Domain-specific hardware is becoming a promising topic in the backdrop of improvement slow down for general-purpose processors due to the foreseeable end of Moore's Law. Machine learning, especially ...deep neural networks (DNNs), has become the most dazzling domain witnessing successful applications in a wide spectrum of artificial intelligence (AI) tasks. The incomparable accuracy of DNNs is achieved by paying the cost of hungry memory consumption and high computational complexity, which greatly impedes their deployment in embedded systems. Therefore, the DNN compression concept was naturally proposed and widely used for memory saving and compute acceleration. In the past few years, a tremendous number of compression techniques have sprung up to pursue a satisfactory tradeoff between processing efficiency and application accuracy. Recently, this wave has spread to the design of neural network accelerators for gaining extremely high performance. However, the amount of related works is incredibly huge and the reported approaches are quite divergent. This research chaos motivates us to provide a comprehensive survey on the recent advances toward the goal of efficient compression and execution of DNNs without significantly compromising accuracy, involving both the high-level algorithms and their applications in hardware design. In this article, we review the mainstream compression approaches such as compact model, tensor decomposition, data quantization, and network sparsification. We explain their compression principles, evaluation metrics, sensitivity analysis, and joint-way use. Then, we answer the question of how to leverage these methods in the design of neural network accelerators and present the state-of-the-art hardware architectures. In the end, we discuss several existing issues such as fair comparison, testing workloads, automatic compression, influence on security, and framework/hardware-level support, and give promising topics in this field and the possible challenges as well. This article attempts to enable readers to quickly build up a big picture of neural network compression and acceleration, clearly evaluate various methods, and confidently get started in the right way.
Spiking neural networks (SNNs) are promising in ascertaining brain-like behaviors since spikes are capable of encoding spatio-temporal information. Recent schemes, e.g., pre-training from artificial ...neural networks (ANNs) or direct training based on backpropagation (BP), make the high-performance supervised training of SNNs possible. However, these methods primarily fasten more attention on its spatial domain information, and the dynamics in temporal domain are attached less significance. Consequently, this might lead to the performance bottleneck, and scores of training techniques shall be additionally required. Another underlying problem is that the spike activity is naturally non-differentiable, raising more difficulties in supervised training of SNNs. In this paper, we propose a spatio-temporal backpropagation (STBP) algorithm for training high-performance SNNs. In order to solve the non-differentiable problem of SNNs, an approximated derivative for spike activity is proposed, being appropriate for gradient descent training. The STBP algorithm combines the layer-by-layer spatial domain (SD) and the timing-dependent temporal domain (TD), and does not require any additional complicated skill. We evaluate this method through adopting both the fully connected and convolutional architecture on the static MNIST dataset, a custom object detection dataset, and the dynamic N-MNIST dataset. Results bespeak that our approach achieves the best accuracy compared with existing state-of-the-art algorithms on spiking networks. This work provides a new perspective to investigate the high-performance SNNs for future brain-like computing paradigm with rich spatio-temporal dynamics.
Conductive elastic composites have been used widely in soft electronics and soft robotics. These composites are typically a mixture of conductive fillers within elastomeric substrates. They can sense ...strain via changes in resistance resulting from separation of the fillers during elongation. Thus, most elastic composites exhibit a negative piezoconductive effect, i.e. the conductivity decreases under tensile strain. This property is undesirable for stretchable conductors since such composites may become less conductive during deformation. Here, we report a liquid metal-filled magnetorheological elastomer comprising a hybrid of fillers of liquid metal microdroplets and metallic magnetic microparticles. The composite's resistivity reaches a maximum value in the relaxed state and drops drastically under any deformation, indicating that the composite exhibits an unconventional positive piezoconductive effect. We further investigate the magnetic field-responsive thermal properties of the composite and demonstrate several proof-of-concept applications. This composite has prospective applications in sensors, stretchable conductors, and responsive thermal interfaces.
The interaction between protein and DNA plays an essential function in various critical natural processes, like DNA replication, transcription, splicing, and repair. Studying the binding affinity of ...proteins to DNA helps to understand the recognition mechanism of protein-DNA complexes. Since there are still many limitations on the protein-DNA binding affinity data measured by experiments, accurate and reliable calculation methods are necessarily required. So we put forward a computational approach in this paper, called PreDBA, that can forecast protein-DNA binding affinity effectively by using heterogeneous ensemble models. One hundred protein-DNA complexes are manually collected from the related literature as a data set for protein-DNA binding affinity. Then, 52 sequence and structural features are obtained. Based on this, the correlation between these 52 characteristics and protein-DNA binding affinity is calculated. Furthermore, we found that the protein-DNA binding affinity is affected by the DNA molecule structure of the compound. We classify all protein-DNA compounds into five classifications based on the DNA structure related to the proteins that make up the protein-DNA complexes. In each group, a stacked heterogeneous ensemble model is constructed based on the obtained features. In the end, based on the binding affinity data set, we used the leave-one-out cross-validation to evaluate the proposed method comprehensively. In the five categories, the Pearson correlation coefficient values of our recommended method range from 0.735 to 0.926. We have demonstrated the advantages of the proposed method compared to other machine learning methods and currently existing protein-DNA binding affinity prediction approach.
