Book

Parallel Distributed Processing: Explorations in the Microstructure of Cognition

Authors:
... The production of the past form both with the internal change and the -t/d suffixation may thus be understood as an intermediate mastery of the internal-change production. L2 speakers may also be influenced by such models as sleep/slept or keep/kept (Rumelhart & McClelland 1986). Rumelhart and McClelland (1986) noted that three of their four responses with this type of past-tense change were verbs ending in /p/, similarly to the sleep/keep models above. ...
... L2 speakers may also be influenced by such models as sleep/slept or keep/kept (Rumelhart & McClelland 1986). Rumelhart and McClelland (1986) noted that three of their four responses with this type of past-tense change were verbs ending in /p/, similarly to the sleep/keep models above. They understand this as the participants' "sensitivity to the regular and subregular patterns of the English past tense" (Rumelhart & McClelland 1986: 34). ...
... Such instances have also been previously described in classic literature as double past markers (in the case of -t-insertions) and using affixes for both present and past (in the case of -s-insertions). In their computer simulation, Rumelhart and McClelland (1986), too, found the double past marking on seven responses to the input word. Even though they attribute such forms to errors that children and adults occasionally make, they also notice that the doubling occurred for those verbs whose stem ended in /p/ or /k/ and whose correct past-tense form should be created by the addition of /t/ and whose stem ended in /p/ or /k/ (e.g. ...
Article
Full-text available
The present study investigates the production of novel morphologically inflected forms in second-language learners of English with Czech as L1. The study attempts to investigate which production model (single-or dual-route) best accounts for L2 learners' morphological productivity when forming regular past forms of novel words. Additionally, it explores the possible interference effects of L1. 88 English L2 learners and 9 native speakers heard sentences in which a new activity was described with a novel word (The baby likes to dize. Look, there it is dizing. Everyday it dizes.) and past-tense forms were elicited (So yesterday it…). The results revealed that for native speakers the likelihood of a verb being produced in a regular past-tense form was inversely related to its phonological similarity to existing irregular verbs (replicating previous studies). L2 speakers showed a development in this direction: While for the A1 to B1 participants similarity to existing irregulars did not matter, B2 and C1 participants appeared to be sensitive to these similarities and behaved comparably to native speakers. In addition to the form analysis, the reaction-times results showed that the lowest language levels used their L1 as a performance facilitator (with slower performance with novel words that do not respect the phonology of the participants' L1), while proficient learners and native speakers were not sensitive to this property of the novel words. The results suggest that the L2 acquisition of the English past-tense is characterized by a development from the mastery of mechanistic rules to the refinement of their application based on analogical patterns extracted from existing verbs, with Czech promoting the production at the earliest proficiency stages.
... This paper proposes a deep learning framework to tackle these challenges. Specifically, we use a recurrent neural network (RNN) [5] to map the historical observations to a fixedsize representation, called state information, while accounting arXiv:2112.04075v1 [cs.IT] 8 Dec 2021 ...
... In this problem, the mapping F : R T D × R T M → R V outputs the designed control action v, possibly subject to some constraints. We show that with minor modifications to the proposed deep learning framework for solving (4), we can also tackle the problem (5). Note that since in solving (5), we may not have access to labeled data for the desired output v, the training of the deep neural network for solving (5) falls into the unsupervised learning paradigm. ...
... , R − 1 and use the activation function in the last layer to ensure any potential constraint that the final output must satisfy. Finally, we train the neural network architecture in Fig. 2 by employing stochastic gradient descent algorithms in order to minimize the empirical average MSE (over the training set) for problem (4) or to maximize average utility function J (·, ·) for problem (5). ...
Preprint
This paper proposes a deep learning approach to a class of active sensing problems in wireless communications in which an agent sequentially interacts with an environment over a predetermined number of time frames to gather information in order to perform a sensing or actuation task for maximizing some utility function. In such an active learning setting, the agent needs to design an adaptive sensing strategy sequentially based on the observations made so far. To tackle such a challenging problem in which the dimension of historical observations increases over time, we propose to use a long short-term memory (LSTM) network to exploit the temporal correlations in the sequence of observations and to map each observation to a fixed-size state information vector. We then use a deep neural network (DNN) to map the LSTM state at each time frame to the design of the next measurement step. Finally, we employ another DNN to map the final LSTM state to the desired solution. We investigate the performance of the proposed framework for adaptive channel sensing problems in wireless communications. In particular, we consider the adaptive beamforming problem for mmWave beam alignment and the adaptive reconfigurable intelligent surface sensing problem for reflection alignment. Numerical results demonstrate that the proposed deep active sensing strategy outperforms the existing adaptive or nonadaptive sensing schemes.
... However, these body models may not explicitly exist anywhere but may be implicitly realized in distributed patterns of synaptic connections, for example. This is a typical property expected from neural representations in the tradition of parallel distributed processing, also known as connectionism (Rumelhart and McClelland, 1986); see also Ramsey (2007). Hence, the inputs from the body models are marked with dashed arrows in Fig. 2. The flow and mixing of tactile and proprioceptive information is inspired by what is known from neurophysiology (see, e.g., Dijkerman and de Haan, 2007, for a review). ...
... First, we added the temporal dimension and distinguished between short-term, online, and longterm, offline, body representations (body state representation vs. body model). The former can be viewed as neural activations in a certain brain area; the latter may be implicitly encoded in synaptic weights-in line with the nature of representations in connectionist models in general (Rumelhart and McClelland, 1986). Second, we separated the model of body size from the model of body shape. ...
Preprint
Neurocognitive models of higher-level somatosensory processing have emphasised the role of stored body representations in interpreting real-time sensory signals coming from the body (Longo, Azanon and Haggard, 2010; Tame, Azanon and Longo, 2019). The need for such stored representations arises from the fact that immediate sensory signals coming from the body do not specify metric details about body size and shape. Several aspects of somatoperception, therefore, require that immediate sensory signals be combined with stored body representations. This basic problem is equally true for humanoid robots and, intriguingly, neurocognitive models developed to explain human perception are strikingly similar to those developed independently for localizing touch on humanoid robots, such as the iCub, equipped with artificial electronic skin on the majority of its body surface (Roncone et al., 2014; Hoffmann, 2021). In this chapter, we will review the key features of these models, discuss their similarities and differences to each other, and to other models in the literature. Using robots as embodied computational models is an example of synthetic methodology or 'understanding by building' (e.g., Hoffmann and Pfeifer, 2018), computational embodied neuroscience (Caligiore et al., 2010) or 'synthetic psychology of the self' (Prescott and Camilleri, 2019). Such models have the advantage that they need to be worked out into every detail, making any theory explicit and complete. There is also an additional way of (pre)validating such a theory other than comparing to the biological or psychological phenomenon studied by simply verifying that a particular implementation really performs the task: can the robot localize where it is being touched (see https://youtu.be/pfse424t5mQ)?
