Model-Based Prediction of an Effective Adhesion Parameter Guiding Multi-Type Cell Segregation

The process of cell-sorting is essential for development and maintenance of tissues. Mathematical modeling can provide the means to analyze the consequences of different hypotheses about the underlying mechanisms. With the Differential Adhesion Hypothesis, Steinberg proposed that cell-sorting is determined by quantitative differences in cell-type-specific intercellular adhesion strengths. An implementation of the Differential Adhesion Hypothesis is the Differential Migration Model by Voss-Böhme and Deutsch. There, an effective adhesion parameter was derived analytically for systems with two cell types, which predicts the asymptotic sorting pattern. However, the existence and form of such a parameter for more than two cell types is unclear. Here, we generalize analytically the concept of an effective adhesion parameter to three and more cell types and demonstrate its existence numerically for three cell types based on in silico time-series data that is produced by a cellular-automaton implementation of the Differential Migration Model. Additionally, we classify the segregation behavior using statistical learning methods and show that the estimated effective adhesion parameter for three cell types matches our analytical prediction. Finally, we demonstrate that the effective adhesion parameter can resolve a recent dispute about the impact of interfacial adhesion, cortical tension and heterotypic repulsion on cell segregation.     

基于Mel频谱值和深度学习网络的鸟声识别算法*

为了增强网络对鸟鸣声信号的特征学习能力并提高识别精度,提出一种基于深度残差收缩网络和扩张卷积的鸟声识别方法。首先,提取鸟鸣声信号的对数梅尔特征及其一阶和二阶差分系数组成logMel特征集作为网络模型的输入;其次,通过深度残差收缩网络自动学习噪声阈值,减少噪声干扰;然后,引入扩张卷积增大卷积核感受野并利用注意力机制使网络更关注关键帧特征;最后,通过双向长短时记忆网络从学到的局部特征中学习长期依赖关系。以百鸟数据birdsdata鸟声库中的19种中国常见鸟类作为实验对象,识别正确率可以达到96.58%,并对比模型在不同信噪比数据下的识别结果,结果表明该模型在噪声环境下的识别效果优于现有模型。     

Discrete Infomax Codes for Supervised Representation Learning

For high-dimensional data such as images, learning an encoder that can output a compact yet informative representation is a key task on its own, in addition to facilitating subsequent processing of data. We present a model that produces discrete infomax codes (DIMCO); we train a probabilistic encoder that yields k-way d-dimensional codes associated with input data. Our model maximizes the mutual information between codes and ground-truth class labels, with a regularization which encourages entries of a codeword to be statistically independent. In this context, we show that the infomax principle also justifies existing loss functions, such as cross-entropy as its special cases. Our analysis also shows that using shorter codes reduces overfitting in the context of few-shot classification, and our various experiments show this implicit task-level regularization effect of DIMCO. Furthermore, we show that the codes learned by DIMCO are efficient in terms of both memory and retrieval time compared to prior methods.     

Select cognitive deficits in Vasoactive Intestinal Peptide deficient mice

Background  

The neuropeptide vasoactive intestinal peptide (VIP) is widely distributed in the adult central nervous system where this peptide functions to regulate synaptic transmission and neural excitability. The expression of VIP and its receptors in brain regions implicated in learning and memory functions, including the hippocampus, cortex, and amygdala, raise the possibility that this peptide may function to modulate learned behaviors. Among other actions, the loss of VIP has a profound effect on circadian timing and may specifically influence the temporal regulation of learning and memory functions.     

