Search

Wild giant pandas, an endangered species exclusive to China, are a focus of conservation efforts. The behavior of giant pandas reflects their health conditions and activity capabilities, which play an important role in formulating and implementing conservation measures. Researching and developing efficient behavior recognition methods based on deep learning can significantly advance the study of wild giant panda behavior. This study introduces, for the first time, a transformer-based behavior recognition method termed PandaFormer, which employs time-spatial attention to analyze behavioral temporal patterns and estimate activity spaces. The method integrates advanced techniques such as cross-fusion recurrent time encoding and transformer modules, which handle both temporal dynamics and spatial relationships within panda behavior videos. First, we design cross-fusion recurrent time encoding to represent the occurrence time of behaviors effectively. By leveraging the multimodal processing capability of the transformer, we input time and video tokens into the transformer module to explore the relation between behavior and occurrence time. Second, we introduce relative temporal weights between video frames to enable the model to learn sequential relationships. Finally, considering the fixed position of the camera during recording, we propose a spatial attention mechanism based on the estimation of the panda's activity area. To validate the effectiveness of the model, a video dataset of wild giant pandas, encompassing five typical behaviors, was constructed. The proposed method is evaluated on this video-level annotated dataset. It achieves a Top-1 accuracy of 92.25 % and a mean class precision of 91.19 %, surpassing state-of-the-art behavior recognition algorithms by a large margin. Furthermore, the ablation experiments validate the effectiveness of the proposed temporal and spatial attention mechanisms. In conclusion, the proposed method offers an effective way of studying panda behavior and holds potential for application to other wildlife species.

Job scheduling of high performance cluster is a crucial task that affects the efficiency and performance of the system. The accuracy of job runtime prediction is one of the key factors that influences the quality of job scheduling. In this paper, we propose a novel method for job runtime prediction based on Transformer with plain connection and attention mechanism. The proposed method utilizes the job category information obtained by clustering the historical log datasets, and selects six-dimensional features that are highly correlated with job runtime. We divide the datasets into multiple job sets according to the length of job runtime, train and predict each job set separately. We evaluate the proposed method on the HPC2N dataset, and compare it with several existing methods. The results show that the proposed method achieves an average accuracy of 0.892, with 15.2% MAPE, and outperforms other methods in terms of prediction performance and training time. The proposed method can be applied to improve the efficiency and quality of job scheduling in high performance cluster. [ABSTRACT FROM AUTHOR]

Wild giant pandas, an endangered species exclusive to China, are a focus of conservation efforts. The behavior of giant pandas reflects their health conditions and activity capabilities, which play an important role in formulating and implementing conservation measures. Researching and developing efficient behavior recognition methods based on deep learning can significantly advance the study of wild giant panda behavior. This study introduces, for the first time, a transformer-based behavior recognition method termed PandaFormer, which employs time-spatial attention to analyze behavioral temporal patterns and estimate activity spaces. The method integrates advanced techniques such as cross-fusion recurrent time encoding and transformer modules, which handle both temporal dynamics and spatial relationships within panda behavior videos. First, we design cross-fusion recurrent time encoding to represent the occurrence time of behaviors effectively. By leveraging the multimodal processing capability of the transformer, we input time and video tokens into the transformer module to explore the relation between behavior and occurrence time. Second, we introduce relative temporal weights between video frames to enable the model to learn sequential relationships. Finally, considering the fixed position of the camera during recording, we propose a spatial attention mechanism based on the estimation of the panda's activity area. To validate the effectiveness of the model, a video dataset of wild giant pandas, encompassing five typical behaviors, was constructed. The proposed method is evaluated on this video-level annotated dataset. It achieves a Top-1 accuracy of 92.25 % and a mean class precision of 91.19 %, surpassing state-of-the-art behavior recognition algorithms by a large margin. Furthermore, the ablation experiments validate the effectiveness of the proposed temporal and spatial attention mechanisms. In conclusion, the proposed method offers an effective way of studying panda behavior and holds potential for application to other wildlife species. • Time and video are integrated for wild panda behavior recognition. • The relative order of video frames enhances model performance. • An active region attention module for background noise suppression is presented. [ABSTRACT FROM AUTHOR]

