Showing papers on "Pooling published in 2015"

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

Abstract: Existing deep convolutional neural networks (CNNs) require a fixed-size (e.g., 224 $\times$ 224) input image. This requirement is “artificial” and may reduce the recognition accuracy for the images or sub-images of an arbitrary size/scale. In this work, we equip the networks with another pooling strategy, “spatial pyramid pooling”, to eliminate the above requirement. The new network structure, called SPP-net, can generate a fixed-length representation regardless of image size/scale. Pyramid pooling is also robust to object deformations. With these advantages, SPP-net should in general improve all CNN-based image classification methods. On the ImageNet 2012 dataset, we demonstrate that SPP-net boosts the accuracy of a variety of CNN architectures despite their different designs. On the Pascal VOC 2007 and Caltech101 datasets, SPP-net achieves state-of-the-art classification results using a single full-image representation and no fine-tuning. The power of SPP-net is also significant in object detection. Using SPP-net, we compute the feature maps from the entire image only once, and then pool features in arbitrary regions (sub-images) to generate fixed-length representations for training the detectors. This method avoids repeatedly computing the convolutional features. In processing test images, our method is 24-102 $\times$ faster than the R-CNN method, while achieving better or comparable accuracy on Pascal VOC 2007. In ImageNet Large Scale Visual Recognition Challenge (ILSVRC) 2014, our methods rank #2 in object detection and #3 in image classification among all 38 teams. This manuscript also introduces the improvement made for this competition.

Learning Deep Features for Discriminative Localization

Abstract: In this work, we revisit the global average pooling layer proposed in [13], and shed light on how it explicitly enables the convolutional neural network to have remarkable localization ability despite being trained on image-level labels. While this technique was previously proposed as a means for regularizing training, we find that it actually builds a generic localizable deep representation that can be applied to a variety of tasks. Despite the apparent simplicity of global average pooling, we are able to achieve 37.1% top-5 error for object localization on ILSVRC 2014, which is remarkably close to the 34.2% top-5 error achieved by a fully supervised CNN approach. We demonstrate that our network is able to localize the discriminative image regions on a variety of tasks despite not being trained for them

Efficient object localization using Convolutional Networks

Abstract: Recent state-of-the-art performance on human-body pose estimation has been achieved with Deep Convolutional Networks (ConvNets). Traditional ConvNet architectures include pooling and sub-sampling layers which reduce computational requirements, introduce invariance and prevent over-training. These benefits of pooling come at the cost of reduced localization accuracy. We introduce a novel architecture which includes an efficient ‘position refinement’ model that is trained to estimate the joint offset location within a small region of the image. This refinement model is jointly trained in cascade with a state-of-the-art ConvNet model [21] to achieve improved accuracy in human joint location estimation. We show that the variance of our detector approaches the variance of human annotations on the FLIC [20] dataset and outperforms all existing approaches on the MPII-human-pose dataset [1].

Generalizing Pooling Functions in Convolutional Neural Networks: Mixed, Gated, and Tree

Abstract: We seek to improve deep neural networks by generalizing the pooling operations that play a central role in current architectures. We pursue a careful exploration of approaches to allow pooling to learn and to adapt to complex and variable patterns. The two primary directions lie in (1) learning a pooling function via (two strategies of) combining of max and average pooling, and (2) learning a pooling function in the form of a tree-structured fusion of pooling filters that are themselves learned. In our experiments every generalized pooling operation we explore improves performance when used in place of average or max pooling. We experimentally demonstrate that the proposed pooling operations provide a boost in invariance properties relative to conventional pooling and set the state of the art on several widely adopted benchmark datasets; they are also easy to implement, and can be applied within various deep neural network architectures. These benefits come with only a light increase in computational overhead during training and a very modest increase in the number of model parameters.

