One-trial learning and biphasic time course of performance in the goldfish

Goldfish (N = 408) spontaneously swam against flowing water into a calm-water well. After a single trial punished by brief electric shock, the fish avoided the well, as indexed by increased latencies of reentry. Avoidance declined during the first minute after shock, then rose to a peak 1 hour later. The biphasic time course is compatible with the two-store theory of memory formation.     

Equipotentiality quantified: the anatomical distribution of the engram

Sensory events and the representation of past experience cause distinctive changes in the electrical activity of widespread regions of the brain. These regions have similar roles in the engram in the sense that they all seem to participate in responses to external events and in subsequent representations of these events. However, the relative contribution of these processes to the activity of different brain regions is quantitatively different, in that some regions are much more strongly affected than others. These results may constitute the basis for reconciliation of localizationist and antilocalizationist views of brain function.     

基于迁移学习的棉花识别

提高智能采棉机效率的一个重要途径是实现单个、重叠和遮挡棉花的识别,避免误采摘和漏采摘。针对不同形态棉花的识别,常规的特征提取方法难以达到令人满意的结果,因而采用基于迁移学习的棉花识别方法和基于迁移模型的特征提取与极限学习机(extreme learning machine,ELM)相结合的方法进行棉花识别研究。首先更改AlexNet、GoogleNet、ResNet-50模型分类层和设置相关参数,用训练好的迁移模型对棉花验证集识别,然后利用训练好的迁移模型进行棉花数据集特征提取,再用训练集的特征训练ELM模型,统计不同隐含层神经元个数的ELM模型对棉花的识别准确率。AlexNet、GoogleNet、ResNet-50迁移模型识别率依次为92.03%、93.19%、93.68%;使用特征提取再与ELM结合的方法,准确率比对应迁移模型分别提高了1.97、1.34、1.55百分点。结果表明,迁移模型对小样本棉花识别也有较高准确率,基于特征提取与ELM相结合的方法可进一步提高准确率。     

基于迁移学习的番茄叶片病害图像分类

针对卷积神经网络对番茄病害识别需训练参数较多,训练非常耗时的问题,将迁移学习应用于AlexNet卷积神经网络,对病害叶片和健康叶片共10种类别的番茄叶片进行分类研究。使用14 529张番茄叶片病害图像,随机选择70%作为训练集,30%作为验证集,对AlexNet卷积神经网络模型结构进行迁移,利用在Imagenet图像数据集上训练成熟的AlexNet模型和其参数对番茄叶片病害识别。在训练过程中,固定低层网络参数不变,微调高层网络参数,将番茄病害图像输入到网络中训练网络高层参数,用训练好的模型对10种类别的番茄叶片分类,并进行了20组试验。结果表明:该算法在训练迭代474次时使网络模型很好的收敛,网络对验证集的测试平均准确率达到95.62%,与从零开始训练的AlexNet卷积神经网络相比,本研究算法缩短了训练时间,平均准确率提高了5.6%。采用迁移学习所建立的病害分类模型能够对10种类别的番茄叶片病害快速准确地分类。     

基于迁移学习的水稻病虫害识别

【目的】水稻病虫害是引起水稻减产的重要因素。准确地识别水稻病虫害类型,及 时采取有效的针对性预防措施,有助于避免因水稻减产带来的经济损失。然而,聚焦于人 脸和花草等常见事物的识别技术,在农业领域特别是水稻病虫害识别领域应用较少,而 目前已有的水稻病虫害识别研究存在数据量小和数据种类不够丰富等问题。【方法】文 章搜集了2.0372 万张水稻病虫害图片,并以此构建了完整的水稻病虫害识别数据集,基 于迁移学习的思想,在ResNet50 的预训练模型基础上构建了一个针对16 种主要水稻病 虫害识别的深度模型。同时,考虑实际应用的需要,搜集了9 928 张其他图片（包括人 像、汽车等）,结合9 675 张水稻病虫害图片,构建了一个二分类数据过滤模型,以此来 避免非水稻病虫害图片被识别为某一类病虫害的不合理结果。【结果】有预训练模型验 证结果的top-1 准确率达到了95.23%,F1 系数为77.83%,相较无预训练模型top-1 准确 率提升了24.51%,F1 系数提升了56.66%。数据过滤模型的过滤准确度达到了99.60%。 【结论】基于迁移学习的水稻病虫害识别模型,使水稻病虫害识别结果更加准确。非水稻病 虫害过滤模型,有效地解决了实际应用中非水稻病虫害图片被错分为某一类水稻病虫害的 问题。     