[Display omitted] • A hybrid deep learning model is developed to forecast active and reactive power. • Concept of explicit time encoding is proposed as a factor to improve forecasting. • SHapley Additive exPlanations framework is employed to enhance model interpretability. • Power management is formulated based on SHapley Additive exPlanations findings. Nowadays, residential households, including both consumers and emerging prosumers, have exhibited a growing demand for active/reactive power. This demand surge arises from activities such as charging electrical devices, leveraging flexible resources, and integrating renewable energy sources. To meet this escalating demand effectively, operators must ensure the provision of an ample supply of active/reactive power. Achieving this necessitates the identification of influential factors and the generation of precise forecasts for active/reactive power demand. Hence, this work proposes an efficient hybrid deep learning model consisting of long short-term memory and self-Attention (LSTM-Attention). This model incorporates explicit time encoding to forecast one-hour-ahead consumption of active and reactive power using real-time data from residential units. The integration of both models represents a strategic development for model robustness. Leveraging the inherent strengths of both architectures allows for a synergistic compensation that addresses limitations within each, contributing to an overall effective forecasting model. Moreover, the Shapley Additive Explanations (SHAP) framework was employed for model interpretability, and the investigation underscores the pivotal role of incorporating temporal features into active and reactive power forecasting. SHAP findings can be effectively applied in power management strategies to optimally enhance demand response. Finally, to evaluate the effectiveness of the proposed model, a comprehensive array of performance metrics was employed. The results demonstrate a superior forecast accuracy of the proposed model compared to alternative forecasting models. The proposed model achieved the lowest root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) for active power with a value of 0.0256, 0.0181, and 14.255 %, respectively. The formulated forecasting method can also significantly contribute to the industrial sector by improving the accuracy of active/reactive power forecasting, thereby enhancing model interpretability and identifying the most critical factors. [ABSTRACT FROM AUTHOR]

The prediction of future consumption is vital for building management in general and green buildings in particular. A certified green building should meet an evaluation criterion of resource efficiency throughout its life cycle. This paper presents a hybrid model that combines a profile averaging model and a machine learning model, namely random forest regressor method, for improved far horizon forecasting (10% improvement when the prediction horizon is 1 week). The proposed hybrid model is modular, explainable, and does not entail a large dataset. The feature analysis highlighted the importance of explicit time encoding by feeding the time as an input feature to the forecasting model (importance factor of 35%). [ABSTRACT FROM AUTHOR]

Activity recognition based on channel state information (CSI) plays an increasingly important role in human computer interaction. However most CSI activity recognition systems need to re-collect a large amount of samples and retrain model when they are used in new environments or recognize new types of activities, which greatly reduces the practicality of CSI activity recognition. To address this problem we design an adaptable CSI activity recognition system based on meta-learning, which only needs to fine-tune model with very little train effort when it is used in new environments or recognize new types of activities. Specifically, we first use meta-learning algorithm to get the pre-trained model that adapts to task distribution, when the environment or activity category changes, our system doesn't need to retrain the model and has maximal performance after updates the pre-trained model through one or more gradient steps computed with a small amount of samples from new activities. To prevent the loss of CSI time information after feature extraction with multi-layer CNN, we add time encoding on CSI data as the input of CNN neural network. Considering that CSI data may be labeled incorrectly during labeling process, we improve categorical cross entropy loss(CCE) to enhance the system's robustness to these mislabeled data. We test our system on the gesture dataset and the body activity dataset, and the experimental results show that our system achieves average accuracy of 72 percent with one sample of each new activity and 89.6 percent with five samples of each new activity. [ABSTRACT FROM AUTHOR]

