ABLE: Attention based learning for enzyme classification

Classifying proteins into their respective enzyme class is an interesting question for researchers for a variety of reasons. The open source Protein Data Bank (PDB) contains more than 1,60,000 structures, with more being added everyday. This paper proposes an attention-based bidirectional-LSTM model (ABLE) trained on over sampled data generated by SMOTE to analyse and classify a protein into one of the six enzyme classes or a negative class using only the primary structure of the protein described as a string by the FASTA sequence as an input. We achieve the highest F1-score of 0.834 using our proposed model on a dataset of proteins from the PDB. We baseline our model against eighteen other machine learning and deep learning networks, including CNN, LSTM, Bi-LSTM, GRU, and the state-of-the-art DeepEC model. We conduct experiments with two different oversampling techniques, SMOTE and ADASYN. To corroborate the obtained results, we perform extensive experimentation and statistical testing.     

自适应光学系统变形镜控制电压预测

在校正大气湍流畸变波前相差的自适应光学系统中,利用基于Levenberg-Marquardt学习算法的非线性反向传播神经网络技术（LMBP）对变形镜控制电压进行预测。以对受横向风影响的大气湍流畸变波前斜率数据为研究对象,通过数值仿真方法,研究了基于LMBP算法的自适应光学系统变形镜电压非线性预测控制算法。通过实验发现,预测电压和变形镜实际控制电压拟合效果良好。讨论了回溯帧数对预测效果的影响,并与基于递推最小二乘（RLS）算法的线性预测算法进行比较。对比结果表明,基于LMBP算法的非线性电压预测方法比基于递推最小二乘法的线性电压预测方法能更有效地降低系统由伺服延迟引起的误差。

     

网络环境下智能化学习模式构建与实践研究

随着计算机技术和网络技术的飞速发展,以及各种移动设备的应用与普及,网络环境下的学习方式已经成为新时期现代化教学方式发展的新方式.但目前网络环境下的学习方式还存在如上课形式单一、学习资源不足、学生自主学习能力弱、课堂互动较少、各种教学活动不易于控制等问题,那么就急需构建网络环境下新的学习方式.将智能化融入教学过程中不仅能...     

Learning hierarchical spatio-temporal pattern for human activity prediction

Human activity prediction has become increasingly valuable in many applications. This paper, initially from the perspective of cognition science, presents a novel approach to learning a hierarchical spatio-temporal pattern of human activities to predict ongoing activities from videos that contain only the onsets of the activities. Spatio-temporal pattern can be learned by a Hierarchical Self-Organizing Map (HSOM), which consists of two self-organizing maps (i.e., action map and actionlet map) connected via associative links trained by Hebbian learning. Ongoing activities can be predicted by Variable order Markov Model (VMM), which provides the means for capturing both large and small order Markov dependencies based on the training actionlet sequences. Experiments of the proposed method on four challenging 3D action datasets captured by commodity depth cameras show promising results.     

Content-aware image resizing based on random-carving with probability

To improve the running speed of image resizing,a fast content-aware image resizing algorithm was proposed based on the threshold learning and random-carving with probability.Firstly the important map was calculated by combining the graph-based visual saliency map and gradient map.Then the image threshold value was obtained by radial basis function (RBF) neural network learning.And by the threshold,the original image was separated into the protected part and the unprotected part which was corresponding to the important part and the unimportant part of the original image individually.Finally,the two parts were allocated different resizing scales and the random-carving with probability was applied to them respectively.Experiments results show that the proposed algorithm has lower time cost comparing to the state-of-arts algorithms in MSRA image database,and has a better visual perception on image resizing.     

Survey on differential privacy and its progress

With the arrival of the era of big data sharing,data privacy protection issues will be highlighted.Since its introduction in 2006,differential privacy technology has been widely researched in data mining and data publishing.In recent years,Google,Apple and other companies have introduced differential privacy technology into the latest products,and differential privacy technology has become the focus of academia and industry again.Firstly,the traditional centralized model of differential privacy was summarized,from the perspective of analysis of data mining and data released in the differential privacy way.Then the latest local differential privacy regarding data collection and data analysis based on the local model was described,involving crowdsourcing with random response technology,BloomFilter,statistical inference techniques.Finally,the main problems and solutions of differential privacy technology were summarized.     

计算机应用设计3DMAX的学习感受

3DMAX软件是一款专业服务艺术设计的三维设计软件，由美国Autdesk传媒娱乐部开发，运行在Win32和Win64平台的计算机图形软件。当前3DMAX软件运用在多个领域，并由于高性价比、易上手、用户多易交流等优势得到广泛认可，业界甚至可以通过对3DMAX软件能力，去衡量个人或企业的水准。鉴于此，研究3DMAX的运用和学习能够深刻读者对该软件的认识，进一步提升相关学者的软件学习能力。     

Chaos in social learning with multiple true states

Most existing social learning models assume that there is only one underlying true state. In this work, we consider a social learning model with multiple true states, in which agents in different groups receive different signal sequences generated by their corresponding underlying true states. Each agent updates his belief by combining his rational self-adjustment based on the external signals he received and the influence of his neighbors according to their communication. We observe chaotic oscillation in the belief evolution, which implies that neither true state could be learnt correctly by calculating the largest Lyapunov exponents and Hurst exponents.     

Adaptive visual target detection and tracking using weakly supervised incremental appearance learning and RGM-PHD tracker

Multiple visual target tracking is a challenging problem due to various uncertainties including occlusion, miss-detection and noisy measurement. Most tracking approaches utilize an object-specific detector, pre-trained on many labeled images, to provide suitable measurements for their tracking system. In this paper, we use a simple background subtraction detector which only needs the background image to localize targets independent of their shape or type. In order to cope with the uncertainties resulted by the detector, we propose an adaptive appearance model and develop an incremental appearance learning algorithm to learn the target appearances in time. The proposed method employs the background information and our defined keypoints’ miss-matched history to adapt the target appearances within different frames. Furthermore, we combine Refined Gaussian Mixture Probability Hypothesis Density (RGM-PHD) tracker with the detectors to keep target trajectories and handle uncertainties. The experiments conducted on several video datasets show the effectiveness of our proposed method.