... It is only suitable for binary linear problems, limiting the application in complex issues. The secondgeneration ANN is the multi-layer perceptron, in which the back propagation neural network (BPNN) is one of the representatives [6]. The BPNN solves the weight adjustment problem of multi-layer neural networks, and the principles inspired the deep learning models such as convolutional neural networks (CNN). ...
... Rights reserved. 6. Generate spike trains: The generated sliding window sequence contains the normalized amplitudes in [0,1]. ...
Article
Full-text available
The spiking neural network (SNN) is the third generation of artificial neural networks. The transmission and expression of information in SNN are performed by spike trains, making the SNN have the advantages of high calculation speed and low power consumption. Recently, researchers have employed the SNN to recognize surface electromyography (sEMG) signals, but problems are still left. The sEMG encoders may cause information loss, and the network decoders may cause poor training performance. The strength of the neuron stimulated can be expressed by the frequency of the input or output spikes (namely firing rate). Inspired by the firing rate principle, we proposed the smoothed frequency-domain decomposition encoder, which converts the sEMG to spike trains. Furthermore, we also proposed the network efferent energy decoder, which converts the network output to recognizing results. The employed SNN is a three-layer fully-connected network trained by the grey wolf optimizer. The proposed methods are verified by a hand gestures recognition task. A total of 11 subjects participated in the experiment, and sEMG signals were acquired from five commonly used hand gestures by three sEMG sensors. The results indicate that the loss function can be reduced to below 0.4, and the average gesture recognizing accuracy is 91.21%. These results show the potential of using the proposed methods for the actual prosthesis. In the future, we will optimize the SNN training method to improve the training speed and stability.
... FHs. Connectionist methods of data representation can be categorized into two types: Localist and Distributed. In localist representation, each unit is associated with a single feature or concept and each concept is represented by one and only one unit [216]. Localist representation is simple to use and easy to code but not feasible for a component, structure-based data such as FHs. ...
... Localist representation is simple to use and easy to code but not feasible for a component, structure-based data such as FHs. In distributed representations, a single concept is represented by a combination of multiple units and each unit can be a part of multiple concepts [216]. and hair which are C1 -00000001 and C2 -00000010 as presented in Figure 3-1 (a). ...
Thesis
Full-text available
This thesis explores the idea that features extracted from deep neural networks (DNNs) through layered weight analysis are knowledge components and are transferable. Among the components extracted from the various layers, middle layer components are shown to constitute knowledge that is mainly responsible for the accuracy of deep architectures including deep autoencoders (DAEs), deep belief networks (DBNs) and DNNs. The proposed component-based transfer of knowledge is shown to be efficient when applied to a variety of benchmark datasets including handwritten character recognition, image recognition, speech analysis, gene expression, as well as hierarchical feature datasets. The importance of hidden layer and its position in the topology of Artificial Neural Networks (ANNs) is under-researched in comparison to the deployment of new architectures, components and learning algorithms. This thesis addresses this imbalance by providing an insight into what actually is learned by a neural network. This is because recent advances in layer-wise training enable us to explore systematically and rigorously the features that expose hidden layer by hidden layer in deep architectures. The key contribution of this research is providing a transferable component model by extracting knowledge components from hidden layers. This thesis also provides an approach to determine the contribution of individual layers, thus providing an insight into the topological constraints that require addressing while designing a transfer learning model. Such transfer learning can mitigate the problem of needing to train each neural network ‘from scratch.’ This is important since deep learning currently can be slow and require large amounts of processing power. “Warm started” deep learning may open new avenues of research, especially in areas where ‘portable’ deep architectures can be deployed for decision making.
... 6 Second, the general character of neural processing before conscious awareness is quite unlike the serial character of conscious awareness (Cisek & Kalaska, 2010). The principal feature of preconscious neural processing is massive parallel processing (McClelland & Rumelhart, 1988;Rumelhart & McClelland, 1986). With respect to emotion, multiple neural processes are executed in parallel and each well before consciousness. ...
... As to the first of the two neuroscience considerations, there is the Lodge and Taber's 'hot cognition' account. But, as to the second, there has been no recognition of the fundamental character of preconscious processing as massively multi-channel concurrent appraisals, especially as to emotion (McClelland & Rumelhart, 1988;Rumelhart & McClelland, 1986). ...
Conference Paper
Full-text available
Emotion has become an increasing influential area of research in psychology, political psychology, political science and other social sciences. Normally research is driven by theory. As such it is worth considering how well the current emotion research programs meet the requirements of a full blown theory. Among these, in alphabetical order, are: appraisal theories; emotion regulation; and, valence based accounts. After a brief overview of what elements individually and collective constitute a theory of emotion, I evaluate each as to the plausible claim of being a theory of emotion. I find that the worthy ambition to develop a full theory of emotion awaits fulfillment.
...  RESTRICTED BOLTZMANN MACHINE (RBM) RBM is a kind of ANN which consists of decision making units and uniformly connected neurons. It is a non-linear graphical developed model which represents probabilistic distribution made of observational, hidden or visible vectors [82][83]. RBM can model the binary numbers into two layers. ...
Article
Full-text available
The main and pivot part of electric companies is the load forecasting. Decision-makers and think tank of power sectors should forecast the future need of electricity with large accuracy and small error to give uninterrupted and free of load shedding power to consumers. The demand of electricity can be forecasted amicably by many Machine Learning (ML), Deep Learning (DL) and Artificial Intelligence (AI) techniques among which hybrid methods are most popular. The present technologies of load forecasting and present work regarding combination of various ML, DL and AI algorithms are reviewed in this paper. The comprehensive review of single and hybrid forecasting models with functions; advantages and disadvantages are discussed in this paper. The comparison between the performance of the models in terms of Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE) values are compared and discussed with literature of different models to support the researchers to select the best model for load prediction. This comparison validates the fact that the hybrid forecasting models will provide a more optimal solution. INDEX TERMS load forecasting, machine learning, load shedding, root mean squared error, mean absolute percentage error.
... [74,81,103]. The ongoing success of using momentum methods for training [91,100] and constructing [69,78,111] neural networks warrants an extensive investigation of these strategies in both worlds. ...
Article
Full-text available
We investigate numerous structural connections between numerical algorithms for partial differential equations (PDEs) and neural architectures. Our goal is to transfer the rich set of mathematical foundations from the world of PDEs to neural networks. Besides structural insights, we provide concrete examples and experimental evaluations of the resulting architectures. Using the example of generalised nonlinear diffusion in 1D, we consider explicit schemes, acceleration strategies thereof, implicit schemes, and multigrid approaches. We connect these concepts to residual networks, recurrent neural networks, and U-net architectures. Our findings inspire a symmetric residual network design with provable stability guarantees and justify the effectiveness of skip connections in neural networks from a numerical perspective. Moreover, we present U-net architectures that implement multigrid techniques for learning efficient solutions of partial differential equation models, and motivate uncommon design choices such as trainable nonmonotone activation functions. Experimental evaluations show that the proposed architectures save half of the trainable parameters and can thus outperform standard ones with the same model complexity. Our considerations serve as a basis for explaining the success of popular neural architectures and provide a blueprint for developing new mathematically well-founded neural building blocks.
