In this experiment, a phoneme classification model has been developed using a Deep Neural Network based framework. The experiment is conducted in two phases. In the first phase, phoneme classification task has been pe...
详细信息
ISBN:
(纸本)9781538633335
In this experiment, a phoneme classification model has been developed using a Deep Neural Network based framework. The experiment is conducted in two phases. In the first phase, phoneme classification task has been performed. The deep-structured model provided good overall classification accuracy of 87.8%. All the phonemes are classified with precision and recall values. A confusion matrix of all the Bengali phonemes is derived. Using the confusion matrix, the phonemes are classified into nine groups. These nine groups provided better overall classification accuracy of 98.7%, and a new confusion matrix is derived for this nine groups. A lower confusion rate is observed this time. In the second phase of the experiment, the nine groups are reclassified into 15 groups using the manner of articulation based knowledge and the deep-structured model is retrained. The system provided 98.9% of overall classification accuracy this time. This result is almost equal to the overall accuracy which was observed for nine groups. But as the nine groups are redivided into 15 groups, the phoneme confusion in a single group became less which leads to a better phoneme classification model.
Hyperspectral unmixing is a challenging inverse problem that involves determining the fractional abundances of the representive material (endmembers) in each pixel. In this paper, we develop a neural network autoencod...
详细信息
ISBN:
(纸本)9781509049516
Hyperspectral unmixing is a challenging inverse problem that involves determining the fractional abundances of the representive material (endmembers) in each pixel. In this paper, we develop a neural network autoencoder, that dynamically exploits the sparsity of the abundances and enforces the abundance sum constraint (ASC) for hyperspectral unmixing. Instead of using the conventional mean square error (MSE) objective function, we use the spectral information divergence (SID) measure. Experiments are performed using a real hyperspectral dataset and we compare results obtained using both MSE and SID. It is demonstrated by qualitative inspection that using SID gives significantly better results than using MSE.
Deep learning is a very noteworthy technic that is take into consideration in the several fields. One of the most attractive subjects that need more attention in the prediction accuracy is fraud detection. As the deep...
详细信息
ISBN:
(纸本)9781538626405
Deep learning is a very noteworthy technic that is take into consideration in the several fields. One of the most attractive subjects that need more attention in the prediction accuracy is fraud detection. As the deep network can gradually learn the concepts of any complicated problem, using this technic in this realm is very beneficial. To do so, we propose a deep autoencoder to extract best features from the information of the credit card transactions and then append a softmax network to determine the class labels. Regarding the effect of features in such data employing an overcomplete autoencoder can map data to a high dimensional space and using the sparse models leads to be in a discriminative space that is useful for classification aims. The benefit of this method is the generality virtues that we can use such networks in several realms e.g. national intelligence, cyber security, marketing, medical informatics and so on. Another advantage is the ability to facing big datasets. As the learning phase is offline we can use it for a huge amount of data and generalize that is earned. Results can reveal the advantages of proposed method comparing to the state of the arts.
Deep learning has shown impressive performance on hard perceptual problems. However, researchers found deep learning systems to be vulnerable to small, specially crafted perturbations that are imperceptible to humans....
详细信息
ISBN:
(纸本)9781450349468
Deep learning has shown impressive performance on hard perceptual problems. However, researchers found deep learning systems to be vulnerable to small, specially crafted perturbations that are imperceptible to humans. Such perturbations cause deep learning systems to mis-classify adversarial examples, with potentially disastrous consequences where safety or security is crucial. Prior defenses against adversarial examples either targeted specific attacks or were shown to be ineffective. We propose MagNet, a framework for defending neural network classifiers against adversarial examples. MagNet neither modifies the protected classifier nor requires knowledge of the process for generating adversarial examples. MagNet includes one or more separate detector networks and a reformer network. The detector networks learn to differentiate between normal and adversarial examples by approximating the manifold of normal examples. Since they assume no specific process for generating adversarial examples, they generalize well. The reformer network moves adversarial examples towards the manifold of normal examples, which is effective for correctly classifying adversarial examples with small perturbation. We discuss the intrinsic difficulties in defending against whitebox attack and propose a mechanism to defend against gray-box attack. Inspired by the use of randomness in cryptography, we use diversity to strengthen MagNet. We show empirically that MagNet is effective against the most advanced state-of-the-art attacks in blackbox and graybox scenarios without sacrificing false positive rate on normal examples.