An examination of "transfer of learning" by nucleic Acid

Nucleic acid extracted from brains of trained animals and injected intraperitoneally into naive animals produced no "transfer of learning" effect on several tasks under many conditions. P(325)-Labeled RNA was not found in the brain after intraperitoneal administration. Even intraventricular injections of nucleic acid produced no "transfer" effect.     

基于残差网络与迁移学习的柑橘类型识别

传统的柑橘分类依靠人工进行辨识再手动完成分拣,整个过程耗费时间且成本高昂。对此提出了一种基于迁移学习与残差网络的柑橘图像分类方法。对Kaggle获取的20738张共8类柑橘的图像按7：3比例进行划分得到数据集。在此数据集上不同网络对于柑橘分类性能差异以及迁移学习对经典卷积模型在图像分类任务中的性能提升进行探究,实验以损失值、精准率、召回率等为性能评价指标。实验结果表明,在多种模型中,残差神经网络能获得比其他网络更高的准确率,使用迁移学习初始化网络参数能显著提高柑橘分类的准确度,降低模型过拟合的概率,实现对8类柑橘的准确识别分类,最终分类准确率达到99.9%,对柑橘分类具有指导意义。     

基于迁移学习的芒果成熟度分类算法研究

成熟度判别是芒果采收及储藏的重要依据.为了适应移动设备硬件算力的限制,对比了传统机器学习方法和迁移学习方法在芒果成熟度鉴别方面的表现,优选最佳模型并开发了芒果成熟度分类软件.试验采集了不同成熟度的小台农图像100张,进行数据扩充后按照8:2划分为训练集和测试集,以准确率、F1值和预测时间作为模型评价指标,分别采用k近邻(K-NearestNeighbor, KNN)、支持向量机(Support vector machine, SVM)、朴素贝叶斯(Naive bayes, NB)、决策树(Decision tree, DT)机器学习算法模型和AlexNet、ResNet18、VGG16、GoogleNet和SqueezeNet迁移学习算法模型进行训练和测试,并对比分析各模型的表现.结果表明：机器学习虽然运算速度快,但分类的准确率明显低于迁移学习,迁移学习的分类准确率均在90%以上.但综合考虑模型的分类准确率和计算能力认为Resnet18表现最佳,它在迭代20次后准确率达到98.75%,而测试时间仅为74.66 ms,优于其它深度学习模型.     

The dynamics of interhemispheric compensatory processes in mental imagery

The capacity to generate and analyze mental visual images is essential for many cognitive abilities. We combined triple-pulse transcranial magnetic stimulation (tpTMS) and repetitive TMS (rTMS) to determine which distinct aspect of mental imagery is carried out by the left and right parietal lobe and to reveal interhemispheric compensatory interactions. The left parietal lobe was predominant in generating mental images, whereas the right parietal lobe was specialized in the spatial comparison of the imagined content. Furthermore, in case of an rTMS-induced left parietal lesion, the right parietal cortex could immediately compensate such a left parietal disruption by taking over the specific function of the left hemisphere.     