Knowledge Reuse of Multi-Agent Reinforcement Learning in Cooperative Tasks

With the development and appliance of multi-agent systems, multi-agent cooperation is becoming an important problem in artificial intelligence. Multi-agent reinforcement learning (MARL) is one of the most effective methods for solving multi-agent cooperative tasks. However, the huge sample complexity of traditional reinforcement learning methods results in two kinds of training waste in MARL for cooperative tasks: all homogeneous agents are trained independently and repetitively, and multi-agent systems need training from scratch when adding a new teammate. To tackle these two problems, we propose the knowledge reuse methods of MARL. On the one hand, this paper proposes sharing experience and policy within agents to mitigate training waste. On the other hand, this paper proposes reusing the policies learned by original teams to avoid knowledge waste when adding a new agent. Experimentally, the Pursuit task demonstrates how sharing experience and policy can accelerate the training speed and enhance the performance simultaneously. Additionally, transferring the learned policies from the N-agent enables the (N+1)–agent team to immediately perform cooperative tasks successfully, and only a minor training resource can allow the multi-agents to reach optimal performance identical to that from scratch.     

An electronic implementation of amoeba anticipation

In nature, the capability of memorizing environmental changes and recalling past events can be observed in unicellular organisms like amoebas. Pershin and Di Ventra have shown that such learning behavior can be mimicked in a simple memristive circuit model consisting of an LC (inductance capacitance) contour and a memristive device. Here, we implement this model experimentally by using an Ag/TiO_2?x /Al memristive device. A theoretical analysis of the circuit is used to gain insight into the functionality of this model and to give advice for the circuit implementation. In this respect, the transfer function, resonant frequency, and damping behavior for a varying resistance of the memristive device are discussed in detail.     

Probabilistic prediction in scale-free networks: diameter changes

In complex systems, responses to small perturbations are too diverse to definitely predict how much they would be, and then such diverse responses can be predicted in a probabilistic way. Here we study such a problem in scale-free networks, for example, the diameter changes by the deletion of a single vertex for various in silico and real-world scale-free networks. We find that the diameter changes are indeed diverse and their distribution exhibits an algebraic decay with an exponent zeta asymptotically. Interestingly, the exponent zeta is robust as zeta approximately 2.2(1) for most scale-free networks and insensitive to the degree exponents gamma as long as 2相似文献   

Detecting Errors with Zero-Shot Learning

Error detection is a critical step in data cleaning. Most traditional error detection methods are based on rules and external information with high cost, especially when dealing with large-scaled data. Recently, with the advances of deep learning, some researchers focus their attention on learning the semantic distribution of data for error detection; however, the low error rate in real datasets makes it hard to collect negative samples for training supervised deep learning models. Most of the existing deep-learning-based error detection algorithms solve the class imbalance problem by data augmentation. Due to the inadequate sampling of negative samples, the features learned by those methods may be biased. In this paper, we propose an AEGAN (Auto-Encoder Generative Adversarial Network)-based deep learning model named SAT-GAN (Self-Attention Generative Adversarial Network) to detect errors in relational datasets. Combining the self-attention mechanism with the pre-trained language model, our model can capture semantic features of the dataset, specifically the functional dependency between attributes, so that no rules or constraints are needed for SAT-GAN to identify inconsistent data. For the lack of negative samples, we propose to train our model via zero-shot learning. As a clean-data tailored model, SAT-GAN tries to recognize error data as outliers by learning the latent features of clean data. In our evaluation, SAT-GAN achieves an average F1-score of 0.95 on five datasets, which yields at least 46.2% F1-score improvement over rule-based methods and outperforms state-of-the-art deep learning approaches in the absence of rules and negative samples.     

Boot Camp: A Novel Intensive Approach to Voice Therapy

It is well known in the disciplines of neurobiology, exercise physiology, motor learning, and psychotherapy that desirable learning and behavior changes occur primarily from practice that involves high-intensity overload, variability, and specificity of training. We propose a novel treatment approach called intensive short-term voice therapy that uses these practice parameters for recalcitrant dysphonia. Intensive short-term voice therapy involves multiple sessions with a variety of clinicians, incorporating multiple simultaneous therapeutic approaches. The intensive short-term voice therapy approach is characterized by voice therapy for 1–4 successive days each with an average of 5 hours of therapy and five clinicians. This form of intensive voice therapy provides rigorous practice, involving not only overload but also opportunities for specificity and individuality thereby facilitating better transfer of learned skills. This article discusses the conceptual, theoretical, and practical foundations of this novel therapy approach.     