In online shopping, users usually express their intent through search queries. However, these queries are often ambiguous. For example, it is more likely (and easier) for users to write a query like "high-end bik...
详细信息
ISBN:
(纸本)9781450349130
In online shopping, users usually express their intent through search queries. However, these queries are often ambiguous. For example, it is more likely (and easier) for users to write a query like "high-end bike" than "21 speed carbon frames jamis or giant road bike". It is challenging to interpret these ambiguous queries and thus search result accuracy suffers. A user oftentimes needs to go through the frustrating process of refining search queries or self-teaching from possibly unstructured information. However, shopping is indeed a structured domain, that is composed of category hierarchy, brands, product lines, features, etc. It would be much better if a shopping site could understand users' intent through this structure, present organized information, and then find the items with the right categories, brands or features. In this paper we study the problem of inferring the latent intent from unstructured queries and mapping them to structured attributes. We present a novel framework that jointly learns this knowledge from user consumption behaviors and product metadata. We present a hybrid Long Shortterm Memory (LSTM) [10] joint model that is accurate and robust, even though user queries are noisy and product catalog is rapidly growing. Our study is conducted on a largescale dataset from Google Shopping, that is composed of millions of items and user queries along with their click responses. Extensive qualitative and quantitative evaluation shows that the proposed model is more accurate, concise, and robust than multiple possible alternatives. In terms of information retrieval (IR) performance, our model is able to improve the quality of current Google Shopping production system, which is a very strong baseline.
Spectral band power features are one of the most widely used features in the studies of electroencephalogram (EEG)-based emotion recognition. The power spectral density of EEG signals is partitioned into different ban...
详细信息
ISBN:
(纸本)9780769562155
Spectral band power features are one of the most widely used features in the studies of electroencephalogram (EEG)-based emotion recognition. The power spectral density of EEG signals is partitioned into different bands such as delta, theta, alpha and beta band etc. Though based on neuroscientific findings, the partition of frequency bands is somewhat on an adhoc basis, and the definition of frequency ranges of the bands of interest can vary between studies. On the other hand, it is also arguable that one definition of power bands could perform equally well on all subjects. In this paper, we propose to use autoencoder to automatically learn from each subject the salient frequency components from power spectral density estimated as periodogram by Fast Fourier Transform (FFT). We propose a network architecture especially for EEG feature extraction, one that adopts hidden unit clustering with added pooling neuron per cluster. The classification accuracy with features extracted by our proposed method is benchmarked against that with standard power features. Experimental results show that our proposed feature extraction method achieves accuracy ranging from 44% to 59% for three-emotion classification. We also see a 4-20% accuracy improvement over standard band power features.
Last decades have witnessed a vast amount of interest and research in recommendation systems. Collaborative filtering which uses the known preferences of a group of users to make recommendations or predictions of the ...
详细信息
ISBN:
(纸本)9781450346757
Last decades have witnessed a vast amount of interest and research in recommendation systems. Collaborative filtering which uses the known preferences of a group of users to make recommendations or predictions of the unknown preferences for other users, is one of the most successful approaches to build recommendation systems. Most previous collaborative filterin approaches employ the matrix factorization techniques to learn latent user feature profile and item feature profiles Also many subsequent works are proposed to incorporate users' social network information and items' attributions to further improve recommendation performance under the matrix factorization framework. However, the matrix factorization based methods may not make full use of the rating information, leading to unsatisfying performance. Recently deep learning has been approved to be able to fin good representations in natural language processing, image classification and so on. Along this line, we propose a collaborative ranking framework via REpresent4tion learning with Pair-wise constraints (REAP for short), in which autoencoder is used to simultaneously learn the latent factors of both users and items and pair-wise ranked loss define by (user, item) pairs is considered. Extensive experiments are conducted on five data sets to demonstrate the effectiveness of the proposed framework.
