China
Africa
Explore chapters and articles related to this topic
Fundamental Knowledge
Published in Peng Liu, Wang Chao, Computational Advertising, 2020
The above methods show that the optimization methods of neural network are not closely related to its model structure, which makes it possible to develop a general neural network expression and optimization tool. At present, the open source neural network software tools mainly include Tensorflow [47], Caffe [48], and Maxnet [49]. What they have in common is the ability to express a graph representation corresponding to the model structure by employing a relatively convenient programming tool, and the tool itself can perform a BP optimization process. In fact, the solution procedure of neural network is far from that simple; after going through multiple links of propagation, problems such as gradient vanish or gradient explode often crop up, making it important to stress the design of network structure and selection of optimization methods and parameters.
(Q)SAR and Extrapolation
Published in Keith R. Solomon, Theo C.M. Brock, Dick de Zwart, Scott D. Dyer, Leo Posthuma, Sean M. Richards, Hans Sanderson, Paul K. Sibley, Paul J. van den Brink, Extrapolation Practice for Ecotoxicological Effect Characterization of Chemicals, 2008
Hans Sanderson, Scott D. Dyer, J. Vincent Nabholz
The artificial neural network (ANN) is a relatively new technique and possibly the preferred one for current and future (Q)SAR development. Basically, ANNs can be regarded as multinonlinear regression methods. Thus, the neural network software simply multiplies the input by a set of weights that in a nonlinear way transforms the input to an output value. ANNs are a form of a multiprocessor computer system that aim to mimic the way the human brain works with 1) simple processing elements, 2) a high degree of interconnectedness, 3) simple scalar messages, and 4) adaptive interactions between the elements. Roughly speaking, neural networks operate by assigning individual weights to the single descriptors in such a way that the endpoints of the training set are mimicked by the calculation. In this way, the network is trained, which simply means that it uses these examples to establish (learn) the actual relations that exist between the input descriptors and the endpoint by setting these weights. When the relationships between the descriptors and the endpoints are established, the model can subsequently be used for prediction endpoints unknown to the neural network by feeding the network with respective descriptors corresponding to these compounds. The major risk in applying ANN is overtraining, which eventually leads to erroneous predictions, as the model will try to include minor variations in the data of the training set as being significant. ANN can exclusively be applied for interpolation (limited to the model domain). Two of the major advantages of the ANN approach are the generality and applicability of the generated models. Disadvantages include slow data processing, the need for considerable computer power, and a lack of transparency; indeed, ANN models are virtually black box models (Carlsen 2003; Eriksson et al. 2003).
Reinforcement learning based optimal decision making towards product lifecycle sustainability
Published in International Journal of Computer Integrated Manufacturing, 2022
Yang Liu, Miying Yang, Zhengang Guo
In many cases, modelling and making inferences from data-driven problems can be reduced to the combination of supervised learning and unsupervised learning. Some domain-specific considerations do apply when it comes to what models to learn and what learning algorithm to use. Black-box models typically require the least amount of work. In particular, deep neural network software tools such as TensorFlow (Abadi et al. 2016) are fairly mature. On the other hand, data on products can be costly to collect, and some prior knowledge may exist, for example, in the form of physics-based models of product behaviour, or at least on the structure of the problem. This suggests using more tailored models, and perhaps even probabilistic inference methods are needed in case of a major data shortage. However, although mature tools also exist for probabilistic modelling (Carpenter et al. 2017), such domain-specific models typically require more work to implement. Determining which approach is the most suitable has to be done on a case-by-case basis, but black-box models provide a good starting point.