E-Learning Performance Prediction: Mining the Feature Space of Effective Learning Behavior

Learning analysis provides a new opportunity for the development of online education, and has received extensive attention from scholars at home and abroad. How to use data and models to predict learners’ academic success or failure and give teaching feedback in a timely manner is a core problem in the field of learning analytics. At present, many scholars use key learning behaviors to improve the prediction effect by exploring the implicit relationship between learning behavior data and grades. At the same time, it is very important to explore the association between categories and prediction effects in learning behavior classification. This paper proposes a self-adaptive feature fusion strategy based on learning behavior classification, aiming to mine the effective E-learning behavior feature space and further improve the performance of the learning performance prediction model. First, a behavior classification model (E-learning Behavior Classification Model, EBC Model) based on interaction objects and learning process is constructed; second, the feature space is preliminarily reduced by entropy weight method and variance filtering method; finally, combined with EBC Model and a self-adaptive feature fusion strategy to build a learning performance predictor. The experiment uses the British Open University Learning Analysis Dataset (OULAD). Through the experimental analysis, an effective feature space is obtained, that is, the basic interactive behavior (BI) and knowledge interaction behavior (KI) of learning behavior category has the strongest correlation with learning performance.And it is proved that the self-adaptive feature fusion strategy proposed in this paper can effectively improve the performance of the learning performance predictor, and the performance index of accuracy(ACC), F1-score(F1) and kappa(K) reach 98.44%, 0.9893, 0.9600. This study constructs E-learning performance predictors and mines the effective feature space from a new perspective, and provides some auxiliary references for online learners and managers.     

Dynamics of self-organization in complex adaptive networks

We study the dynamical behavior of complex adaptive automata during unsupervised learning of periodic training sets. A new technique for their analysis is presented and applied to an adaptive network with distributed memory. We show that with general imput pattern sequences, the system can display behavior that ranges from convergence into simple fixed points and oscillations to chaotic wanderings. We also test the ability of the self-organized automaton to recognize spatial patterns, discriminate between them, and to elicit meaningful information out of noisy inputs. In this configuration we determine that the higher the ratio of excitation to inhibition, the broader the equivalence class into which patterns are put together.     

Task’s Choice: Pruning-Based Feature Sharing (PBFS) for Multi-Task Learning

In most of the existing multi-task learning (MTL) models, multiple tasks’ public information is learned by sharing parameters across hidden layers, such as hard sharing, soft sharing, and hierarchical sharing. One promising approach is to introduce model pruning into information learning, such as sparse sharing, which is regarded as being outstanding in knowledge transferring. However, the above method performs inefficiently in conflict tasks, with inadequate learning of tasks’ private information, or through suffering from negative transferring. In this paper, we propose a multi-task learning model (Pruning-Based Feature Sharing, PBFS) that merges a soft parameter sharing structure with model pruning and adds a prunable shared network among different task-specific subnets. In this way, each task can select parameters in a shared subnet, according to its requirements. Experiments are conducted on three benchmark public datasets and one synthetic dataset; the impact of the different subnets’ sparsity and tasks’ correlations to the model performance is analyzed. Results show that the proposed model’s information sharing strategy is helpful to transfer learning and superior to the several comparison models.     