Target-dependent twitter sentiment classification with rich automatic features

Abstract: Target-dependent sentiment analysis on Twitter has attracted increasing research attention. Most previous work relies on syntax, such as automatic parse trees, which are subject to noise for informal text such as tweets. In this paper, we show that competitive results can be achieved without the use of syntax, by extracting a rich set of automatic features. In particular, we split a tweet into a left context and a right context according to a given target, using distributed word representations and neural pooling functions to extract features. Both sentiment-driven and standard embeddings are used, and a rich set of neural pooling functions are explored. Sentiment lexicons are used as an additional source of information for feature extraction. In standard evaluation, the conceptually simple method gives a 4.8% absolute improvement over the state-of-the-art on three-way targeted sentiment classification, achieving the best reported results for this task.

Delegation and pooling in international organizations

Abstract: We conceive authority of an international organization as latent in two independent dimensions: delegation by states to international agents and pooling in collective decision making bodies. We theorize that delegation and pooling are empirically as well as conceptually different. Delegation is an effort to deal with the transaction costs of cooperation which are greater in larger, broader, and correspondingly more complex organizations. Pooling reflects the tension between protecting or surrendering the national veto. This paper theorizes that delegation and pooling are constrained by two basic design features: a) the scope of an IO’s policy portfolio and b) the scale of its membership. We test these hypotheses with a new cross-sectional dataset that provides detailed and reliable information on IO decision making. Our major finding is that the design of international organizations is framed by stark and intelligible choices, but in surprising ways. Large membership organizations tend to have both more delegation and more pooling. The broader the policy scope of an IO, the more willing are its members to delegate, but the less willing they are to pool authority.

Spectral representations for convolutional neural networks

Abstract: Discrete Fourier transforms provide a significant speedup in the computation of convolutions in deep learning. In this work, we demonstrate that, beyond its advantages for efficient computation, the spectral domain also provides a powerful representation in which to model and train convolutional neural networks (CNNs). We employ spectral representations to introduce a number of innovations to CNN design. First, we propose spectral pooling, which performs dimensionality reduction by truncating the representation in the frequency domain. This approach preserves considerably more information per parameter than other pooling strategies and enables flexibility in the choice of pooling output dimensionality. This representation also enables a new form of stochastic regularization by randomized modification of resolution. We show that these methods achieve competitive results on classification and approximation tasks, without using any dropout or max-pooling. Finally, we demonstrate the effectiveness of complex-coefficient spectral parameterization of convolutional filters. While this leaves the underlying model unchanged, it results in a representation that greatly facilitates optimization. We observe on a variety of popular CNN configurations that this leads to significantly faster convergence during training.

Domain-size pooling in local descriptors: DSP-SIFT

Abstract: We introduce a simple modification of local image descriptors, such as SIFT, based on pooling gradient orientations across different domain sizes, in addition to spatial locations. The resulting descriptor, which we call DSP-SIFT, outperforms other methods in wide-baseline matching benchmarks, including those based on convolutional neural networks, despite having the same dimension of SIFT and requiring no training.

HDR-VDP-2.2: a calibrated method for objective quality prediction of high-dynamic range and standard images

Abstract: With the emergence of high-dynamic range (HDR) imaging, the existing visual signal processing systems will need to deal with both HDR and standard dynamic range (SDR) signals. In such systems, computing the objective quality is an important aspect in various optimization processes (e.g., video encoding). To that end, we present a newly calibrated objective method that can tackle both HDR and SDR signals. As it is based on the previously proposed HDR-VDP-2 method, we refer to the newly calibrated metric as HDR-VDP-2.2. Our main contribution is toward improving the frequency-based pooling in HDR-VDP-2 to enhance its objective quality prediction accuracy. We achieve this by formulating and solving a constrained optimization problem and thereby finding the optimal pooling weights. We also carried out extensive cross-validation as well as verified the performance of the new method on independent databases. These indicate clear improvement in prediction accuracy as compared with the default pooling weights. The source codes for HDR-VDP-2.2 are publicly available online for free download and use.