基于图像处理和深度迁移学习的芒果果实病状识别

目的 基于计算机层析成像(Computed tomography, CT)设备所得芒果CT序列图像,实现芒果内部品质的无损检测和病状识别分类。方法 利用分段函数法、中值滤波结合双边滤波,实现芒果图像增强;采用局部自适应阈值法,实现二值化处理;采用种子填充法进行区域填充;最后通过差影法准确提取芒果果实内部组织的坏损区域。基于深度迁移学习模型,对未处理和已处理的芒果图像数据开展训练和测试,通过AlexNet和GoogLeNet深度学习网络开展迁移学习,调整超参数完成训练过程的网络微调,在不同模型中对比未处理和已处理的芒果测试集在模型上的分类结果。结果 基于未处理数据集,GoogLeNet模型在学习率为0.0002下训练,Accuracy和Macro-average指标分别为98.79%和98.41%。基于已处理数据集,GoogLeNet模型在学习率为0.0002下训练,Accuracy和Macro-average指标分别为100%和100%。深度迁移学习模型在已处理数据集下的模型分类指标较未处理的数据集下有较大的提升。基于同一数据集且超参数一致时,GoogLeNet网络的分类效果明显优于AlexNet网络。结论 设定学习率为0.0002、迭代轮数为3、最小批值为64,基于GoogLeNet网络开展深度迁移学习训练,将所得模型作为最终的分类模型。     

基于迁移学习的嫩江市主要农作物遥感分类

【目的 】机器学习模型在农作物分类研究领域有着较高精度,但如何利用历史样本用于当前时间的作物分类是一个难点。迁移学习的核心思想在于找到已有数据与新数据之间的相似性,文章旨在探索迁移学习方法使用历史样本进行作物分类的可靠性。【方法 】该文以嫩江市为研究区域,基于实地采样数据与遥感数据,用随机森林（Random Forest,RF）分类器,结合多种遥感指数,对2020—2021年嫩江市玉米与大豆种植区域进行分类;利用动态时间规整方法,以2020—2021年实地采样数据生成2022年的分类样本,用RF对2022年嫩江市的玉米与大豆种植区域进行分类。【结果 】（1）对2020—2021年玉米与大豆种植区域进行分类,RF的平均总体精度达到97.8%。（2）对动态时间规整方法生成的2022年玉米与大豆种植区域进行分类,RF的总体精度达到87.5%。【结论 】基于迁移学习的作物识别方法达到较高精度,具有实践意义,可提高历史时期样本的利用效率。     

Near-synchronous interhemispheric termination of the last glacial maximum in mid-latitudes

Isotopic records from polar ice cores imply globally asynchronous warming at the end of the last glaciation. However, 10Be exposure dates show that large-scale retreat of mid-latitude Last Glacial Maximum glaciers commenced at about the same time in both hemispheres. The timing of retreat is consistent with the onset of temperature and atmospheric CO2 increases in Antarctic ice cores. We suggest that a global trend of rising summer temperatures at the end of the Last Glacial Maximum was obscured in North Atlantic regions by hypercold winters associated with unusually extensive winter sea ice.     

基于迁移学习和卷积神经网络的生菜鲜重估测

为提高设施生产中对各生长阶段生菜鲜重的无损估测精度进而更好地指导生产，提出一种利用生菜冠层图像为输入，基于迁移学习技术和卷积神经网络估测鲜重的方法，对比分析AlexNet、VGG-16、GoogLeNet和ResNet-18模型迁移学习后在生菜鲜重估测任务上的效果；同时，对比不同迁移学习方法对模型性能的影响，通过冻结卷积层和减少全连接层改善模型的参数量和训练速度。结果表明:1)AlexNet和VGG-16两种模型能较好的实现生菜鲜重的估测，AlexNet模型的决定系数R²为0.928 0，标准均方根误差NRMSE为19.08%，VGG-16模型的R²为0.938 0，NRMSE为17.71%，但VGG-16模型存在参数量大训练慢的问题，综合考虑选取AlexNet模型迁移学习后作为生菜鲜重估测模型；2)与全新学习方法相比，在预训练模型基础上对生菜鲜重数据集进行迁移学习，可以明显提升生菜鲜重估测模型的训练速度和准确度；3)冻结卷积层能显著加快模型的训练速度，训练时间可减少18%，减少全连接层在保持精度的前提下能大幅度减少模型的参数量。基于迁移学习的卷积神经网络模型可用于生菜鲜重的快速估测，该方法也可以拓展应用到其他叶类蔬菜的鲜重估测中。     

Auditory evoked response: meaningfulness of stimuli and interhemispheric asymmetry

Interhemispheric asymmetries of different magnitudes were observed in human cortical auditory evoked responses to speech and sound-effect stimuli. The wave with peak asymmetry occurred 100 milliseconds after signal onset. The amount of asymmetry of the amplitude of this wave was related to the meaningfulness to the subject of the auditory stimulus rather than to the mere use of verbal versus nonverbal materials.     