Unmanned aerial vehicle (UAV)-based multispectral remote sensing has shown great potential for precision agriculture. However, there are many problems in data acquisition, processing and application, ...which have stunted its development. In this study, a narrowband Mini-MCA6 multispectral camera and a sunshine-sensor-equipped broadband Sequoia multispectral camera were mounted on a multirotor micro-UAV. They were used to simultaneously collect multispectral imagery and soil–plant analysis development (SPAD) values of maize at multiple sampling points in the field, in addition to the spectral reflectances of six standard diffuse reflectance panels with different reflectance values (4.5%, 20%, 30%, 40%, 60% and 65%). The accuracies of the reflectance and vegetation indices (VIs) derived from the imagery were compared, and the effectiveness and accuracy of the SPAD prediction from the normalized difference vegetation index (NDVI) and red-edge NDVI (reNDVI) under different nitrogen treatments were examined at the plot level. The results show that the narrowband Mini-MCA6 camera could produce more accurate reflectance values than the broadband Sequoia camera, but only if the appropriate calibration method (the nonlinear subband empirical line method) was adopted, especially in visible (blue, green and red) bands. However, the accuracy of the VIs was not completely dependent on the accuracy of the reflectance, i.e., the NDVI from Mini-MCA6 was slightly better than that from Sequoia, whereas Sequoia produced more accurate reNDVI than did Mini-MCA6. At the plot level, reNDVI performed better than NDVI in SPAD prediction regardless of which camera was employed. Moreover, the reNDVI had relatively low sensitivity to the vegetation coverage and was insignificantly affected by environmental factors (e.g., exposed sandy soil). This study indicates that UAV multispectral remote sensing technology is instructive for precision agriculture, but more effort is needed regarding calibration methods for vegetation, postprocessing techniques and robust quantitative studies.
The establishment of either forest or grassland on degraded cropland has been proposed as an effective method for climate change mitigation because these land use types can increase soil carbon (C) ...stocks. This paper synthesized 135 recent publications (844 observations at 181 sites) focused on the conversion from cropland to grassland, shrubland or forest in China, better known as the ‘Grain‐for‐Green’ Program to determine which factors were driving changes to soil organic carbon (SOC). The results strongly indicate a positive impact of cropland conversion on soil C stocks. The temporal pattern for soil C stock changes in the 0–100 cm soil layer showed an initial decrease in soil C during the early stage (<5 years), and then an increase to net C gains (>5 years) coincident with vegetation restoration. The rates of soil C change were higher in the surface profile (0–20 cm) than in deeper soil (20–100 cm). Cropland converted to forest (arbor) had the additional benefit of a slower but more persistent C sequestration capacity than shrubland or grassland. Tree species played a significant role in determining the rate of change in soil C stocks (conifer < broadleaf, evergreen < deciduous forests). Restoration age was the main factor, not temperature and precipitation, affecting soil C stock change after cropland conversion with higher initial soil C stock sites having a negative effect on soil C accumulation. Soil C sequestration significantly increased with restoration age over the long‐term, and therefore, the large scale of land‐use change under the ‘Grain‐for‐Green’ Program will significantly increase China's C stocks.
In this study, six biomass fuels are washed with deionized water at different temperatures, by a well designed test setup. The effects of water washing on fuel properties, pyrolysis and combustion ...characteristics, and ash fusibility of biomass fuels are studied via fuel analysis, thermogravimetric analysis and ash fusion temperature measurements. A better method for evaluating the removal efficiency of element contained in biomass has been proposed. The results show that potassium, sulfur and chlorine contained in biomass, which may be harmful to boiler operation, can be effectively removed by washing. After washing, the hemicellulose and cellulose peaks move apart in the derivative thermogravimetric profile and the devolatilization begins at a higher temperature during pyrolysis and combustion processes. Washing also delays the char combustion. Except for candlenut wood and rice hull, washing significantly increases the ash fusion temperatures for biomass fuels. As water temperature increases, the removal efficiencies of potassium, amorphous silica and ash increase for all the six biomass fuels, and the temperature intervals between the deformation temperature, softening temperature and fluid temperature decrease only for washed samples of wheat straw and corn stalk.
► Water washing improves fuel properties for all the six biomass fuels. ► The removal efficiencies of K, SiO2 and ash increase as water temperature rises. ► Water washing delays the devolatilization and the char combustion. ► The temperature intervals between DT, ST, and FT decrease as water temperature rises. ► Suggestions for industrial application of water washing are given.