Spectral Representations for Convolutional Neural Networks

Abstract: Discrete Fourier transforms provide a significant speedup in the computation of convolutions in deep learning. In this work, we demonstrate that, beyond its advantages for efficient computation, the spectral domain also provides a powerful representation in which to model and train convolutional neural networks (CNNs). We employ spectral representations to introduce a number of innovations to CNN design. First, we propose spectral pooling, which performs dimensionality reduction by truncating the representation in the frequency domain. This approach preserves considerably more information per parameter than other pooling strategies and enables flexibility in the choice of pooling output dimensionality. This representation also enables a new form of stochastic regularization by randomized modification of resolution. We show that these methods achieve competitive results on classification and approximation tasks, without using any dropout or max-pooling. Finally, we demonstrate the effectiveness of complex-coefficient spectral parameterization of convolutional filters. While this leaves the underlying model unchanged, it results in a representation that greatly facilitates optimization. We observe on a variety of popular CNN configurations that this leads to significantly faster convergence during training.

Fractional max-pooling

Abstract: Convolutional networks almost always incorporate some form of spatial pooling, and very often it is max-pooling with = 2. Max-pooling act on the hidden layers of the network, reducing their size by an integer multiplicative factor . The amazing by product of discarding 75% of your data is that you build into the network a degree of invariance with respect to translations and elastic distortions. However, if you simply alternate convolutional layers with max-pooling layers, performance is limited due to the rapid reduction in spatial size, and the disjoint nature of the pooling regions. We have formulated a fractional version of max-pooling where is allowed to take non-integer values. Our version of maxpooling is stochastic as there are lots of different ways of constructing suitable pooling regions. We find that our form of fractional max-pooling reduces overfitting on a variety of datasets: for instance, we improve on the state of the art for CIFAR-100 without even using dropout.

Techniques for sharing network security event information

Abstract: This disclosure provides techniques for pooling and searching network security events reported by multiple sources. As information representing a security event is received from one source, it is searched against a central or distributed database representing events reported from multiple, diverse sources (e.g., different client networks). Either the search or correlated results can be filtered and/or routed according at least one characteristic associated with the networks, for example, to limit correlation to events reported by what are presumed to be similarly situated networks. The disclosed techniques facilitate faster identification of high-relevancy security event information, and thereby help facilitate faster threat identification and mitigation. Various techniques can be implemented as standalone software (e.g., for use by a private network) or for a central pooling and/or query service. This disclosure also provides different examples of actions that can be taken in response to search results.

On the Expressive Power of Deep Learning: A Tensor Analysis

Abstract: It has long been conjectured that hypotheses spaces suitable for data that is compositional in nature, such as text or images, may be more efficiently represented with deep hierarchical networks than with shallow ones. Despite the vast empirical evidence supporting this belief, theoretical justifications to date are limited. In particular, they do not account for the locality, sharing and pooling constructs of convolutional networks, the most successful deep learning architecture to date. In this work we derive a deep network architecture based on arithmetic circuits that inherently employs locality, sharing and pooling. An equivalence between the networks and hierarchical tensor factorizations is established. We show that a shallow network corresponds to CP (rank-1) decomposition, whereas a deep network corresponds to Hierarchical Tucker decomposition. Using tools from measure theory and matrix algebra, we prove that besides a negligible set, all functions that can be implemented by a deep network of polynomial size, require exponential size in order to be realized (or even approximated) by a shallow network. Since log-space computation transforms our networks into SimNets, the result applies directly to a deep learning architecture demonstrating promising empirical performance. The construction and theory developed in this paper shed new light on various practices and ideas employed by the deep learning community.

Max-Pooling Dropout for Regularization of Convolutional Neural Networks

Abstract: Recently, dropout has seen increasing use in deep learning. For deep convolutional neural networks, dropout is known to work well in fully-connected layers. However, its effect in pooling layers is still not clear. This paper demonstrates that max-pooling dropout is equivalent to randomly picking activation based on a multinomial distribution at training time. In light of this insight, we advocate employing our proposed probabilistic weighted pooling, instead of commonly used max-pooling, to act as model averaging at test time. Empirical evidence validates the superiority of probabilistic weighted pooling. We also compare max-pooling dropout and stochastic pooling, both of which introduce stochasticity based on multinomial distributions at pooling stage.