In event-based sensing, many sensors independently and asynchronously emit events when there is a change in their input. Event-based sensing can present significant improvements in power efficiency when compared to traditional sampling, because (1) the output is a stream of events where the important information lies in the timing of the events, and (2) the sensor can easily be controlled to output information only when interesting activity occurs at the input. Moreover, event-based sampling can often provide better resolution than standard uniform sampling. Not only does this occur because individual event-based sensors have higher temporal resolution (Rebecq et al., 2021) it also occurs because the asynchrony of events within a sensor and therefore across sensors allows for less redundant and more informative encoding. We would like to explain how such curious results come about. To do so, we use ideal time encoding machines as a proxy for event-based sensors. We explore time encoding of signals with low rank structure, and apply the resulting theory to video. We then see how the asynchronous firing across time encoding machines can couple spatial sampling density with temporal resolution, leading to better reconstruction, whereas, in frame-based video, temporal resolution depends solely on the frame-rate and spatial resolution solely on the pixel grid used. [ABSTRACT FROM AUTHOR]

Linguistics research shows that languages differ as to how they differentiate future from present events. Economics research finds that when the grammatical structure of a language disassociates the future from the present, speakers of the language also disassociate the future from the present in their behaviors. This study examines how linguistically induced time perception relates to cross-country variation in investment efficiency. We find that underinvestment is less prevalent in countries where there is a weaker time disassociation in the language. The results from both a within-country analysis based on firms headquartered in different regions of Switzerland and an analysis based on the birthplace information of U.S. firms' chief executive officers confirm the relation between languages and investment efficiency. Collectively, the results suggest that time encoding in languages influences speakers' cognition and their investment decisions. This paper was accepted by Shivaram Rajgopal, accounting. [ABSTRACT FROM AUTHOR]

We investigate time encoding as an alternative method to classical sampling, and address the problem of reconstructing classes of non-bandlimited signals from time-based samples. We consider a sampling mechanism based on first filtering the input, before obtaining the timing information using a time encoding machine. Within this framework, we show that sampling by timing is equivalent to a non-uniform sampling problem, where the reconstruction of the input depends on the characteristics of the filter and on its non-uniform shifts. The classes of filters we focus on are exponential and polynomial splines, and we show that their fundamental properties are locally preserved in the context of non-uniform sampling. Leveraging these properties, we then derive sufficient conditions and propose novel algorithms for perfect reconstruction of classes of non-bandlimited signals such as: streams of Diracs, sequences of pulses and piecewise constant signals. Next, we extend these methods to operate with arbitrary filters, and also present simulation results on synthetic noisy data. [ABSTRACT FROM AUTHOR]

The paper presents a novel transmitter architecture for short-range asynchronous wireless communication, applicable to simultaneous multi-user wireless acquisition of biological signals. The analog signal, provided from an analog biosensor, is transformed to time information using an Integral Pulse Frequency Modulator (IPFM) as a Time-Encoding Machine. The IPFM generates a time-encoded unipolar pulse train, maintaining the linear dependence of the output pulse distance on analog input voltage. The system enables continuous acquisition of the signals from multiple sensors in which each transmitter has unique feedback loop delay used for multi-user coding. IPFM pulses trigger the Impulse Radio Ultra-Wideband pulse generator directly, providing two ultra-wideband (UWB) pulses per each IPFM pulse. Due to the lack of internal clock signal and microprocessor-free multi-user coding, the circuitry satisfies the requirements of multi-user coding energy efficiency and size reduction, which are crucial demands in biomedical applications. The proposed Time-Encoded UWB (TE-UWB) transmitter is implemented in 0.18 $\mu{\text{m}}$ CMOS technology. Measurement results of the IPFM transfer function for input voltage ranging from 0.15 to 1.5 ${\rm V}$ are presented, providing the dependence of the IPFM pulse time distance on analog input voltage and power consumption dependence on the input voltage level. For continuous monitoring operation, total power consumption of the transmitter circuitry for the maximum input voltage is 10.8 $\mu{\text{W}}$ , while for the lowest input voltage it increases to 40.48  $\mu{\text{W}}$. The circuit occupies 0.14 ${\text{mm}}^2$. [ABSTRACT FROM AUTHOR]