Upon perception by plant cells, the immunity hormone jasmonate (JA) triggers a genome-wide transcriptional program, which is largely regulated by the master transcription factor MYC2. The function of ...MYC2 depends on its physical and functional interaction with MED25, a subunit of the Mediator transcriptional co-activator complex. In addition to interacting with MYC2 and RNA polymerase II for preinitiation complex formation, MED25 also interacts with multiple genetic and epigenetic regulators and controls almost every step of MYC2-dependent transcription, including nuclear hormone receptor activation, epigenetic regulation, mRNA processing, transcriptional termination, and chromatin loop formation. These diversified functions have ascribed MED25 to a signal-processing and signal-integrating center during JA-regulated gene transcription. This review is focused on the interactions of MED25 with diverse transcriptional regulators and how these mechanistic interactions contribute to the initiation, amplification, and fine tuning of the transcriptional output of JA signaling.
Artificial neural networks (ANNs), a popular path towards artificial intelligence, have experienced remarkable success via mature models, various benchmarks, open-source datasets, and powerful ...computing platforms. Spiking neural networks (SNNs), a category of promising models to mimic the neuronal dynamics of the brain, have gained much attention for brain inspired computing and been widely deployed on neuromorphic devices. However, for a long time, there are ongoing debates and skepticisms about the value of SNNs in practical applications. Except for the low power attribute benefit from the spike-driven processing, SNNs usually perform worse than ANNs especially in terms of the application accuracy. Recently, researchers attempt to address this issue by borrowing learning methodologies from ANNs, such as backpropagation, to train high-accuracy SNN models. The rapid progress in this domain continuously produces amazing results with ever-increasing network size, whose growing path seems similar to the development of deep learning. Although these ways endow SNNs the capability to approach the accuracy of ANNs, the natural superiorities of SNNs and the way to outperform ANNs are potentially lost due to the use of ANN-oriented workloads and simplistic evaluation metrics.
In this paper, we take the visual recognition task as a case study to answer the questions of “what workloads are ideal for SNNs and how to evaluate SNNs makes sense”. We design a series of contrast tests using different types of datasets (ANN-oriented and SNN-oriented), diverse processing models, signal conversion methods, and learning algorithms. We propose comprehensive metrics on the application accuracy and the cost of memory & compute to evaluate these models, and conduct extensive experiments. We evidence the fact that on ANN-oriented workloads, SNNs fail to beat their ANN counterparts; while on SNN-oriented workloads, SNNs can fully perform better. We further demonstrate that in SNNs there exists a trade-off between the application accuracy and the execution cost, which will be affected by the simulation time window and firing threshold. Based on these abundant analyses, we recommend the most suitable model for each scenario. To the best of our knowledge, this is the first work using systematical comparisons to explicitly reveal that the straightforward workload porting from ANNs to SNNs is unwise although many works are doing so and a comprehensive evaluation indeed matters. Finally, we highlight the urgent need to build a benchmarking framework for SNNs with broader tasks, datasets, and metrics.
Aucubin (Au), an iridoid glycoside from natural plants, has antioxidative and anti-inflammatory bioactivities; however, its effects on a traumatic brain injury (TBI) model remain unknown. We explored ...the potential role of Au in an H
O
-induced oxidant damage in primary cortical neurons and weight-drop induced-TBI in a mouse model.
In vitro experiments, the various concentrations of Au (50 μg/ml, 100 μg/ml, or 200 μg/ml) were added in culture medium at 0 h and 6 h after neurons stimulated by H
O
(100 μM). After exposed for 12 h, neurons were collected for western blot (WB), immunofluorescence, and M29,79-dichlorodihydrofluorescein diacetate (DCFH-DA) staining. In vivo experiments, Au (20 mg/kg or 40 mg/kg) was administrated intraperitoneally at 30 min, 12 h, 24 h, and 48 h after modeling. Brain water content, neurological deficits, and cognitive functions were measured at specific time, respectively. Cortical tissue around focal trauma was collected for WB, TdT-mediated dUTP Nick-End Labeling (TUNEL) staining, Nissl staining, quantitative real time polymerase chain reaction (q-PCR), immunofluorescence/immunohistochemistry, and enzyme linked immunosorbent assay (ELISA) at 72 h after TBI. RNA interference experiments were performed to determine the effects of nuclear factor erythroid-2 related factor 2 (Nrf2) on TBI mice with Au (40 mg/kg) treatment. Mice were intracerebroventricularly administrated with lentivirus at 72 h before TBI establishment. The cortex was obtained at 72 h after TBI and used for WB and q-PCR.
Au enhanced the translocation of Nrf2 into the nucleus, activated antioxidant enzymes, suppressed excessive generation of reactive oxygen species (ROS), and reduced cell apoptosis both in vitro and vivo experiments. In the mice model of TBI, Au markedly attenuated brain edema, histological damages, and improved neurological and cognitive deficits. Au significantly suppressed high mobility group box 1 (HMGB1)-mediated aseptic inflammation. Nrf2 knockdown in TBI mice blunted the antioxidant and anti-inflammatory neuroprotective effects of the Au.
Taken together, our data suggest that Au provides a neuroprotective effect in TBI mice model by inhibiting oxidative stress and inflammatory responses; the mechanisms involve triggering Nrf2-induced antioxidant system.
PDF
