Artificial neural networks have shown great success in solving real-world problems in recent years. Nowadays, most of the widely used neural network algorithms are running on silicon-based computers, where the resourc...
详细信息
Artificial neural networks have shown great success in solving real-world problems in recent years. Nowadays, most of the widely used neural network algorithms are running on silicon-based computers, where the resource requirement and energy consumption become a challenge when the network size grows. In contrast, brain contains trillions of neurons and synapses which naturally processes information with far less energy as compared to silicon-based computers. In vitro biological neural networks created from dissociated neurons may be used for computing and performing machine learning tasks. In this dissertation, learning algorithms are explored for a network of neurons with detailed biological properties with feedforward and recurrent structures. For feedforward networks, a two-layer hybrid bio-silicon platform is constructed and a five-step design method is proposed for the fast development of living neural network algorithms. Neural variations and dynamics are verified by fitting model parameters with biological experimental results. Random connections are generated under different connection probabilities to vary network sparsity. A multi-layer perceptron algorithm is tested with biological constraints to investigate the impact of neural variations and random connections. The results show that a reasonable inference accuracy can be achieved despite the presence of neural variations and random network connections. A new adaptive pre-processing technique is proposed to ensure good learning accuracy with different living neural network sparsity. On the training side, a supervised STDP-based learning algorithm is proposed for networks with biological constraints. For recurrent spiking neural networks (RSNN), temporal dynamics are studied with detailed biological features. An automatic fitting tool is used to match the precise spike timing of the in vitro neurons and the modeled neurons to get the fitted neuron parameters. Model fidelity and learning performance of dif
To achieve a real-time behavior in wireless communication systems, the multi-rate support (MRS) provided by the IEEE 802.11 Wireless LAN standard may reveal particularly advantageous. Unfortunately, the most widesprea...
详细信息
To achieve a real-time behavior in wireless communication systems, the multi-rate support (MRS) provided by the IEEE 802.11 Wireless LAN standard may reveal particularly advantageous. Unfortunately, the most widespread rate adaptation algorithms designed for general purpose applications proved to be unsuitable for the challenging real-time scenario. This has led to the definition of purposely designed algorithms such as RSIN, a rate adaptation technique based on the SNR measurement, which showed very good performance in terms of timeliness and reliability. The goal of this paper is to propose an improvement of RSIN that extends its applicability to a wider range of applications. To this aim, we introduce RSIN-E, an enhanced version of RSIN based on an estimation of the SNR obtained through a learning algorithm. In detail, this paper first provides an exhaustive description of both the proposed learning algorithm and the relevant estimation procedure. Then it presents an extensive performance assessment of RSIN-E carried out via both experimental sessions and simulations. The obtained results confirm the effectiveness of the proposed technique and highlight that its performance figures are comparable with those of RSIN, and significantly better than those of Minstrel, a widespread rate adaptation algorithm adopted by most general purpose applications. (C) 2017 Elsevier B.V. All rights reserved.
In this paper the Linguistic Information Feedback-based Dynamical Fuzzy System (LIFDFS) proposed earlier by the authors is first introduced. The principles of alpha-level sets and Backpropagation Through Time (BTT) ar...
详细信息
ISBN:
(纸本)0780385667
In this paper the Linguistic Information Feedback-based Dynamical Fuzzy System (LIFDFS) proposed earlier by the authors is first introduced. The principles of alpha-level sets and Backpropagation Through Time (BTT) are also briefly discussed We employ these two methods to derive an explicit learning algorithm for the feedback parameters of the LIFDFS. With this training algorithm, our LITDFS becomes a potential candidate in solving real-time modeling and prediction problems.
In this paper, we propose a new learning algorithm for multilayer feedforward neural networks, which converges faster and achieves a better classification accuracy than the conventional backpropagation learning algori...
详细信息
In this paper, we propose a new learning algorithm for multilayer feedforward neural networks, which converges faster and achieves a better classification accuracy than the conventional backpropagation learning algorithm for pattern classification. In the conventional backpropagation learning algorithm, weights are adjusted to reduce the error or cost function that reflects the differences between the computed and the desired outputs. In the proposed learning algorithm, we view each term of the output layer as a function of weights and adjust the weights directly so that the output neurons produce the desired outputs. Experiments with remotely sensed data show the proposed algorithm consistently performs better than the conventional backpropagation learning algorithm in terms of classification accuracy and convergence speed.