A fully digital algorithm to shape the spectrum of dead-time distortion in power inverters is presented. Dead-time is required to avoid short circuits of the power source by the legs of a power inverter due to the finite turn-onand turn-offtimes of the switches. Dead-time modifies the pulsewidths of the pulsewidth modulated (PWM) signal causing high harmonic distortion. The proposed approach is based on the time-to-digital conversion of the pulsewidths at the output of the power stage and does not require to measure the sign of the current, which is the preferred approach for dead-time compensation algorithms. The effect of the parasitic capacitance of the switching device that distorts the switching waveform is also analyzed and corrected. Furthermore, the quantization noise produced by digital PWM is also reduced by the proposed approach and has a minimal computational cost. Hardware-in-the-loop simulations are provided to show the effectiveness of the proposed approach, and experimental results using an H-bridge voltage-source inverter are included. A minimum total harmonic distortion plus noise (THD+N) of $\text{0.027}\%$ was achieved for a $\text{1}$ -kHz input driving an inductive load. [ABSTRACT FROM AUTHOR]

This paper presents a compact-area, low-power, highly digital analog-to-digital converter (ADC) for audio applications. The proposed converter is implemented using only oscillators and digital circuitry, without operational amplifiers nor other highly linear circuits. The ADC consists of two twin second-order \Sigma \Delta modulators, which can work both individually or in a pseudodifferential configuration. The proposed system has been implemented in a 0.13- \mu \text{m} standard CMOS technology. The single-ended configuration occupies an active area of 0.02 mm2, is powered at 1.8 V with a current consumption of 155 \mu \text{A}$ , and achieves an A-weighted dynamic range (DR) of 98 dB-A. The pseudodifferential configuration achieves 103 dB-A of A-weighted DR at the expense of doubling the area and power consumption. [ABSTRACT FROM AUTHOR]

• This article reports recent developments in time-to-digital conversion (TDC) techniques. • A comprehensive picture of major trends and advancements of the TDC design is presented. • The architectures, implementations and performance of TDCs including time resolution, conversion range, conversion time, power consumption, and converter nonlinearities are discussed in details. Time-to-digital converters (TDCs) are key components of time-mode circuits and enablers for digital processing of analog signals encoded in time. Since design of time-mode circuits facilitates replacing analog blocks with digital circuitry, TDCs pave the way to hardware-efficient and purely digital architectures in deep-submicron semiconductor technology. The design and implementation of modern TDCs is heterogeneous, shaped in multiple directions, and driven by CMOS process downscaling along with application demands. The substantial research effort is made in more intensive digital implementations, optimization of techniques for high resolution, increasing input range and linearity, reduced conversion time, power consumption and silicon area. The calibration and mismatch-tolerant, as well as anti-PVT-variation and anti-metastability design techniques are of growing importance in order to alleviate imperfections of nanoscale CMOS technologies. The paper surveys recent developments of time-to-digital conversion techniques to give a possibly comprehensive picture of major trends and design advancements. Finally, we highlight TDC design challenges for cutting-edge applications such as All-Digital Phase Locked Loops for high data rate wirelesss communication systems operating in the millimeter-wave band. [ABSTRACT FROM AUTHOR]

In modern Western music, melody is commonly conveyed by pitch changes in the highest-register voice, whereas meter or rhythm is often carried by instruments with lower pitches. An intriguing and recently suggested possibility is that the custom of assigning rhythmic functions to lower-pitch instruments may have emerged because of fundamental properties of the auditory system that result in superior time encoding for low pitches. Here we compare rhythm and synchrony perception between low- and high-frequency tones, using both behavioral and EEG techniques. Both methods were consistent in showing no superiority in time encoding for low over high frequencies. However, listeners were consistently more sensitive to timing differences between two nearly synchronous tones when the high-frequency tone followed the low-frequency tone than vice versa. The results demonstrate no superiority of low frequencies in timing judgments but reveal a robust asymmetry in the perception and neural coding of synchrony that reflects greater tolerance for delays of low- relative to high-frequency sounds than vice versa. We propose that this asymmetry exists to compensate for inherent and variable time delays in cochlear processing, as well as the acoustical properties of sound sources in the natural environment, thereby providing veridical perceptual experiences of simultaneity. [ABSTRACT FROM AUTHOR]