On the identification of sales forecasting models in the presence of promotions

Abstract: Shorter product life cycles and aggressive marketing, among other factors, have increased the complexity of sales forecasting. Forecasts are often produced using a Forecasting Support System that integrates univariate statistical forecasting with managerial judgment. Forecasting sales under promotional activity is one of the main reasons to use expert judgment. Alternatively, one can replace expert adjustments by regression models whose exogenous inputs are promotion features (price, display, etc). However, these regression models may have large dimensionality as well as multicollinearity issues. We propose a novel promotional model that overcomes these limitations. It combines Principal Component Analysis to reduce the dimensionality of the problem and automatically identifies the demand dynamics. For items with limited history, the proposed model is capable of providing promotional forecasts by selectively pooling information across established products. The performance of the model is compared against forecasts provided by experts and statistical benchmarks, on weekly data; outperforming both substantially.

STAP: Spatial-Temporal Attention-Aware Pooling for Action Recognition

Abstract: Human action recognition is valuable for numerous practical applications, e.g., gaming, video surveillance, and video search. In this paper we hypothesize that the classification of actions can be boosted by designing a smart feature pooling strategy under the prevalently used bag-of-words-based representation. Founded on automatic video saliency analysis, we propose the spatial-temporal attention-aware pooling scheme for feature pooling. First, the video saliencies are predicted using the video saliency model, and the localized spatial-temporal features are pooled at different saliency levels and video-saliency-guided channels are formed. Saliency-aware matching kernels are thus derived as the similarity measurement of these channels. Intuitively, the proposed kernels calculate the similarities of the video foreground (salient areas) or background (nonsalient areas) at different levels. Finally, the kernels are fed into popular support vector machines for action classification. Extensive experiments on three popular data sets for action classification validate the effectiveness of our proposed method, which outperforms the state-of-the-art methods, namely 95.3% on UCF Sports (better by 4.0%), 87.9% on YouTube data set (better by 2.5%), and achieves comparable results on Hollywood2 dataset.

Inside-Outside Net: Detecting Objects in Context with Skip Pooling and Recurrent Neural Networks

Abstract: It is well known that contextual and multi-scale representations are important for accurate visual recognition. In this paper we present the Inside-Outside Net (ION), an object detector that exploits information both inside and outside the region of interest. Contextual information outside the region of interest is integrated using spatial recurrent neural networks. Inside, we use skip pooling to extract information at multiple scales and levels of abstraction. Through extensive experiments we evaluate the design space and provide readers with an overview of what tricks of the trade are important. ION improves state-of-the-art on PASCAL VOC 2012 object detection from 73.9% to 76.4% mAP. On the new and more challenging MS COCO dataset, we improve state-of-art-the from 19.7% to 33.1% mAP. In the 2015 MS COCO Detection Challenge, our ION model won the Best Student Entry and finished 3rd place overall. As intuition suggests, our detection results provide strong evidence that context and multi-scale representations improve small object detection.

Free-Form Region Description with Second-Order Pooling

Abstract: Semantic segmentation and object detection are nowadays dominated by methods operating on regions obtained as a result of a bottom-up grouping process (segmentation) but use feature extractors developed for recognition on fixed-form (e.g. rectangular) patches, with full images as a special case. This is most likely suboptimal. In this paper we focus on feature extraction and description over free-form regions and study the relationship with their fixed-form counterparts. Our main contributions are novel pooling techniques that capture the second-order statistics of local descriptors inside such free-form regions. We introduce second-order generalizations of average and max-pooling that together with appropriate non-linearities, derived from the mathematical structure of their embedding space, lead to state-of-the-art recognition performance in semantic segmentation experiments without any type of local feature coding. In contrast, we show that codebook-based local feature coding is more important when feature extraction is constrained to operate over regions that include both foreground and large portions of the background, as typical in image classification settings, whereas for high-accuracy localization setups, second-order pooling over free-form regions produces results superior to those of the winning systems in the contemporary semantic segmentation challenges, with models that are much faster in both training and testing.