... A neural network model was trained with either an orthography-phonology or phonology-orthography mapping task, corresponding to reading aloud visually presented words, and spelling spoken words, respectively. Our focus is on the PDP framework developed by Rumelhart et al. (1986) that provides natural accounts of the exploitation of multiple, simultaneous, and often mutual constraints. To examine the ease with which the model can generate the target output for a word, we measured the closeness of the model's output to the target by calculating the mean squared error (MSE) that serves as a reflection of how difficult it was for the model to learn the GPC/PGC mappings of each word. ...
Preprint
Full-text available
Research on orthographic consistency in English words has selectively identified different sub-syllabic units in isolation (grapheme, onset, vowel, coda, rime), yet there is no comprehensive assessment of how these measures affect word identification when taken together. To study which aspects of consistency are more psychologically relevant, we investigated their independent and composite effects on human reading behaviour using large-scale databases. Study 1 found effects of both forward (orthography-to-phonology) and backward consistency (phonol-ogy to orthography) on adults' naming responses. Study 2 found backward but no forward consistency effects on both visual and auditory lexical decision tasks, with the composite consistency across onset, vowel, and coda (OVC) being the best predictor. In Study 3, we explicitly modeled the reading process with forward and backward flow in bidirectionally connected neural networks. The model captured latent dimensions of quasi-regular mapping that explain additional variance in human reading and spelling behaviour, compared to the established measures. Together, the results suggest interactive activation between phonological and or-thographic word representations. They also validate the role of computational analyses of language to better understand how print maps to sound, and what affects reading complexity.
... Rumelhart and McClelland [58] define the ANN as a network that comprises a series of nodes-or process elements (PE)-with a certain information storage capacity. These PE are composed by an input vector (x 1 , x 2 , . . . ...
Article
Full-text available
Environmentally friendly behaviour and the equitable and sustainable use of natural resources can contribute to solving various environmental, economic, and social problems in different countries. The analysis of the perception of young students is important because schools are suitable for educating future generations and shaping their attitudes to also include a greater concern for the environment. This research aims to determine the degree of influence that a series of Likert-type questions of knowledge, attitudes, and behaviours about sustainable development has on a series of items of the student profile (gender, age, course, and household members) in a developing country. For this, an artificial neural network is used that allows us not only to quantify the degree of influence but also to obtain an estimation of the student’s profile according to the responses obtained on sustainable development. The network developed allows us to obtain, through a determined collection of answers to questions about sustainable development, the estimation of a specific profile of a student from a developing country. This can be useful to educational communities interested in optimising economic resources through sustainable development, allowing them to know which issues they should focus more (or less) on according to the profile of the student they are targeting.
... This derivative is generally figured in two passes; forward pass, an input vector from the preparation set is connected to the input units of the network and is spread through the network layer by layer creating the last output, and backward pass, the output of the network is compared with the optimal output and the subsequent error is then propagated in reverse through the network modifying the network. Mcclelland et al., (1986) presented a momentum term δ to equation (15) to accelerate the learning procedure, while keeping away from the instability of the algorithm. ...
Chapter
This chapter analyses efficiency of support vector regression (SVR), artificial neural networks (ANNs), and structural vector autoregressive (SVAR) models in terms of in-sample forecasting of portfolio inflows (PIs). Time series daily data sourced from Rand Merchant Bank (RMB) covering the period of 1st March 2004 to 1st February 2016 were used. Mean squared error, root mean squared error, mean absolute error, mean absolute squared error, and root mean scaled log error were used to evaluate model performance. The results showed that SVR has the best modelling performance when compared to others. In determining factors that affect allocation of PIs into South Africa based on SVAR, 69% of the variation was explained by pull factors while 9% was explained by push factor. Hence, SVR model is more accurate than ANNs. This chapter therefore recommends that banking sector particularly RMB should use machine learning technique in modelling PIs for a better financial solution.
... There is, nevertheless, a vast literature on sigma-pi networks in general, e.g. [12,13], which is not surprising since such networks define a large class of possible systems. ...
Article
Full-text available
Feature-product networks (FP-nets) are inspired by end-stopped cortical cells with FP-units that multiply the outputs of two filters. We enhance state-of-the-art deep networks, such as the ResNet and MobileNet, with FP-units and show that the resulting FP-nets perform better on the Cifar-10 and ImageNet benchmarks. Moreover, we analyze the hyperselectivity of the FP-net model neurons and show that this property makes FP-nets less sensitive to adversarial attacks and JPEG artifacts. We then show that the learned model neurons are end-stopped to different degrees and that they provide sparse representations with an entropy that decreases with hyperselectivity.
... O'Loughlin and Thagard (2000) described another computational model that explains the lack of central coherence in autism. Their connectionist model is based on a network of constraint satisfaction [47], and it was believed that the lack of central cohesion is due to a very high level of inhibition compared to the stimulation level. This model is attractive in that it uses both normal and autistic cognition. ...
Article
Autism is an advanced neurological disease that affect communication and social behaviors, including attention -one of the fundamental skills to learn about the world around us. Autistic people have difficulty moving their attention from one point to another fluently. Due to the high prevalence of autism and its increasing progression, and the need to address common disorders in patients, this study aimed to implement and simulate a computational model for attention deficit disorder in autistic patients using MATLAB. This computational model has three components: context-sensitive reinforcement learning, contextual processing, and automation that can teach a shift-shift task automatically. At first, the model functions like normal people, but its performance gets closer to autistic people after changing a single parameter. This study demonstrates that even a simple computational model can be used for normal and abnormal developmental cases using a neural network reinforcement learning approach and provide valuable insights into autism.
... • Artificial neural networks were developed, in principle, as mathematical models of the information processing capabilities of the biological brain [ [2], [3]]. In this study were take into account two types of neural networks: multilayer perceptron (MLP) and Recurrent Neural Networks (RNN). ...
... However, these models are limited in what they can do. The development of "parallel distributed processing" models has attempted to overcome the limitations of linear systems [42]. Such models consist of a very large number of nonlinear functions, often referred to as neural networks or forests of decision trees. ...