In this brief, a new modulation scheme called sample-and-hold asynchronous sigma-delta modulation (SH-ASDM), the extension of the classical asynchronous sigma-delta scheme (ASDM), is introduced. The SH-ASDM is aimed to provide an instantaneous time encoding instead of the mean value encoding realized in the ASDM. This eliminates the effect of the dynamic range reduction of output pulses lengths, which appears for high-frequency input signals in ASDM. SH-ASDM outperforms the ASDM by about 3 dB in the signal-to-noise ratio quantization characteristics, in respect to the worst case input signal. The new analytic model of ASDM is also proposed, which decomposes this type of time-encoding into a linear filtering, sampling, and nonlinear sample transformation. [ABSTRACT FROM AUTHOR]

Oversampled ADCs based on voltage-controlled oscillators have been analyzed using statistical models inherited from sigma-delta modulation. This letter shows that the discrete Fourier transform of a VCO-ADC output sequence can be calculated analytically for single tone inputs. The calculation is based on the transformation of the VCO output into a pulse frequency modulated signal that can be represented by a trigonometric series. Knowledge of the VCO-ADC output spectrum allows accurate evaluation of the SNDR dependence with the VCO oscillation frequency and gain constant. The SNDR predictions of the proposed model have been compared to behavioral simulations displaying only a deviation of 0.7 dB. [ABSTRACT FROM AUTHOR]

We present algorithms for identifying multidimensional receptive fields directly from spike trains produced by biophysically-grounded neuron models. We demonstrate that only the projection of a receptive field onto the input stimulus space may be perfectly identified and derive conditions under which this identification is possible. We also provide detailed examples of identification of neural circuits incorporating spatiotemporal and spectrotemporal receptive fields. [ABSTRACT FROM AUTHOR]

This paper describes a delay-and-addition cell that enables direct signal processing of pulse with modulation (PWM) encoded signals. The cell can be considered functionally equivalent to a switched capacitor integrator. However, both its input and output are synchronous PWM signals. As a difference to a switched capacitor integrator, the circuit does not require operational amplifiers and is composed of passive RC circuits, switches, comparators and digital logic. Circuit implementation non idealities such as offset and propagation delays have also been analyzed. The main advantage of this circuit is the possibility to operate at a low voltage. The paper shows measurements of a demonstration circuit implementing a first order filter. As an application example, the filter is used to attenuate the quantization noise of a sigma-delta signal, delivering a continuously varying PWM waveform from a synchronous bitstream. [ABSTRACT FROM AUTHOR]

We investigate architectures for time encoding and time decoding of visual stimuli such as natural and synthetic video streams (movies, animation). The architecture for time encoding is akin to models of the early visual system. It consists of a bank of filters in cascade with single-input multi-output neural circuits. Neuron firing is based on either a threshold-and-fire or an integrate-and-fire spiking mechanism with feedback. We show that analog information is represented by the neural circuits as projections on a set of band-limited functions determined by the spike sequence. Under Nyquist-type and frame conditions, the encoded signal can be recovered from these projections with arbitrary precision. For the video time encoding machine architecture, we demonstrate that band-limited video streams of finite energy can be faithfully recovered from the spike trains and provide a stable algorithm for perfect recovery. The key condition for recovery calls for the number of neurons in the population to be above a threshold value. [ABSTRACT FROM AUTHOR]

This paper shows the operating principle and experimental results of a new continuous-time sigma-delta modulator architecture. The proposed modulator does not require a multibit quantizer nor a mismatch-shaping digital-to-analog converter to produce a multibit noise-shaped output. Instead, its quantizer encodes the loop filter output in a binary signal using a time encoding technique similar to pulsewidth modulation. This binary signal is used to generate both the analog feedback loop signal and the digital output. A proof-of-concept chip in 0.35-μm CMOS achieves 10 bits of resolution within a signal bandwidth of 1.2 MHz using a first-order modulator. [ABSTRACT FROM AUTHOR]