ABSTRACT Pooling Multiple Case Studies Using Synthetic Controls: An Application to Minimum Wage Policies *

Abstract: We propose a simple, distribution-free method for pooling synthetic control case studies using the mean percentile rank. We also test for heterogeneous treatment effects using the distribution of estimated ranks, which has a known form. We propose a cross-validation based procedure for model selection. Using 29 cases of state minimum wage increases between 1979 and 2013, we find a sizable, positive and statistically significant effect on the average teen wage. We do detect heterogeneity in the wage elasticities, consistent with differential bites in the policy. In contrast, the employment estimates suggest a small constant effect not distinguishable from zero.

Resource Pooling and Cost Allocation Among Independent Service Providers

Abstract: We study a situation where several independent service providers collaborate by complete pooling of their resources and customer streams into a joint service system. These service providers may represent such diverse organizations as hospitals that pool beds or maintenance firms that pool repairmen. We model the service systems as Erlang delay systems M/M/s queues that face a fixed cost rate per server and homogeneous delay costs for waiting customers. We examine rules to fairly allocate the collective costs of the pooled system amongst the participants by applying concepts from cooperative game theory. We consider both the case where players' numbers of servers are exogenously given and the scenario where any coalition picks an optimal number of servers. By exploiting new analytical properties of the continuous extension of the classic Erlang delay function, we provide sufficient conditions for the games under consideration to possess a core allocation i.e., an allocation that gives no group of players an incentive to split off and form a separate pool and to admit a population monotonic allocation scheme whereby adding extra players does not make anyone worse off. This is not guaranteed in general, as illustrated via examples.

Multiregion Bilinear Convolutional Neural Networks for Person Re-Identification

Abstract: In this work we propose a new architecture for person re-identification. As the task of re-identification is inherently associated with embedding learning and non-rigid appearance description, our architecture is based on the deep bilinear convolutional network (Bilinear-CNN) that has been proposed recently for fine-grained classification of highly non-rigid objects. While the last stages of the original Bilinear-CNN architecture completely removes the geometric information from consideration by performing orderless pooling, we observe that a better embedding can be learned by performing bilinear pooling in a more local way, where each pooling is confined to a predefined region. Our architecture thus represents a compromise between traditional convolutional networks and bilinear CNNs and strikes a balance between rigid matching and completely ignoring spatial information. We perform the experimental validation of the new architecture on the three popular benchmark datasets (Market-1501, CUHK01, CUHK03), comparing it to baselines that include Bilinear-CNN as well as prior art. The new architecture outperforms the baseline on all three datasets, while performing better than state-of-the-art on two out of three. The code and the pretrained models of the approach can be found at this https URL.

Analysis of MILP Techniques for the Pooling Problem

Abstract: The pq-relaxation for the pooling problem can be constructed by applying McCormick envelopes for each of the bilinear terms appearing in the so-called pq-formulation of the pooling problem. This relaxation can be strengthened by using piecewise-linear functions that over- and under-estimate each bilinear term. Although there is a significant amount of empirical evidence to show that such piecewise-linear relaxations, which can be written as mixed-integer linear programs (MILPs), yield good bounds for the pooling problem, to the best of our knowledge, no formal result regarding the quality of these relaxations is known. In this paper, we prove that the ratio of the upper bound obtained by solving piecewise-linear relaxations (objective function is maximization) to the optimal objective function value of the pooling problem is at most n, where n is the number of output nodes. Furthermore for any ϵ > 0 and for any piecewise-linear relaxation, there exists an instance where the ratio of the relaxation value t...

Signaling to Partially Informed Investors in the Newsvendor Model

Abstract: We analyze a signaling game between the manager of a firm and an investor in the firm. The manager has private information about the firm's demand and cares about the short-term stock price assigned by the investor. Previous research has shown that under continuous decision choices and the Intuitive Criterion refinement, the least-cost separating equilibrium will result, in which a low-quality firm chooses its optimal capacity and a high-quality firm over-invests in order to signal its quality to investors. We build on this research by showing the existence of pooling outcomes in which low-quality firms over-invest and high-quality firms under-invest so as to provide identical signals to investors. The pooling equilibrium is practically appealing because it yields a Pareto improvement compared to the least-cost separating equilibrium. Distinguishing features of our analysis are that: (i) we allow the capacity decision to have either discrete or continuous support, and (ii) we allow beliefs to be refined based on either the Undefeated refinement or the Intuitive Criterion refinement. We find that the newsvendor model parameters impact the likelihood of a pooling outcome, and this impact changes in both sign and magnitude depending on which refinement is used.