Article
Full-text available
The use of artificial intelligence (AI) systems in biomedical and clinical settings can disrupt the traditional doctor-patient relationship, which is based on trust and transparency in medical advice and therapeutic decisions. When the diagnosis or selection of a therapy is no longer made solely by the physician, but to a significant extent by a machine using algorithms, decisions become untransparent. Skill learning is the most common application of machine learning algorithms in clinical decision making. These are a class of very general algorithms (artificial neural networks, classifiers, etc.) which are tuned based on examples to optimize the classification of new, unseen cases. It is pointless to ask for an explanation for a decision. A detailed understanding of the mathematical details of an AI algorithm may be possible for experts in statistics or computer science. But when it comes to the fate of human beings, this "developer's explanation" is not sufficient. The concept of explainable AI (XAI) as a solution to this problem is attracting increasing scientific and regulatory interest. This review focuses on the requirement that XAIs must be able to explain in detail the decisions made by the AI to the experts in the field.
... However, Q-learning has a drawback in that the number of states increases explosively as the number of input variables increases, and its memory usage also increases sharply since it should store all the state-action relations in a table. Accordingly, many related works [11][12][13][14][15][16] adopt Deep Q-Network (DQN) [19], which combines Q-learning with an artificial neural network [20]. ...
Article
Full-text available
A network composed of unmanned aerial vehicles (UAVs), serving as base stations (UAV-BS network), is emerging as a promising component in next-generation communication systems. In the UAV-BS network, the optimal positioning of a UAV-BS is an essential requirement to establish line-of-sight (LoS) links for ground users. A novel deep Q-network (DQN)-based learning model enabling the optimal deployment of a UAV-BS is proposed. Moreover, without re-learning of the model and the acquisition of the path information of ground users, the proposed model presents the optimal UAV-BS trajectory while ground users move. Specifically, the proposed model optimizes the trajectory of a UAV-BS by maximizing the mean opinion score (MOS) for ground users who move to various paths. Furthermore, the proposed model is highly practical because, instead of the locations of individual mobile users, an average channel power gain is used as an input parameter. The accuracy of the proposed model is validated by comparing the results of the model with those of a mathematical optimization solver.
... Perhaps surprisingly, the derivatives toward the weights in the hidden layers (e.g., input to hidden weights) have an intuitive and relatively simple formulation in terms of the prediction errors of the weights in the subsequent layer (as explained in detail next). The realization that these resulting derivatives are so simple, and of similar form as the delta-rule derivatives, led to a widespread application of hidden-layer models in cognitive science, starting from the 1980s (Rumelhart et al., 1986b;Werbos, 1982) The learning algorithm works as follows. Consider again a mean square error (MSE) function that we aim to minimize for the generic model depicted in figure 5.2: ...
... Besides the unique function approximation capabilities of ANN, their burgeoning popularity is arguably due to two more factors. The first is the possibility to easily compute the cost function's gradient ∇ w J(w) using the chain rule for differentiation, which leads to the popular back-propagation algorithm (Rumelhart and McClelland, 1989). This allows to use an arsenal of optimization tools for the training. ...
Preprint
Full-text available
Big data and machine learning are driving comprehensive economic and social transformations and are rapidly re-shaping the toolbox and the methodologies of applied scientists. Machine learning tools are designed to learn functions from data with little to no need of prior knowledge. As continuous developments in experimental and numerical methods improve our ability to collect high-quality data, machine learning tools become increasingly viable and promising also in disciplines rooted in physical principles. These notes explore how machine learning can be integrated and combined with more classic methods in fluid dynamics. After a brief review of the machine learning landscape, we show how many problems in fluid mechanics can be framed as machine learning problems and we explore challenges and opportunities. We consider several relevant applications: aeroacoustic noise prediction, turbulence modelling, reduced-order modelling and forecasting, meshless integration of (partial) differential equations, super-resolution and flow control. While this list is by no means exhaustive, the presentation will provide enough concrete examples to offer perspectives on how machine learning might impact the way we do research and learn from data.
... La sortie désigne la dernière couche du réseau comportant un neurone par catégorie. Un CNN applique un ensemble de perceptrons multicouches (PMC), présentant une famille de réseau neuronal artificiel, caractérisés par des combinaisons de séparateurs linéaires permettant de produire un séparateur global non-linéaire [144], conçus pour réduire le temps de calcul. En effetune architecture de CNN est formée par un empilement de couches de traitement (voir Figure II.13) ; la couche de convolution (CONV) permettant le traitement des données d'un champ récepteur par un filtre de convolution ; la couche de pooling (POOL) permettant la compression d'information en réduisant la taille de l'image intermédiaire (souvent par sous-échantillonnage) ; la couche de correction (ReLU), souvent appelée par abus « ReLU » en référence à la fonction d'activation (Unité de rectification linéaire) ; la couche entièrement connectée (« fully connected »), qui est une couche de type perceptron et finalement LOSS présentant la couche de calcul de la fonction d'erreur. ...
Thesis
Le diagnostic des lésions hépatiques est une tâche complexe surtout lorsque les nodules détectés sont de petites tailles. Dans ce cas, il devient très difficile de connaitre leurs natures (tumeur bénigne ou maligne, type de lésion, etc). Dans des cas similaires, il faut répéter des examens cliniques pendant plusieurs mois pour voir l’évolution des masses hépatiques. Afin de mieux répondre à ces problèmes, il faut trouver des solutions informatiques qui servent à l’optimisation du diagnostic des tumeurs du foie. Dans le contexte de la classification des lésions hépatiques, nous avons développé une première approche ontologique (OntHCC) pour l’aide au diagnostic, à la stadification et au choix de traitement des tumeurs CHC (Carcinome Hépatocellulaire). Cette approche est fondée sur l’analyse d’images IRM de foies infectés et sur des rapports radiologiques. Par la suite, nous avons proposé une deuxième approche ontologique (MROnt) pour la modélisation de l’information médicale contenue dans les rapports radiologiques, dans le cadre du diagnostic et de suivi de tumeurs du foie. La détection automatique des tumeurs du foie nécessite un processus de diagnostic primaire en utilisant obligatoirement les images médicales (par exemple IRM ou scanner). Pour ce faire, nous avons intégré l’apprentissage profond dans la classification d’images IRM avec prise de contraste. Dans la suite de la thèse et afin d’accroitre la performance du processus de classification des images, nous avons intégré les connaissances sémantiques. L’objectif est de profiter de la base de connaissances offerte par les ontologies pour décrire les images médicales et fournir des informations sur les tumeurs détectées (par exemple, le type, la taille et le stade). En outre, notre approche consiste à développer un CNN multi-label afin de supporter les ontologies développées (OntHCC et MROnt). Nous montrons l’efficacité des approches et prototypes proposés dans ces travaux de thèse à travers des évaluations numériques comparatives et des études de cas.
... Economic motivation is also one of the important reasons for collusion in the bidding process [17]. e reason for this is that the costs and risks of collusion are low compared to the benefits. ...