Abstract: The parity-violation experiment at Jefferson Lab (Newport News, VA) is designed to determine the contribution of strange/anti-strange quark pairs to the intrinsic properties of the proton. In the forward-angle part of the experiment, the asymmetry in the cross-section was measured for elastic scattering by counting the recoil protons corresponding to the two beam-helicity states. Due to the high accuracy required to measure the few-part-per-million asymmetry, the experiment was based on a custom experimental setup with its own associated electronics and data acquisition (DAQ) system. Highly specialized time-encoding electronics provided time-of-flight spectra for each detector for each helicity state. More conventional electronics, processing only a small fraction of the events, was used for monitoring (mainly FastBus). The time-encoding electronics and the DAQ system have been designed to handle events from the 128 detector pairs at a mean rate of 2MHz per detector pair with low deadtime and with minimal helicity-correlated systematic errors. In this paper, we outline the general architecture and the main features of the electronics and the DAQ system dedicated to forward-angle measurements. [Copyright &y& Elsevier]

The implementation of a protection scheme is central to ecotourism. Drawing upon the theoretical framework of critical discourse analysis, the present study accounts for the way protection action is represented in visitors' books at a Greek reserve (Dadia forest). Specifically, proceeding to content analysis and considering a wide range of linguistic features (vocabulary, syntax, ergativity, aspect, and temporal and illocutionary indices), we study whether visitors, through the way they construct protection acts in their texts, display knowledge about and concern for environmental issues within an ecotourism context. The analysis suggests that visitors seem not to be particularly concerned or informed about the protection action undertaken for Dadia forest. With the green aspect of ecotourism rather neglected, this research empirically verifies that ecotourism conceals a consumerist essence under a green wrapping, functioning ideologically. Moreover, the present study offers qualitative insights into ecotourism research, disclosing how the ideological function of ecotourism is evident in the way visitors represent protection acts, since the version of protection action they construct diffuses their responsibilities for the environment. [ABSTRACT FROM AUTHOR]

We introduce a novel class of time encoding machines (TEMs) that exhibit multiplicative coupling, and, feedforward and feedback. We show that a machine with multiplicative coupling is I/O equivalent with an integrate-and-fire neuron with a variable threshold sequence. The same result holds for a TEM with feed-forward while a machine with feedback is I/O equivalent with an asynchronous sigma/delta modulator with variable thresholds. For all TEMs, an input band-limited signal can be perfectly recovered from the zero crossings of the modulated signal and the threshold sequence. We present the optimal decoding algorithm and give conditions for perfect signal recovery. [ABSTRACT FROM AUTHOR]

Abstract: Time encoding is a mechanism of mapping amplitude information into a time sequence. We show that multichannel time encoding using filter banks and integrate-and-fire neurons provides, under natural conditions, an invertible representation of information, i.e., a sensory stimulus can be recovered from its multidimensional spike train representation loss-free. We describe an algorithm for perfect stimulus recovery and derive conditions that guarantee its convergence. [Copyright &y& Elsevier]

Time encoding is a formal method of mapping amplitude information into a time sequence. We show that under simple conditions, bandlimited stimuli encoded with an integrate-and-fire neuron with an absolute refractory period can be recovered loss-free from the neural spike train at its output. We provide an algorithm for perfect recovery and derive conditions for its convergence. [Copyright &y& Elsevier]

Abstract Time-encoding circuits operate in an asynchronous mode and thus are very suitable for ultra-wideband applications. However, this asynchronous mode leads to nonuniform sampling that requires computationally complex decoding algorithms to recover the input signals. In the encoding and decoding process, many non-idealities in circuits and the computing system can affect the final signal recovery. In this article, the sources of the distortion are analyzed for proper parameter setting. In the analysis, the decoding problem is generalized as a function approximation problem. The characteristics of the bases used in existing algorithms are examined. These bases typically require long time support to reach good frequency property. Long time support not only increases computation complexity, but also increases approximation error when the signal is reconstructed through short patches. Hence, a new approximation basis, the Gaussian basis, which is more compact both in time and frequency domain, is proposed. The reconstruction results from different bases under different parameter settings are compared.