Design consideration and modelling studies of ultrasound and ultraviolet combined approach for shelf-life enhancement of pine apple juice

Although both ultraviolet (UV) radiation and ultrasound (US) treatment have their capabilities in microbial inactivation, applying any one method alone may require a high dose for complete inactivation, which may affect the sensory and nutritional properties of pineapple juice. Hence, this study was intended to analyse and optimise the effect of combined US and UV treatments on microbial inactivation without affecting the selected quality parameters of pineapple juice. US treatment (33 kHz) was done at three different time intervals, viz. 10 min, 20 min and 30 min., after which, juice samples were subjected to UV treatment for 10 min at three UV dosage levels, viz. 1 J/cm², 1.3 J/cm², and 1.6 J/cm². The samples were evaluated for total colour difference, pH, total soluble solids (TSS), titrable acidity (TA), and ascorbic acid content; total bacterial count and total yeast count; and the standardization of process parameters was done using Response Surface Methodology and Artificial Neural Network. The results showed that the individual, as well as combined treatments, did not significantly impact the physicochemical properties while retaining the quality characteristics. It was observed that combined treatment resulted in 5 log cycle reduction in bacterial and yeast populations while the individual treatment failed. From the optimization studies, it was found that combined US and UV treatments with 22.95 min and1.577 J/cm² ensured a microbiologically safe product while retaining organoleptic quality close to that of fresh juice.     

基于快速神经网络架构搜索的鲁棒图像水印网络算法

为解决深度学习在图像水印算法中计算量大且模型冗余的问题，提高图像水印算法在抵抗噪声、旋转和剪裁等攻击时的鲁棒性，提出基于快速神经网络架构搜索（neural architecture search，NAS）的鲁棒图像水印网络算法。通过多项式分布学习快速神经网络架构搜索算法，在预设的搜索空间中搜索最优网络结构，进行图像水印的高效嵌入与鲁棒提取。首先，将子网络中线性连接的全卷积层设置为独立的神经单元结构，并参数化表示结构单元内节点的连接，预先设定结构单元内每个神经元操作的搜索空间；其次，在完成一个批次的数据集训练后，依据神经元操作中的被采样次数和平均损失函数值动态更新概率；最后，重新训练搜索完成的网络。水印网络模型的参数量较原始网络模型缩减了92%以上，大大缩短了模型训练时间。由于搜索得到的网络结构更为紧凑，本文算法具有较高的时间性能和较好的实验效果，在隐藏图像时，对空域信息的依赖比原始网络更少。对改进前后的2个网络进行了大量鲁棒性实验，对比发现，本文算法在CIFAR-10数据集上对抵抗椒盐噪声和旋转、移除像素行（列）等攻击优势显著；在ImageNet数据集上对抵抗椒盐高斯噪声、旋转、中值滤波、高斯滤波、JPEG压缩、裁剪等攻击优势显著，特别是对随机移除行（列）和椒盐噪声有较强的鲁棒性。     

Hyperparameter on-line learning of stochastic resonance based threshold networks

Aiming at training the feed-forward threshold neural network consisting of nondifferentiable activation functions, the approach of noise injection forms a stochastic resonance based threshold network that can be optimized by various gradient-based optimizers. The introduction of injected noise extends the noise level into the parameter space of the designed threshold network, but leads to a highly non-convex optimization landscape of the loss function. Thus, the hyperparameter on-line learning procedure with respective to network weights and noise levels becomes of challenge. It is shown that the Adam optimizer, as an adaptive variant of stochastic gradient descent, manifests its superior learning ability in training the stochastic resonance based threshold network effectively. Experimental results demonstrate the significant improvement of performance of the designed threshold network trained by the Adam optimizer for function approximation and image classification.     

Universality of deep convolutional neural networks

Deep learning has been widely applied and brought breakthroughs in speech recognition, computer vision, and many other domains. Deep neural network architectures and computational issues have been well studied in machine learning. But there lacks a theoretical foundation for understanding the approximation or generalization ability of deep learning methods generated by the network architectures such as deep convolutional neural networks. Here we show that a deep convolutional neural network (CNN) is universal, meaning that it can be used to approximate any continuous function to an arbitrary accuracy when the depth of the neural network is large enough. This answers an open question in learning theory. Our quantitative estimate, given tightly in terms of the number of free parameters to be computed, verifies the efficiency of deep CNNs in dealing with large dimensional data. Our study also demonstrates the role of convolutions in deep CNNs.     