Article
Full-text available
Due to the national economic development form and social development demand, in recent years, the government has been vigorously promoting the control of government-enterprise collusion in the bidding process of government projects in order to promote the standardization of the market. How to predict the vertical collusion behavior under different internal and external environments has become an important research content. Although the prediction of individual behavior is difficult to achieve, the prediction of group behavior has certain possibilities. In this paper, we propose a method for predicting and evaluating the vertical collusion behavior of government investment project bidding based on BP neural network analysis optimized by an annealing algorithm. First, drawing on the traditional evaluation model, the evaluation index system of government-enterprise collusion behavior is constructed from five dimensions: internal environment, external environment, policy development, enforcement effort, and feedback channel. Secondly, a machine learning method based on BP neural network optimized by an annealing algorithm is introduced to evaluate the influence of the change of initial conditions on the bidding collusion behavior. This study has certain significance for government managers to discover the problems and causes in policy formulation, which in turn can support the government in further improving the policy utility.
... COGNITIVE CONTROL OF IMPLICIT LEARNING 11 also followed their lead, reducing the target output activations to .1 and .9 instead of using 0 and 1, to make these output predictions more reachable (Rumelhart & McClelland, 1986). On each trial, the prediction of the model was measured as the Luce choice ratio (LCR), which amounts to the activation of the target unit divided by the sum of the activation over the four output units. ...
Article
Implicit sequence learning represents an established paradigm to investigate incidental skill acquisition in a laboratory environment. During a covert task, participants respond to the location of a target appearing over a series of locations according to a complex sequence, which gets violated in a reduced set of control trials. Even though participants are not fully aware of the sequence, they respond faster and more accurately to trials following it, thus expressing sequence knowledge. Recent evidence has challenged the view that such knowledge is applied rigidly and affects performance independently from control influences. Jiménez et al. (2009) highlighted that its expression gets reduced immediately after trials not conforming with the learned sequence-an effect that resembles the congruency sequence effect (CSE) commonly observed in interference tasks. However, such effects can also be alternatively explained in associative terms. In this experimental series we took advantage of the well-known attentional properties of oddball sounds and introduced them as an orthogonal variable with respect to the learning process. We found that oddball sounds also hindered the automatic expression of sequence learning, highlighting an oddball-dependent sequence effect similar to the CSE, but most clearly triggered by cognitive control. Moreover, as illustrated through a simulation with a simple recurrent network (SRN), we showed that the CSE reported in this article under noisier conditions is harder to expect from associative processes. (PsycInfo Database Record (c) 2022 APA, all rights reserved).
... ANNs are very sophisticated techniques able to model very complex functions through artificial units, the neurons, arranged in various architectures. Among the various ANN types, we considered the multilayer perceptron (MLP) [38] for this study, due to its successful application in previous works [16,17]. The MLP architecture generally arranges the neurons in more layers: the input layer includes the neurons where inputs are applied; thus, the output of this layer acts as the input for the next one, and so on, until the final output layer. ...
Article
Full-text available
Air pollution significantly affects public health in many countries. In particular, indoor air quality can be equally, if not more, concerning than outdoor emissions of pollutant gases. However, monitoring the air quality in homes and apartments using chemical analyzers may be not affordable for households due to their high costs and logistical issues. Therefore, a new alternative is represented by low-cost air quality monitors (AQMs) based on low-cost gas sensors (LCSs), but scientific literature reports some limitations and issues concerning the quality of the measurements performed by these devices. It is proven that AQM performance is significantly affected by the calibration model used for calibrating LCSs in outdoor environments, but similar investigations in homes or apartments are quite rare. In this work, the assessment of an AQM based on electrochemical sensors for CO, NO2, and O3 has been performed through an experiment carried out in an apartment occupied by a family of four during their everyday life. The state-of-the-art of the LCS calibration is featured by the use of multivariate linear regression (MLR), random forest regression (RF), support vector machines (SVM), and artificial neural networks (ANN). In this study, we have conducted a comparison of these calibration models by using different sets of predictors through reference measurements to investigate possible differences in AQM performance. We have found a good agreement between measurements performed by AQM and data reported by the reference in the case of CO and NO2 calibrated using MLR (R2 = 0.918 for CO, and R2 = 0.890 for NO2), RF (R2 = 0.912 for CO, and R2 = 0.697 for NO2), and ANN (R2 = 0.924 for CO, and R2 = 0.809 for NO2).
... Encoding a comprehensive social trait space provides the neural basis for rapid spontaneous impressions of faces on multiple trait dimensions. Our present results are in line with the notion that face representations are encoded over a broad and distributed population of neurons [46], which has been conclusively demonstrated in the non-human primate IT cortex [3]. Our results further shed light on how face processing evolves along the visual processing stream where the brain transforms from encoding lowlevel facial features in the higher visual cortex to complex social traits in the amygdala and hippocampus. ...
Article
Full-text available
People instantaneously evaluate faces with significant agreement on evaluations of social traits. However, the neural basis for such rapid spontaneous face evaluation remains largely unknown. Here, we recorded from 490 neurons in the human amygdala and hippocampus and found that the neuronal activity was associated with the geometry of a social trait space. We further investigated the temporal evolution and modulation on the social trait representation, and we employed encoding and decoding models to reveal the critical social traits for the trait space. We also recorded from another 938 neurons and replicated our findings using different social traits. Together, our results suggest that there exists a neuronal population code for a comprehensive social trait space in the human amygdala and hippocampus that underlies spontaneous first impressions. Changes in such neuronal social trait space may have implications for the abnormal processing of social information observed in some neurological and psychiatric disorders.
... Networks. BP (back propagation) neural network is a highly complex nonlinear dynamic analysis system proposed by Rumelhart and McClelland et al. [33] in 1986. It is a network structure connected by various independent units. ...
Article
Full-text available
E-payment has penetrated every aspect of people’s daily life, and the development and application of E-payment technology have made life more convenient. Besides, the process of E-payment has endless hidden dangers, posing a great threat to payment security. In this context, risk assessment and prediction of E-payment are particularly important. Therefore, this paper mainly studies the E-payment analysis and prediction method based on big data technology. Specifically, this paper uses the BP neural network to extract E-payment data feature and compare and analyze the characteristic of different methods; finally, the payment risks are predicted based on the features, the simulation experiments show that the best result is obtained by the method proposed in this paper, and thus the effectiveness of the new method is verified.
... In the mid-1980s, parallel distributed processing described the algorithm based on BP [50]. It can solve the problem of learning the connection weights of hidden layer neurons in multilayer networks, and some issues that could not be solved by a single perceptron before, which shows that ANN has powerful computing ability [51]. ...
