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.     

Prediction of the drag reduction effect of pulsating pipe flow based on machine learning

Prediction of drag reduction effect caused by pulsating pipe flows is examined using machine learning. First, a large set of flow field data is obtained experimentally by measuring turbulent pipe flows with various pulsation patterns. Consequently, more than 7000 waveforms are applied, obtaining a maximum drag reduction rate and maximum energy saving rate of 38.6% and 31.4%, respectively. The results indicate that the pulsating flow effect can be characterized by the pulsation period and pressure gradient during acceleration and deceleration. Subsequently, two machine learning models are tested to predict the drag reduction rate. The results confirm that the machine learning model developed for predicting the time variation of the flow velocity and differential pressure with respect to the pump voltage can accurately predict the nonlinearity of pressure gradients. Therefore, using this model, the drag reduction effect can be estimated with high accuracy.     

紫色球杆菌视紫红质光谱特性的机器学习研究

近年来，机器学习等人工智能技术被应用于蛋白质工程，其在蛋白质结构、功能预测、催化活性等研究中具有独特优势。在未知蛋白质结构的情况下，将蛋白质序列和功能特性与机器学习相结合，基于序列-活性关系（innovative sequence-activity relationship，ISAR）算法，将蛋白质氨基酸序列数字化，用快速傅里叶变换（fast four transform，FFT）进行预处理，再进行偏最小二乘回归建模，可在数据集较少情况下拟合得到最佳模型。通过机器学习对紫色球杆菌视紫红质（gloeobacter violaceus rhodopsin，GR）的突变体蛋白质氨基酸序列与光谱最大吸收波长进行建模，获得了最佳模型。用最佳索引LEVM760106建模得到的确定系数R² 为0.944，均方误差E为11.64。用小波变换进行的预处理，其R² 虽也约为0.944，但E大于11.64，不及FFT进行的预处理。方法较好地解决了蛋白质序列与功能特性之间的数学建模问题，在蛋白质工程中可为预测更优的突变体提供支持。     

A new family of proximity graphs: Class cover catch digraphs

Motivated by issues in machine learning and statistical pattern classification, we investigate a class cover problem (CCP) with an associated family of directed graphs—class cover catch digraphs (CCCDs). CCCDs are a special case of catch digraphs. Solving the underlying CCP is equivalent to finding a smallest cardinality dominating set for the associated CCCD, which in turn provides regularization for statistical pattern classification. Some relevant properties of CCCDs are studied and a characterization of a family of CCCDs is given.     

Action learning versus strategy learning

This article seeks to ascertain whether the strategy‐learning model of Hanaki, Sethi, Erev, and Peterhansl (2003) better accounts for observed behavior than do the various action‐learning models. It does so by measuring the goodness‐of‐fit of the models' predictions against published experimental results for such games as Coordination, Prisoner's Dilemma, and Chicken. The fit is measured via the mean squared deviation (MSD) between the observed behavior and the one predicted by the model. The results show that, for Chicken, the strategy‐learning model fits the observed data much better than do the action‐learning models. The best action‐learning model, on the other hand, fits the observed data well in Coordination. Overall, the strength of the strategy‐learning model is best shown in games where alternations between the two stage‐game Nash equilibria are often observed in the laboratory experiments. © 2004 Wiley Periodicals, Inc. Complexity 9: 41–50, 2004     

Decentralized iterative learning control for a class of large scale interconnected dynamical systems

The problem of decentralized iterative learning control for a class of large scale interconnected dynamical systems is considered. In this paper, it is assumed that the considered large scale dynamical systems are linear time-varying, and the interconnections between each subsystem are unknown. For such a class of uncertain large scale interconnected dynamical systems, a method is presented whereby a class of decentralized local iterative learning control schemes is constructed. It is also shown that under some given conditions, the constructed decentralized local iterative learning controllers can guarantee the asymptotic convergence of the local output error between the given desired local output and the actual local output of each subsystem through the iterative learning process. Finally, as a numerical example, the system coupled by two inverted pendulums is given to illustrate the application of the proposed decentralized iterative learning control schemes.     

Application of a Near-Optimal Reinforcement Learning Controller to a Robotics Problem in Manufacturing: A Hybrid Approach

Optimization theory provides a framework for determining the best decisions or actions with respect to some mathematical model of a process. This paper focuses on learning to act in a near-optimal manner through reinforcement learning for problems that either have no model or the model is too complex. One approach to solving this class of problems is via approximate dynamic programming. The application of these methods are established primarily for the case of discrete state and action spaces. In this paper we develop efficient methods of learning which act in complex systems with continuous state and action spaces. Monte-Carlo approaches are employed to estimate function values in an iterative, incremental procedure. Derivative-free line search methods are used to obtain a near-optimal action in the continuous action space for a discrete subset of the state space. This near-optimal control policy is then extended to the entire continuous state space via a fuzzy additive model. To compensate for approximation errors, a modified procedure for perturbing the generated control policy is developed. Convergence results under moderate assumptions and stopping criteria are established.     

A new type SVM-projected SVM

Support vector machine (SVM), developed by Vapnik et al., is a new and promising technique for classification and regression and has been proved to be competitive with the best available learning machines in many applications. However, the classification speed of SVM is substantially slower than that of other techniques with similar generalization ability. A new type SVM named projected SVM (PSVM), which is a combination of feature vector selection (FVS) method and linear SVM (LSVM), is proposed in present paper. In PSVM, the FVS method is first used to select a relevant subset (feature vectors, FVs) from the training data, and then both the training data and the test data are projected into the subspace constructed by FVs, and finally linear SVM(LSVM) is applied to classify the projected data. The time required by PSVM to calculate the class of new samples is proportional to the count of FVs. In most cases, the count of FVs is smaller than that of support vectors (SVs), and therefore PSVM is faster than SVM in running. Compared with other speeding-up techniques of SVM, PSVM is proved to possess not only speeding-up ability but also de-noising ability for high-noised data, and is found to be of potential use in mechanical fault pattern recognition.     

Sum-fuzzy implementation of a choice function using artificial learning procedure with fixed fraction

In one if his paper Luo transformed the problem of sum-fuzzy rationality into artificial learning procedure and gave an algorithm which used the learning rule of perception. This paper extends the Luo method for finding a sum-fuzzy implementation of a choice function and offers an algorithm based on the artificial learning procedure with fixed fraction. We also present a concrete example which uses this algorithm.     

THPep: A machine learning-based approach for predicting tumor homing peptides

In the present era, a major drawback of current anti-cancer drugs is the lack of satisfactory specificity towards tumor cells. Despite the presence of several therapies against cancer, tumor homing peptides are gaining importance as therapeutic agents. In this regard, the huge number of therapeutic peptides generated in recent years, demands the need to develop an effective and interpretable computational model for rapidly, effectively and automatically predicting tumor homing peptides. Therefore, a sequence-based approach referred herein as THPep has been developed to predict and analyze tumor homing peptides by using an interpretable random forest classifier in concomitant with amino acid composition, dipeptide composition and pseudo amino acid composition. An overall accuracy and Matthews correlation coefficient of 90.13% and 0.76, respectively, were achieved from the independent test set on an objective benchmark dataset. Upon comparison, it was found that THPep was superior to the existing method and holds high potential as a useful tool for predicting tumor homing peptides. For the convenience of experimental scientists, a web server for this proposed method is provided publicly at http://codes.bio/thpep/.