基于双向循环神经网络的汉语语音识别*

当前基于深度神经网络模型中,虽然其隐含层可设置多层,对复杂问题适应能力强,但每层之间的节点连接是相互独立的,这种结构特性导致了在语音序列中无法利用上下文相关信息来提高识别效果,而传统的循环神经网络虽然做出了改进,但是只能对上文信息进行利用。针对以上问题,该文采用可以同时利用语音序列中上下文相关信息的双向循环神经网络模型与深度神经网络模型相结合,并应用于语音识别。构建具有5层隐含层的模型,其中第3层为双向循环神经网络结构,其他层采用深度神经网络结构。实验结果表明:加入了双向循环神经网络结构的模型与其他模型相比,较好地提高了识别正确率;噪声对双向循环神经网络汉语识别有重要影响,尤其是训练集和测试集附加噪声类型不同时,单一的含噪声语音的训练模型无法适应不同噪声类型的语音识别;调整神经网络模型中隐含层神经元数量后,识别正确率并不是一直随着隐含层中神经元数量的增加而增加,神经元数量数目增加到一定程度后正确率出现了降低的趋势。     

贝叶斯优化卷积神经网络公共场所异常声识别*

针对公共场所异常声的感知和识别问题,提出一种基于贝叶斯优化卷积神经网络的识别方法。提取声信号的Gammatone倒谱系数、倍频程功率谱、短时能量和谱质心,组合成声信号的特征图。构建卷积神经网络作为分类器,利用递增的卷积核设置和池化操作处理不同尺度的特征。基于贝叶斯优化算法优化卷积神经网络的模型参数,对包括火苗噼啪声、婴儿啼哭声、烟花燃放声、玻璃破碎声和警报声的5种公共场所异常声进行识别。该方法的识别结果与基于不同的特征提取和分类器方案得到的识别结果进行比较,结果表明该方法的识别效果优于其他特征提取和分类器方案的识别效果。最后分析了该方法在不同信噪比噪声干扰下的识别结果,验证了该方法的有效性。     

Semiotic Aggregation in Deep Learning

Convolutional neural networks utilize a hierarchy of neural network layers. The statistical aspects of information concentration in successive layers can bring an insight into the feature abstraction process. We analyze the saliency maps of these layers from the perspective of semiotics, also known as the study of signs and sign-using behavior. In computational semiotics, this aggregation operation (known as superization) is accompanied by a decrease of spatial entropy: signs are aggregated into supersign. Using spatial entropy, we compute the information content of the saliency maps and study the superization processes which take place between successive layers of the network. In our experiments, we visualize the superization process and show how the obtained knowledge can be used to explain the neural decision model. In addition, we attempt to optimize the architecture of the neural model employing a semiotic greedy technique. To the extent of our knowledge, this is the first application of computational semiotics in the analysis and interpretation of deep neural networks.     

Neural network representations of quantum many-body states

Machine learning is currently the most active interdisciplinary field having numerous applications;additionally,machine-learning techniques are used to research quantum many-body problems.In this study,we first propose neural network quantum states(NNQSs)with general input observables and explore a few related properties,such as the tensor product and local unitary operation.Second,we determine the necessary and sufficient conditions for the representability of a general graph state using normalized NNQS.Finally,to quantify the approximation degree of a given pure state,we define the best approximation degree using normalized NNQSs.Furthermore,we observe that some 7V-qubit states can be represented by a normalized NNQS,such as separable pure states,Bell states and GHZ states.     

卷积神经网络用于近红外光谱古筝面板木材分级

目前,我国乐器制作行业在古筝面板用木材等级的筛选上主要依赖于技师主观评判,但此法缺少科学理论的依据,效率低,客观性及出材率的提高等方面受到限制,无法满足乐器市场的大量需求。实现古筝面板用木材快速、智能化的分级工作是一个急需解决的课题。近红外光谱非常适用于测量含氢的有机物质。古筝面板木材主要化学成分的化学键均由含氢基团组成,不同等级板材的化学成分存在差异,这些差异反映在近红外光谱中,为判断木材等级提供了可能。同时卷积神经网络对非线性数据具有较强的特征提取能力,所以提出一种应用卷积神经网络模型对光谱数据进行分析的方法,进而判别木材的等级。应用了Savitzky Golay一阶、二阶微分两种预处理方法和核主成分分析、连续投影算法两种数据压缩方法,通过所设计的卷积神经网络模型以样本识别准确率和模型构建过程中的损失值作为判定指标选出最佳预处理和数据压缩方法。为了提高模型提取分析光谱数据的能力和避免过拟合现象,应用了多通道卷积核、批量归一化和early stopping策略,将通过两层卷积层提取的特征信息送入全连接层,从而充分提取剩余信息,通过Softmax函数获得板材的最终预测等级,从而确定了最终模型。最终Savitzky Golay一阶微分和核主成分分析为最佳数据处理方法,同时得出用于区分不同等级的古筝面板用木材的主要关键谱带,分别为1 163~1 243, 1 346~1 375和1 525~1 584 nm。将该模型应用于测试集样本,古筝面板用木材的等级识别准确率为95.5%。实验结果表明所提出的方法可以高效地处理光谱数据,有效识别区分不同等级的古筝面板用木材的关键特征,从而为广阔的乐器市场提供一定的技术支持。