Article
Full-text available
Industrial solid waste (ISW) accounts for the most significant proportion of solid waste in China. Improper treatment of ISW will cause significant environmental pollution. As the basis of decision-making and the management of solid waste resource utilization, the accurate prediction of industrial solid waste generation (ISWG) is crucial. Therefore, combined with China’s national conditions, this paper selects 14 influential factors in four aspects: society, economy, environment and technology, and then proposes a new prediction model called the principal component analysis nonlinear autoregressive back propagation (PCA-NARBP) neural network model. Compared with the back propagation (BP) neural network model and nonlinear autoregressive back propagation (NARBP) neural network model, the mean absolute percentage error (MAPE) of this model reaches 1.25%, which shows that it is more accurate, includes fewer errors and is more generalizable. An example is given to verify the effectiveness, feasibility and stability of the model. The forecast results show that the output of ISW in China will still show an upward trend in the next decade, and limit the total amount to about 4.6 billion tons. This can not only provide data support for decision-makers, but also put forward targeted suggestions on the current management situation in China.
... The former was motivated primarily by "Explorations in the Microstructure of Cognition" (Rumelhart & McClelland, 1986) … On the contrary, the development of deep neural models is mainly driven by applications. (Perconti & Plebe, 2020, p. 5). ...
Article
Full-text available
In this article, I describe a novel position on the semantics of artificial intelligence. I present a problem for the current artificial neural networks used in machine learning, specifically with relation to natural language tasks. I then propose that from a metasemantic level, meaning in machines can best be interpreted as radically contextualist. Finally, I consider what this might mean for human-level semantic competence from a comparative perspective.
Article
Full-text available
As one of the core tasks in the field of natural language processing, syntactic analysis has always been a hot topic for researchers, including tasks such as Questions and Answer (Q&A), Search String Comprehension, Semantic Analysis, and Knowledge Base Construction. This paper aims to study the application of deep learning and neural network in natural language syntax analysis, which has significant research and application value. This paper first studies a transfer-based dependent syntax analyzer using a feed-forward neural network as a classifier. By analyzing the model, we have made meticulous parameters of the model to improve its performance. This paper proposes a dependent syntactic analysis model based on a long-term memory neural network. This model is based on the feed-forward neural network model described above and will be used as a feature extractor. After the feature extractor is pretrained, we use a long short-term memory neural network as a classifier of the transfer action, and the characteristics extracted by the syntactic analyzer as its input to train a recursive neural network classifier optimized by sentences. The classifier can not only classify the current pattern feature but also multirich information such as analysis of state history. Therefore, the model is modeled in the analysis process of the entire sentence in syntactic analysis, replacing the method of modeling independent analysis. The experimental results show that the model has achieved greater performance improvement than baseline methods.
Preprint
Full-text available
It is challenging to specify the role of the default mode network (DMN) in human behaviour. Contemporary theories, based on functional magnetic resonance imaging (MRI), suggest that the DMN is insulated from the external world, which allows it to support perceptually-decoupled states and to integrate external and internal information in the construction of abstract meanings. To date, the neuronal architecture of the DMN has received relatively little attention. Understanding the cytoarchitectural composition and connectional layout of the DMN will provide novel insights into its role in brain function. We mapped cytoarchitectural variation within the DMN using a cortical type atlas and a histological model of the entire human brain. Next, we used MRI acquired in healthy young adults to explicate structural wiring and effective connectivity. We discovered profound diversity of DMN cytoarchitecture. Connectivity is organised along the most dominant cytoarchitectural axis. One side of the axis is the prominent receiver, whereas the other side remains more insulated, especially from sensory areas. The structural heterogeneity of the DMN engenders a network-level balance in communication with external and internal sources, which is distinctive, relative to other functional communities. The neuronal architecture of the DMN suggests it is a protuberance from the core cortical processing hierarchy and holds a unique role in information integration.
Article
Full-text available
Cognitive network science is an emerging approach that uses the mathematical tools of network science to map the relationships among representations stored in memory to examine how that structure might influence processing. In the present study, we used computer simulations to compare the ability of a well-known model of spoken word recognition, TRACE, to the ability of a cognitive network model with a spreading activation-like process to account for the findings from several previously published behavioral studies of language processing. In all four simulations, the TRACE model failed to retrieve a sufficient number of words to assess if it could replicate the behavioral findings. The cognitive network model successfully replicated the behavioral findings in Simulations 1 and 2. However, in Simulation 3a, the cognitive network did not replicate the behavioral findings, perhaps because an additional mechanism was not implemented in the model. However, in Simulation 3b, when the decay parameter in spreadr was manipulated to model this mechanism the cognitive network model successfully replicated the behavioral findings. The results suggest that models of cognition need to take into account the multi-scale structure that exists among representations in memory, and how that structure can influence processing.
Thesis
Full-text available
In the mid 1980’s several bio-inspired approaches emerged to the study of artificial intelligence. Starting from this context and from von Neumann cellular automata, the field of artificial life was developed with the objective to construct artificial systems capable to present similar behaviors to those found in biological phenomena. This thesis recovers the history of artificial life and its relationship with artificial intelligence, presents the difficulties of its development considering cartesian dualism, and demonstrates the possibility of a more adequate way of research based on the hypothesis of continuity between mind and matter, typical of the general semiotics of Charles Sanders Peirce. Through peircean semiotics and using the fundamentals of biosemiotics, the semiotic transposition technique is developed, a set of diagrammatic operations to support the study of artificial life. This technique studies the semiotic processes underlying biological phenomena. Then, through isomorphism, derived from the category theory, a finite automata can be created to computationally express certain aspects of the original biological processes. Throughout the research, the learning and memory behavior of a sea slug species, Aplysia californica, was used as an auxiliary element for the formalization of semiotic transposition. Two other biological phenomena — the genetic translation and the vacancy chain dynamics related to the Pagurus longicarpus, a species of crab — were considered as case studies to demonstrate the general character of the semiotic transposition. It is concluded that the use of semiotic theory as the basis for the study of artificial life constitutes an effective instrument to the creation of bio inspired computational devices.
Chapter
Computational modeling is a powerful tool for studying reading and other complex behaviors. This chapter focuses on the role of computational modeling. It begins by showing that simulations presented as supporting the dual‐route cascade (DRC) model differed from the corresponding behavioral studies. The DRC models were succeeded by a series of hybrid, “connectionist dual process” models. These models replaced the grapheme‐phoneme correspondence rules with connectionist networks, but retained a separate lexical route. The dual‐route theory remains influential in areas where computational modeling results are not well known. These include reading acquisition and instruction, where research and pedagogy still focus on learning pronunciation rules and adding sight words to the lexicon, and in some areas of cognitive neuroscience. The chapter concludes by considering the relevance of computational modeling for understanding how children learn to read.
Article
This study aims to examine the feasibility of predicting surface wind pressure induced by conical vortex using a backpropagation neural network (BPNN) combined with proper orthogonal decomposition (POD), in which a 1:150 scaled model with a large-span retractable roof was tested in wind tunnel under both laminar and turbulent flow conditions. The distributions of mean and fluctuating wind pressure coefficients were first described, and the effects of inflow turbulence, wind direction, roof opening were examined separately. For the prediction of wind pressure, the POD-BPNN model was trained using measurement data from adjacent points. The prediction results are overall satisfactory. The root-mean-square-error (RMSE) between test and predicted data lies mostly within 10%. In particular, the prediction of mean wind pressure is found to be better than that of fluctuating wind pressure. The outcomes in this study highlight that the proposed POD-BPNN model can be well used as a useful tool to predict surface wind pressure.