Scheduling and Power Control for Wireless Multicast Systems via Deep Reinforcement Learning

Multicasting in wireless systems is a natural way to exploit the redundancy in user requests in a content centric network. Power control and optimal scheduling can significantly improve the wireless multicast network’s performance under fading. However, the model-based approaches for power control and scheduling studied earlier are not scalable to large state spaces or changing system dynamics. In this paper, we use deep reinforcement learning, where we use function approximation of the Q-function via a deep neural network to obtain a power control policy that matches the optimal policy for a small network. We show that power control policy can be learned for reasonably large systems via this approach. Further, we use multi-timescale stochastic optimization to maintain the average power constraint. We demonstrate that a slight modification of the learning algorithm allows tracking of time varying system statistics. Finally, we extend the multi-time scale approach to simultaneously learn the optimal queuing strategy along with power control. We demonstrate the scalability, tracking and cross-layer optimization capabilities of our algorithms via simulations. The proposed multi-time scale approach can be used in general large state-space dynamical systems with multiple objectives and constraints, and may be of independent interest.     

Prediction of photonic crystal emitter efficiency using an optimized fuzzy learning approach

Photonic crystals (PC) have attracted much attention over the last decade for their unique ability to control light propagation. Researchers suggested the use of metallic photonic crystal with network topology as high efficiency thermal emitters. A necessary precursor to the deployment of such crystals in practical systems is fast accurate prediction of the emission characteristics and efficiency from a photonic lattice. Conventional models that simulate the photonic response of PC are computationally expensive and can take up to a few hours on several parallel processors to realize the emitter efficiency for a given PC structure. Therefore, a practical design process with trial and error cannot be done in a reasonable amount of time.In this article we suggest the use of a fuzzy learning approach to establish a model that can be used to predict emitter efficiency from such systems. The widely studied metallic PC Lincoln log structure is used as a case study. We show that the proposed method can estimate the efficiency of any PC Lincoln log structure much faster than any existing method and is by no means bound to this specific geometry. The case study presented here was chosen only because of recent high interest in it and the abundance of literature data on the example structure. The learning process using Fuzzy set Theory is explained. A multi-objective optimization method to enhance the fuzzy learning process is also outlined. An exemplar case showing the ability of the proposed model to predict the emitter efficiency of a tungsten PC with a bandgap at (10–11.5 μm) is illustrated. We show that once the fuzzy learning is performed, the proposed method can predict the emitter efficiency with 95% accuracy without the need for any expensive computations.     

Sparse identification method of extracting hybrid energy harvesting system from observed data

Hybrid energy harvesters under external excitation have complex dynamical behavior and the superiority of promoting energy harvesting efficiency. Sometimes, it is difficult to model the governing equations of the hybrid energy harvesting system precisely, especially under external excitation. Accompanied with machine learning, data-driven methods play an important role in discovering the governing equations from massive datasets. Recently, there are many studies of data-driven models done in aspect of ordinary differential equations and stochastic differential equations (SDEs). However, few studies discover the governing equations for the hybrid energy harvesting system under harmonic excitation and Gaussian white noise (GWN). Thus, in this paper, a data-driven approach, with least square and sparse constraint, is devised to discover the governing equations of the systems from observed data. Firstly, the algorithm processing and pseudo code are given. Then, the effectiveness and accuracy of the method are verified by taking two examples with harmonic excitation and GWN, respectively. For harmonic excitation, all coefficients of the system can be simultaneously learned. For GWN, we approximate the drift term and diffusion term by using the Kramers-Moyal formulas, and separately learn the coefficients of the drift term and diffusion term. Cross-validation (CV) and mean-square error (MSE) are utilized to obtain the optimal number of iterations. Finally, the comparisons between true values and learned values are depicted to demonstrate that the approach is well utilized to obtain the governing equations for the hybrid energy harvester under harmonic excitation and GWN.     