In this paper, we propose to leverage the emerging deep learning techniques for spatiotemporal modeling and prediction in cellular networks, based on big system data. First, we perform a preliminary analysis for a big...
详细信息
ISBN:
(纸本)9781509053360
In this paper, we propose to leverage the emerging deep learning techniques for spatiotemporal modeling and prediction in cellular networks, based on big system data. First, we perform a preliminary analysis for a big dataset from China Mobile, and use traffic load as an example to show non-zero temporal autocorrelation and non-zero spatial correlation among neighboring Base Stations (BSs), which motivate us to discover both temporal and spatial dependencies in our study. Then we present a hybrid deep learning model for spatiotemporal prediction, which includes a novel autoencoder-based deep model for spatial modeling and Long Short-Term Memory units (LSTMs) for temporal modeling. The autoencoder-based model consists of a Global Stacked autoencoder (GSAE) and multiple Local SAEs (LSAEs), which can offer good representations for input data, reduced model size, and support for parallel and application-aware training. Moreover, we present a new algorithm for training the proposed spatial model. We conducted extensive experiments to evaluate the performance of the proposed model using the China Mobile dataset. The results show that the proposed deep model significantly improves prediction accuracy compared to two commonly used baseline methods, ARIMA and SVR. We also present some results to justify effectiveness of the autoencoder-based spatial model.
Neural network can be used to "remember" speech patterns by encoding speech statistical regularity in network parameters. Clean speech can be "recalled" when noisy speech is input to the network. A...
详细信息
ISBN:
(纸本)9781622767595
Neural network can be used to "remember" speech patterns by encoding speech statistical regularity in network parameters. Clean speech can be "recalled" when noisy speech is input to the network. Adding more hidden layers can increase network capacity. But when the hidden layer size increases (deep network), the network is easily to be trapped to a local solution when traditional training strategy is used. Therefore, the performance of using a deep network sometimes is even worse than using a shallow network. In this study, we explore the greedy layer-wised pretraining strategy to train a deep autoencoder (DAE) for speech restoration, and apply the restored speech for noisy robust speech recognition. The DAB is first pretrained using quasi-Newton optimization algorithm layer by layer in which each layer is regarded as a shallow autoencoder. And the output of the preceding layer is served as the input to the next layer. The pretrained layers are stacked and "unrolled" to be a DAB. The pretrained parameters are served as initial parameters of the DAB which are used to refine training. The trained DAB is used as a filter for speech restoration when noisy speech is given. Noisy robust speech recognition experiments were done to examine the performance of the trained deep network. Experimental results show that the DAB trained with pretraining process significantly improved the performance of speech restoration from noisy input.
The purpose of this thesis is to evaluate if unsupervised anomaly detection, the task of finding anomalies in unlabelled data, can be used as a supportive tool for software life cycle management in finding errors whic...
详细信息
The purpose of this thesis is to evaluate if unsupervised anomaly detection, the task of finding anomalies in unlabelled data, can be used as a supportive tool for software life cycle management in finding errors which are tedious to detect manually. The goal is to apply the techniques of unsupervised machine learning on data-sets that are collected and analysed from a miniature-scaled research vehicle system that resembles the operation of a real automotive vehicles electrical architecture. Using a stacked autoencoder implemented with TensorFlow, the final application is able to detect anomalies within the collected data-sets from the research vehicle. This proves the concept of utilising machine learning for error detection as a viable method. Finally concluding whether the techniques of unsupervised anomaly detection is applicable on a larger scale for real automotive vehicles.
暂无评论