Presentation
Full-text available
In this presentation, I discuss the role/influence of the ever-changing human brain-organisation in the registration and interpretation of sensorial stimuli (signals). These both concepts are grounded on the organisation/in-formation framework [1] and defined in a manner similar to Pierce’s semiotics. The discussion relies on recent achievements of cognitive science and on perceiving mathematics as “The science of human cognitive processes and archetypes” [2]. The arguments connect philosophical and mathematical ideas to biological-organisation configurations and lay down a backcloth that allows for advancing a definition of creativity that stems from comparing an intrinsic complexity measure of the intervening organisations, as well as, forcing biological stasis out of balance into another stasis-basin or domain. As defined and due to the mediation of the human nervous system in the perception and thinking processes, creativity refers to a context much wider than mathematics, encompassing literature, music and visual arts. The present work continues discussions held at L’Imagination 2018 and Creativity 2019 congresses. In the proposed abstract setting, creativity implies the instantiation of novel forms of organisation and interpretation enacted by a sudden change, an out-of-the-blue change so to speak. This becomes clear by considering the dynamical behaviour intrinsically associated to “concrete” organisations. The connections with biological elements show that there is a strong natural process that counteracts, but does not prevent, creativity. Namely, homeostasis. To compensate that and maintain human beings creative and being able to readily move away, “thinking out of the box”, we should mimic nature once more and keep active the regulatory processes that swiftly move biological organisations from one homeostatic configuration to another. In the case of human brains and intellectual activity, the majority of these processes relate to synaptic plasticity and local re-organisations in our three brains, or at least in the rational and emotional ones. One way to achieve this flexibility is through the “power of ideas” and “thought-experiments”, by employing theoretical and philosophical thinking to explore wild, never thought, ideas, concepts and possibilities. This is an exercise that, unfortunately, the scientific community at large and its satellite communities (sci-fi, for instance) have progressively abandoned during the last three decades, with the possible exception of physicists and philosophers. It is time to revert this trend in favor of better tackling the complex problems facing mankind. This work aims to contribute to a better understanding of the usually underrated role of biological processes and organisations in human creativity. Such understanding, hopefully, can help us to achieve greater clarity.
Chapter
Full-text available
Die Arbeit befasst sich mit dem wenig diskutierten Thema der Koordination und emergenten Struktur des Wissens in einem evolvierenden Wirtschaftssystem. Ausgehend vom historischen Narrativ der modernen Marktwirtschaft und ihrer Positionierung in der vergleichenden Wirtschaftssystemanalyse werden theoretische Bausteine auf der Basis einer Diskussion der Arbeiten in der Tradition von Smith, Coase, Hayek, Schumpeter und anderen identifiziert und im Rahmen einer Mikro-Meso-Makro Architektur als Komponenten eines einheitlichen Erklärungsansatzes diskutiert.
Article
The development of analytical and computational techniques and growing scientific funds collectively contribute to the rapid accumulation of geoscience data. The massive amount of existing data, the increasing complexity, and the rapid acquisition rates require novel approaches to efficiently discover scientific stories embedded in the data related to geochemistry and cosmochemistry. Machine learning methods can discover and describe the hidden patterns in intricate geochemical and cosmochemical big data. In recent years, considerable efforts have been devoted to the applications of machine learning methods in geochemistry and cosmochemistry. Here, we review the main applications including rock and sediment identification, digital mapping, water and soil quality prediction, and deep space exploration. Research method improvements, such as spectroscopy interpretation, numerical modeling, and molecular machine learning, are also discussed. Based on the up-to-date machine learning/deep learning techniques, we foresee the vast opportunities of implementing artificial intelligence and developing databases in geochemistry and cosmochemistry studies, as well as communicating geochemists/cosmochemists and data scientists.
Article
Full-text available
This paper investigated the development of a hybrid model for wind speed forecast, ranging from 1 to 46 days, in the northeast of Brazil. The prediction system was linked to the widely used numerical weather prediction from the ECMWF global ensemble forecast, with neural networks (NNs) trained using local measurements. The focus of this study was on the post-processing of NNs, in terms of data structure, dimensionality, architecture, training strategy, and validation. Multilayer perceptron NNs were constructed using the following inputs: wind components, temperature, humidity, and atmospheric pressure information from ECMWF, as well as latitude, longitude, sin/cos of time, and forecast lead time. The main NN output consisted of the residue of wind speed, i.e., the difference between the arithmetic ensemble mean, derived from ECMWF, and the observations. By preserving the simplicity and small dimension of the NN model, it was possible to build an ensemble of NNs (20 members) that significantly improved the forecasts. The original ECMWF bias of −0.3 to −1.4 m/s has been corrected to values between −0.1 and 0.1 m/s, while also reducing the RMSE in 10 to 30%. The operational implementation is discussed, and a detailed evaluation shows the considerable generalization capability and robustness of the forecast system, with low computational cost.
Thesis
Full-text available
In the context of foreign and second language learning, multi-word units constitute a particularly interesting phenomenon since they are known to cause problems for learners. One group of multi-word units that causes great difficulty even for advanced learners of English is multi-word verbs. Their acquisition and active usage is a challenge since they are complex both in terms of their grammatical form and their lexical meaning. This study provides a detailed, descriptive investigation of four different categories of multi-word verbs – namely phrasal, phrasal-prepositional, prepositional verbs and verb-noun collocations – in the essays written by intermediate to advanced level of Turkish learners of English. The Turkish sub-corpus (TICLE) of the International Corpus of Learner English (ICLE) has been the basis for the investigation. In order to thoroughly capture difficulties the learners experience in the use multi-word verbs and gain a better understanding of their phraseological competence, both the qualitative and quantitative aspects of the learners’ performance are investigated. An important aim of the study is to determine whether, and if so, to what extent, the learner’s first language (L1) influences their use of multi-word verbs in English. In addition to the learner’s L1, possible effects of other factors (both learner-related and external variables) are also investigated in the context of two categories of multi-word verbs, namely phrasal and phrasal-prepositional verbs – the two verb categories reported to be avoided and/or underused by many learner groups.