Actuarial fairness and solidarity in pooled annuity funds

Abstract: Various types of structures that enable a group of individuals to pool their mortality risk have been proposed in the literature. Collectively, the structures are called pooled annuity funds. Since the pooled annuity funds propose different methods of pooling mortality risk, we investigate the connections between them and find that they are genuinely different for a finite heterogeneous membership profile. We discuss the importance of actuarial fairness, defined as the expected benefits equalling the contributions for each member, in the context of pooling mortality risk and comment on whether actuarial unfairness can be seen as solidarity between members. We show that, with a finite number of members in the fund, the group self-annuitization scheme is not actuarially fair: some members subsidize the other members. The implication is that the members who are subsidizing the others may obtain a higher expected benefit by joining a fund with a more favorable membership profile. However, we find that the subsidies are financially significant only for very small or highly heterogeneous membership profiles.

The Potential of Servicizing as a Green Business Model

Abstract: It has been argued that servicizing business models, under which a firm sells the use of a product rather than the product itself, are environmentally beneficial. The main arguments are: First, under servicizing the firm charges customers based on the product usage. Second, the quantity of products required to meet customer needs may be smaller because the firm may be able to pool customer needs. Third, the firm may have an incentive to offer products with higher efficiency. Motivated by these arguments, we investigate the economic and environmental potential of servicizing business models. We endogenize the firm's choice between a pure sales, a pure servicizing, and a hybrid model with both sales and servicizing options, the pricing decisions and, the resulting customer usage. We consider two extremes of pooling efficacy, viz., no versus strong pooling. We find that under no pooling servicizing leads to higher environmental impact due to production but lower environmental impact due to use. In contrast, under strong pooling, when a hybrid business model is more profitable, it is also environmentally superior. However, a pure servicizing model is environmentally inferior for high production costs as it leads to a larger production quantity even under strong pooling. We also examine the product efficiency choice and find that the firm offers higher efficiency products only under servicizing models with strong pooling.

Subjective Bayesian beliefs

Abstract: A large literature suggests that many individuals do not apply Bayes’ Rule when making decisions that depend on them correctly pooling prior information and sample data. We replicate and extend a classic experimental study of Bayesian updating from psychology, employing the methods of experimental economics, with careful controls for the confounding effects of risk aversion. Our results show that risk aversion significantly alters inferences on deviations from Bayes’ Rule.

Differentiable pooling for unsupervised speaker adaptation

Abstract: This paper proposes a differentiable pooling mechanism to perform model-based neural network speaker adaptation. The proposed technique learns a speaker-dependent combination of activations within pools of hidden units, was shown to work well unsupervised, and does not require speaker-adaptive training. We have conducted a set of experiments on the TED talks data, as used in the IWSLT evaluations. Our results indicate that the approach can reduce word error rates (WERs) on standard IWSLT test sets by about 5–11% relative compared to speaker-independent systems and was found complementary to the recently proposed learning hidden units contribution (LHUC) approach, reducing WER by 6–13% relative. Both methods were also found to work well when adapting with small amounts of unsupervised data - 10 seconds is able to decrease the WER by 5% relative compared to the baseline speaker independent system.

Pooling of income and sharing of consumption within households

Abstract: There is extensive literature in economics and economic psychology on the allocation of household income within the household. In economics this refers to household decisions being independent of who generates the income in the household; in economic psychology it refers to the management of household finances. Here, we consider the link between the two concepts using a Danish expenditure survey providing information on both notions and on the assignment of expenditures. More importantly, we investigate whether either type of pooling is related to the sharing of expenditures between the two partners, and whether there are different correlations between the income distribution and the sharing of expenditures among double-career couples and other couples. We find that in most households the income distribution is correlated with the sharing of consumption—the economic approach—and that this holds true even if the household pools its resources—the economic psychology approach, implying that there is no strong relationship between the two approaches.