The cellular basis of GABA(B)-mediated interhemispheric inhibition

Interhemispheric inhibition is thought to mediate cortical rivalry between the two hemispheres through callosal input. The long-lasting form of this inhibition is believed to operate via γ-aminobutyric acid type B (GABA(B)) receptors, but the process is poorly understood at the cellular level. We found that the firing of layer 5 pyramidal neurons in rat somatosensory cortex due to contralateral sensory stimulation was inhibited for hundreds of milliseconds when paired with ipsilateral stimulation. The inhibition acted directly on apical dendrites via layer 1 interneurons but was silent in the absence of pyramidal cell firing, relying on metabotropic inhibition of active dendritic currents recruited during neuronal activity. The results not only reveal the microcircuitry underlying interhemispheric inhibition but also demonstrate the importance of active dendritic properties for cortical output.     

基于改进的VGG16网络和迁移学习的水稻氮素营养诊断

为实现水稻氮素营养的快速、准确识别,采用改进的VGG 16网络和迁移学习相结合的水稻氮素营养诊断识别方法,以杂交稻‘两优培九’为试验对象进行田间试验,设置4组不同的施氮水平(施氮量分别为0、210、300和390 kg/hm²),在水稻幼穗分化期和齐穗期,扫描获取水稻叶片图像数据;通过图像预处理方法,对数据进行扩充;构建改进的VGG16和迁移学习相结合的网络模型对水稻叶片图像数据进行氮素营养诊断识别。结果表明：1)在幼穗分化期时,改进的VGG16网络的识别准确率为93.1%,模型大小约为迁移学习VGG16模型的1/6,训练时间约为1 261 s。2)在水稻幼穗分化期和齐穗期,该模型微调后的识别准确率均能达到95%以上。基于迁移学习和改进的VGG16网络所建立的水稻氮素营养诊断模型具有较好的泛化能力,可以预测水稻氮素营养状况,为水稻氮素营养诊断提供参考。     

Selective vocal learning in a sparrow

Male swamp sparrows learn their songs; they fail to learn songs of the sympatric song sparrow. Syllables from tape recordings of both species of sparrow were spliced into an array of swamp sparrow-like and song sparrow-like temporal patterns. Swamp sparrows learned only those songs made of swamp sparrow syllables. They did so irrespective of whether the temporal pattern was swamp sparrow-like or song sparrow-like. Selectivity was retained by birds reared in total isolation from adult conspecific sounds.     

Science centers: a potential for learning

The past decade has witnessed a proliferation of science-technology centers, or public places for informal learning about science and technology. Science centers are the only institutions that can provide the general public with participatory exhibits and experiences, together with an accurate scientific interpretation of the materials that are involved. The dramatic rise in attendance and the enthusiasm of repeat visitors to science centers reflect a strong public demand for exhibits designed to help the visitor develop a better understanding of the contemporary scientific issues of society.     

Improved learning in a large-enrollment physics class

We compared the amounts of learning achieved using two different instructional approaches under controlled conditions. We measured the learning of a specific set of topics and objectives when taught by 3 hours of traditional lecture given by an experienced highly rated instructor and 3 hours of instruction given by a trained but inexperienced instructor using instruction based on research in cognitive psychology and physics education. The comparison was made between two large sections (N = 267 and N = 271) of an introductory undergraduate physics course. We found increased student attendance, higher engagement, and more than twice the learning in the section taught using research-based instruction.