transient deviations of the firing rate. These firing-rate deflections are evidence of input-driven adaptation and can be used to quantify how this adaptation component affects the neural activity. Based on previous knowledge of the processes in auditory transduction, we conclude that for the investigated auditory receptor neurons, this adaptation phenomenon is of mechanical origin.

선형 시스템 수동화를 위한 병렬 앞먹임 보상기 설계방법 연구

A passivity-based dynamic output feedback controller design is considered for a finite collection of non-square linear systems. Design of a single controller for a set of plants i.e. simultaneous stabilization is an important issue in the area of robust control design. We first determine a squaring gain matrix and an additional dynamics that is connected to the systems in a feedforward way, then a static passivating control law is designed. Consequently, the actual feedback controller will be the static control law combined with the feedforward dynamics. A necessary and sufficient condition for the existence of the parallel feedforward compensator is given by the static output feedback formulation. In contrast to the previous result [1], a technical condition for constructing the parallel feedforward compensator is removed by proposing a new type of the parallel compensator.

수동화 기법에 의한 비정방 선형 시스템의 강인 제어기 설계

We present a state-space approach to design a passivity-based dynamic output feedback control of a finite collection of non-square linear systems. We first determine a squaring gain matrix and an additional dynamics that is connected to the systems in a feedforward way, then a static passivating (i.e. rendering passive) control law is designed. Consequently, the actual feedback controller will be the static control law combined with the feedforward dynamics. A necessary and sufficient condition for the existence of the parallel feedfornward compensator (PFC) is given by the static output feedback fomulation, which enables to utilize linear matrix inequality (LMI). The effectiveness of the proposed method is illustrated by some examples including the systems which can be stabilized by the proprotional-derivative (PD) control law.

Feed-forward contour integration in primary visual cortex based on asynchronous spike propagation

Most current models of visual contour integration involve iterative lateral or feed-back interactions among neurons in V1 and V2. However, some forms of visual processing are too fast for such time-consuming loops. We propose a model avoiding iterative computation by using the fact that real neurons in the retina or LGN fire asynchronously, with the most activated firing first. Thus, early firing V1 neurons can influence processing of their neighbors which are still integrating information from LGN. By limiting the number of spikes to one per neuron, we show that contour integration can be obtained in a purely feed-forward way.

A quantitative study of auditory-evoked saccadic eye movements in two dimensions.

We investigated the properties of human saccadic eye movements evoked by acoustic stimuli in the two-dimensional frontal plane. These movements proved to be quite accurate, both in azimuth and in elevation, provided the sound source spectrum had a broad bandwidth and a sufficiently long duration. If the acoustic target was a tone, the azimuth of the saccadic end points remained equally accurate, whereas the elevation of the response was related to the frequency of the tone, rather than to the physical position of the target. Saccade elevation accuracy also declined substantially for short-duration noise bursts, although response elevation remained highly correlated with target elevation. The latencies of auditory saccades depended on the amplitude, but not on the direction of the eye movement, suggesting a polar coordinate origin of auditory saccade initiation. We also observed that the trajectories of auditory saccades were often substantially curved. Both a qualitative and a model-based analysis showed that this curvature corrected for errors in the initial direction of the saccade. The latter analysis also suggested that the kinematic properties of auditory saccades could be described by the superposition of two overlapping saccadic eye movements, hypothesized to be based on binaural difference cues and monaural spectral cues in the auditory signal, respectively. It is argued that, although the audio-oculomotor system has to operate in a feedforward way, it must nevertheless have access to an accurate representation of actual and desired eye position. Different models underlying the generation of auditory saccades are discussed.