Improved Q-Learning Algorithm Based on Approximate State Matching in Agricultural Plant Protection Environment

An Unmanned Aerial Vehicle (UAV) can greatly reduce manpower in the agricultural plant protection such as watering, sowing, and pesticide spraying. It is essential to develop a Decision-making Support System (DSS) for UAVs to help them choose the correct action in states according to the policy. In an unknown environment, the method of formulating rules for UAVs to help them choose actions is not applicable, and it is a feasible solution to obtain the optimal policy through reinforcement learning. However, experiments show that the existing reinforcement learning algorithms cannot get the optimal policy for a UAV in the agricultural plant protection environment. In this work we propose an improved Q-learning algorithm based on similar state matching, and we prove theoretically that there has a greater probability for UAV choosing the optimal action according to the policy learned by the algorithm we proposed than the classic Q-learning algorithm in the agricultural plant protection environment. This proposed algorithm is implemented and tested on datasets that are evenly distributed based on real UAV parameters and real farm information. The performance evaluation of the algorithm is discussed in detail. Experimental results show that the algorithm we proposed can efficiently learn the optimal policy for UAVs in the agricultural plant protection environment.     

Complex network perspective on modelling chaotic systems via machine learning

Recent advances have demonstrated that a machine learning technique known as "reservoir computing" is a significantly effective method for modelling chaotic systems. Going beyond short-term prediction, we show that long-term behaviors of an observed chaotic system are also preserved in the trained reservoir system by virtue of network measurements. Specifically, we find that a broad range of network statistics induced from the trained reservoir system is nearly identical with that of a learned chaotic system of interest. Moreover, we show that network measurements of the trained reservoir system are sensitive to distinct dynamics and can in turn detect the dynamical transitions in complex systems. Our findings further support that rather than dynamical equations, reservoir computing approach in fact provides an alternative way for modelling chaotic systems.     

Temporal context and conditional associative learning

Background  

We investigated how temporal context affects the learning of arbitrary visuo-motor associations. Human observers viewed highly distinguishable, fractal objects and learned to choose for each object the one motor response (of four) that was rewarded. Some objects were consistently preceded by specific other objects, while other objects lacked this task-irrelevant but predictive context.     

Attention-Based Fault-Tolerant Approach for Multi-Agent Reinforcement Learning Systems

The aim of multi-agent reinforcement learning systems is to provide interacting agents with the ability to collaboratively learn and adapt to the behavior of other agents. Typically, an agent receives its private observations providing a partial view of the true state of the environment. However, in realistic settings, the harsh environment might cause one or more agents to show arbitrarily faulty or malicious behavior, which may suffice to allow the current coordination mechanisms fail. In this paper, we study a practical scenario of multi-agent reinforcement learning systems considering the security issues in the presence of agents with arbitrarily faulty or malicious behavior. The previous state-of-the-art work that coped with extremely noisy environments was designed on the basis that the noise intensity in the environment was known in advance. However, when the noise intensity changes, the existing method has to adjust the configuration of the model to learn in new environments, which limits the practical applications. To overcome these difficulties, we present an Attention-based Fault-Tolerant (FT-Attn) model, which can select not only correct, but also relevant information for each agent at every time step in noisy environments. The multihead attention mechanism enables the agents to learn effective communication policies through experience concurrent with the action policies. Empirical results showed that FT-Attn beats previous state-of-the-art methods in some extremely noisy environments in both cooperative and competitive scenarios, much closer to the upper-bound performance. Furthermore, FT-Attn maintains a more general fault tolerance ability and does not rely on the prior knowledge about the noise intensity of the environment.     

Prediction of compounds’ biological function (metabolic pathways) based on functional group composition

Efficient in silico screening approaches may provide valuable hints on biological functions of the compound-candidates, which could help to screen functional compounds either in basic researches on metabolic pathways or drug discovery. Here, we introduce a machine learning method (Nearest Neighbor Algorithm) based on functional group composition of compounds to the analysis of metabolic pathways. This method can quickly map small chemical molecules to the metabolic pathway that they likely belong to. A set of 2,764 compounds from 11 major classes of metabolic pathways were selected for study. The overall prediction rate reached 73.3%, indicating that functional group composition of compounds was really related to their biological metabolic functions.