Article
Full-text available
For several years, time-series prediction seems to have been a popular research topic. Sales plans, ECG forecasts, meteorological circumstances, and even COVID-19 spreading projections are among its uses. These implementations have inspired several scientists to develop an optimum forecasting method; however, the modeling method varies as the implementation domain evolves. Telemetry data prediction is an important component of networking and information center control software. As a generalization of such a fuzzy system, the concept of an intuitionistic fuzzified set was created, which has proven to become a highly valuable tool in dealing with indeterminacy (hesitation) as in-network. Indeterminacy is frequently overlooked in applying fuzzified time-series prediction for no obvious cause. We introduce the concept of intuitionistic fuzzified time series within a current study to deal with non-determinism with time-series prediction. Also, it seems to be an intuitionistic fuzzified time-series prediction framework. Using time-series information, the suggested intuitionistic fuzzified time-series predicting approach employs intuitionistic fuzzified logical relationships. The suggested method's effectiveness is tested using two-time sequence data sets. By contrasting the predicted result with some other intuitionistic timing series predicting techniques utilizing root-mean-square inaccuracy and averaged predicting errors, the usefulness of the suggested intuitionistic fuzzified time-series predicting approach is demonstrated.
Preprint
Full-text available
This article presents an artificial intelligence (AI) architecture intended to simulate the human working memory system as well as the manner in which it is updated iteratively. It features several interconnected neural networks designed to emulate the specialized modules of the cerebral cortex. These are structured hierarchically and integrated into a global workspace. They are capable of temporarily maintaining high-level patterns akin to the psychological items maintained in working memory. This maintenance is made possible by persistent neural activity in the form of two modalities: sustained neural firing (resulting in a focus of attention) and synaptic potentiation (resulting in a short-term store). This persistent activity is updated iteratively resulting in incremental changes to the content of the working memory system. As the content stored in working memory gradually evolves, successive states overlap and are continuous with one another. The present article will explore how this architecture can lead to gradual shift in the distribution of coactive representations, ultimately leading to mental continuity between processing states, and thus to human-like cognition.
Chapter
BioremediationBioremediation is the chief applicable methodology to control the pollution and contamination of soilSoil. The in-place treatment along with above ground treatment of contaminated soil has created intense scientific growth. The general pollutants of soil include petroleumPetroleumhydrocarbonsHydrocarbons, heavy metalsHeavy metals, pesticidesPesticides used in agriculturalAgricultural field which alter the characteristics of soil. Although, microbesMicrobes are also beneficial to recover the contaminations present in the soil, several artificial intelligence constructed models help in detection of phytotoxicity of soil.
Book
Durch das Internet hat sich der Zugang zu Nachrichten maßgeblich verändert. Informationen stehen nicht nur unbegrenzt zur Verfügung, sondern sie sind auch zu einem omnipräsenten Bestandteil in digitalen Informationsumgebungen geworden. Dadurch werden Internetnutzer*innen, auch ohne bewusst danach zu suchen, wiederholt mit tagesaktuellen Schlagzeilen konfrontiert, z.B. wenn sie ihren Browser öffnen, oder sich auf sozialen Netzwerkseiten bewegen. Diese kurzen Nachrichtenkontakte haben aufgrund der geringen Informationsmenge wenig Potential für Lerneffekte, können jedoch das Gefühl vermitteln, sich mit einem Thema auszukennen. Vor diesem Hintergrund stellt sich die Frage, inwiefern Nachrichten in digitalen Informationsumgebungen die Entstehung einer Wissensillusion begünstigen, wie sich dieser Prozess erklären lässt und mit welchen Folgen dies verbunden ist. Im theoretischen Teil der Arbeit werden dazu Erkenntnisse zum Gedächtnis, dem Metagedächtnis und der Rolle von Medien für Wissen und Wissenswahrnehmung aufgearbeitet. In Studie 1 wird mit einer experimentellen Studie untersucht, wie sich Nachrichten auf sozialen Netzwerkseiten im Vergleich zu vollständigen Nachrichtenartikeln auf objektives und subjektives Wissen auswirken. Außerdem werden Effekte einer Wissensillusion für Einstellungen und Verhalten untersucht. Studie 2 untersucht mit qualitativen Leitfadeninterviews, welche Rolle Medien für Wissen und Lernen aus Sicht der Nutzer*innen spielen. Diese Erkenntnisse liefern Erklärungen dafür, weshalb und aufgrund welcher Merkmale unterschiedliche Nachrichtenkontakte eine Wissensillusion begünstigen können.
Chapter
Full-text available
En este capitulo se recogen algunas de las tendencias teóricas y metodológicas en ciencias sociales que apuntan hacia la investigación y explicación de los hechos humanos en general (y los politicos y electorales en particular) en ciclos más largos y se analizan algunos de los hechos que caracterizaron la campaña electoral del año 2000 en España.
Article
The training phase of a Back-Propagation (BP) network is an unconstrained optimization problem. The goal of the training is to search an optimal set of connection weights in the manner that the error of the network out put can be minimized. In this paper we developed the Classical Fletcher-Revees (CFRB) method for non-linear conjugate gradient to the scaled conjugate gradient (SFRB say) to train the feed forward neural network. Our development is based on the sufficient descent property and pure conjugacy conditions. Comparative results for (SFRB), (CFRB) and standard Back-Propagation (BP) are presented for some test problems.
Thesis
Full-text available
Improving of machinability properties of the materials which are used in manufacturing industry has been research issue. High corrosion resistance, ductility and tension strength are the essential properties of stainless steels. Stainless steels contain alloy elements such as crom, nickel and molibden which are affect their machinability negatively. For this reason the machining of stainless steels are very difficult. In this study we investigate the machinability of AISI 304 stainless steel using different cutting tools and performance of the used tools. Tool life, surface roughness and cutting forces have been determined experimentally and performed prediction models to constitute the base of determination of optimum cutting conditions, adaptive controls and tool condition monitoring studies.
Article
Full-text available
Language has long been a problem‐case for subsymbolic theories of mind. The reason for this is obvious: Language seems essentially symbolic. However, recent work has developed a potential solution to this problem, arguing that linguistic symbols are public objects which augment a fundamentally subsymbolic mind, rather than components of cognitive symbol‐processing. I shall argue that this strategy cannot work, on the grounds that human language acquisition consists in projecting linguistic structure onto environmental entities, rather than extracting this structure from them.
Article
Full-text available
This paper models unsupervised learning of an identity-based pattern (or copying) in speech called reduplication from raw continuous data with deep convolutional neural networks. We use the ciwGAN architecture (Beguš, 2021a) in which learning of meaningful representations in speech emerges from a requirement that the CNNs generate informative data. We propose a technique to wug-test CNNs trained on speech and, based on four generative tests, argue that the network learns to represent an identity-based pattern in its latent space. By manipulating only two categorical variables in the latent space, we can actively turn an unreduplicated form into a reduplicated form with no other substantial changes to the output in the majority of cases. We also argue that the network extends the identity-based pattern to unobserved data. Exploration of how meaningful representations of identity-based patterns emerge in CNNs and how the latent space variables outside of the training range correlate with identity-based patterns in the output has general implications for neural network interpretability.
ResearchGate has not been able to resolve any references for this publication.