High-performance component manufacturing has increasing needs of robotic grinding process that can achieve accurate material removal. This article proposes a novel material removal model for robotic belt grinding of I...
详细信息
High-performance component manufacturing has increasing needs of robotic grinding process that can achieve accurate material removal. This article proposes a novel material removal model for robotic belt grinding of Inconel 718 based on acoustic sensing and machine learning. The sound signal is collected online by an audio sensor during the grinding process. A novel method to identify the idle running period and eliminate noise is developed using discrete wavelet decomposition (DWD) and fast Fourier transformation (FFT). Statistical features are extracted from each clean acoustic signal segment to better represent and quantify grinding process. A new k-fold eXtreme Gradient Boosting (k-fold-XGBoost) algorithm after training and optimization is integrated into the material removal (MR) model. The test results indicate that the values forecasted by the model are consistent with the measured values. The mean absolute percentage error (MAPE) of material removal evaluated by the model is 4.373%, which shows a better performance than the reported results which are in the range of 6.4 to 8.72%. In comparison with other prediction models, such as optimally pruned extreme learning machine and random forest and support vector regression, k-fold-XGBoost model shows superior results for the same datasets. It can be concluded that the proposed method based on acoustic signal and the ensemble learning model is effective in predicting the material removal despite the complicated grinding environment.
A linear function approximation-based reinforcement learning algorithm is proposed for Markov decision processes with infinite horizon risk-sensitive cost. Its convergence is proved using the "o.d.e. method"...
详细信息
A linear function approximation-based reinforcement learning algorithm is proposed for Markov decision processes with infinite horizon risk-sensitive cost. Its convergence is proved using the "o.d.e. method" for stochastic approximation. The scheme is also extended to continuous state space processes.
Ontology is a semantic analysis and calculation model, which has been applied to many subjects. Ontology similarity calculation and ontology mapping are employed as machine learning approaches. The purpose of this pap...
详细信息
Ontology is a semantic analysis and calculation model, which has been applied to many subjects. Ontology similarity calculation and ontology mapping are employed as machine learning approaches. The purpose of this paper is to study the leave-two-out stability of ontology learning algorithm. Several leave-two-out stabilities are defined in ontology learning setting and the relationship among these stabilities are presented. Furthermore, the results manifested reveal that leave-two-out stability is a sufficient and necessary condition for ontology learning algorithm. (C) 2015 Elsevier Ltd. All rights reserved.
A linear programming formulation of the optimal stopping problem for Markov decision processes is approximated using linear function approximation. Using this formulation, a reinforcement learning scheme based on a pr...
详细信息
A linear programming formulation of the optimal stopping problem for Markov decision processes is approximated using linear function approximation. Using this formulation, a reinforcement learning scheme based on a primal-dual method and incorporating a sampling device called 'split sampling' is proposed and analyzed. An illustrative example from option pricing is also included.
Most of the research in reinforcement learning has been on problems with discrete action spaces. However, many control problems require the application of continuous control signals. In this paper, we present a stocha...
详细信息
Most of the research in reinforcement learning has been on problems with discrete action spaces. However, many control problems require the application of continuous control signals. In this paper, we present a stochastic reinforcement learning algorithm for learning functions with continuous outputs using a connectionist network. We define stochastic units that compute their real-valued outputs as a function of random activations generated using the normal distribution. learning takes place by using our algorithm to adjust the two parameters of the normal distribution so as to increase the probability of producing the optimal real value for each input pattern. The performance of the algorithm is studied by using it to learn tasks of varying levels of difficulty. Further, as an example of a potential application, we present a network incorporating these stochastic real-valued units that learns to perform an underconstrained positioning task using a simulated 3 degree-of-freedom robot arm.
A new learning algorithm is presented for supervised learning of multilayered and interconnected neural networks without using a gradient method. First, fictitious teacher signals for the outputs of each hidden unit a...
详细信息
A new learning algorithm is presented for supervised learning of multilayered and interconnected neural networks without using a gradient method. First, fictitious teacher signals for the outputs of each hidden unit are algebraically determined by an error backpropagation (EBP) method. Then, the weight parameters are determined by using an exponentially weighted least squares (EWLS) method. This is called the EBP-EWLS algorithm for a multilayered neural network, For an interconnected neural network, the mathematical description of the neural network is arranged in the form for which the EBP-EWLS algorithm can be applied, Simulation studies have verified the proposed technique.
暂无评论