Coarse Category Research Articles

Hierarchical neural network with efficient selection inference

The image classification precision is vastly enhanced with the growing complexity of convolutional neural network (CNN) structures. However, the uneven visual separability between categories leads to various difficulties in classification. The hierarchical structure of categories can be leveraged to deal with it, but a few CNNs pay attention to the character of data. Besides, a network model with a hierarchical structure is promising to extract more specific features from the data than current CNNs, since, for the latter, all categories have the same fixed number of layers for feed-forward computation. In this paper, we propose to use category hierarchies to integrate ResNet-style modules to form a hierarchical network model in a top-down manner. To extract abundant discriminative features and improve the computation efficiency, we adopt residual block selection based on coarse categories to allocate different computation paths. Each residual block works as a switch to determine the JUMP or JOIN mode for an individual coarse category. Interestingly, since some categories need less feed-forward computation than others by jumping layers, the average inference time cost is reduced. Extensive experiments show that our hierarchical network achieves higher prediction accuracy with similar FLOPs on CIFAR-10 and CIFAR-100, SVHM, and Tiny-ImageNet datasets compared to original residual networks and other existing selection inference methods.

A Pullback Diagram in the Coarse Category

This paper studies the asymptotic product of two metric spaces. It is well defined if one of the spaces is visual or if both spaces are geodesic. In this case the asymptotic product is the pullback of a limit diagram in the coarse category. Using this product construction we can define a homotopy theory on coarse metric spaces in a natural way. We prove that all finite colimits exist in the coarse category.

The finite coarse shape - inverse systems approach and intrinsic approach

Given an arbitrary category \(\mathcal{C}\), a category \(pro^{*^f}\)-\(\mathcal{C}\) is constructed such that the known \(pro\)-\(\mathcal{C}\) category may be considered as a subcategory of \(pro^{*^f}\)-\(\mathcal{C}\) and that \(pro^{*^f}\)-\(\mathcal{C}\) may be considered as a subcategory of \(pro^*\)-\(\mathcal{C}\). Analogously to the construction of the shape category \(Sh_{(\mathcal{C},\mathcal{D})}\) and the coarse category \(Sh^*_{(\mathcal{C},\mathcal{D})}\), an (abstract) finite coarse shape category \(Sh^{*^f}_{(\mathcal{C},\mathcal{D})}\) is obtained. Between these three categories appropriate faithful functors are defined. The finite coarse shape is also defined by an intrinsic approach using the notion of the \(\epsilon\)-continuity. The isomorphism of the finite coarse shape categories obtained by these two approaches is constructed. Besides, an overview of some basic properties related to the notion of the \(\epsilon\)-continuity is given.

Property A and coarse embeddability for fuzzy metric spaces

Property A is a geometric property originally introduced for discrete metric spaces to provide a sufficient condition for coarse embeddability into Hilbert space, and it is defined via a Følner condition similar in spirit to the classical notion of amenability for groups. In this paper, we define property A for fuzzy metric spaces in the sense of George and Veeramani, show that it is an invariant in the coarse category of fuzzy metric spaces, and provide characterizations of it for uniformly locally finite fuzzy metric spaces. We also show that uniformly locally finite fuzzy metric spaces with property A are coarsely embeddable into Hilbert space.

Clustering by Errors: A Self-Organized Multitask Learning Method for Acoustic Scene Classification.

Acoustic scene classification (ASC) tries to inference information about the environment using audio segments. The inter-class similarity is a significant issue in ASC as acoustic scenes with different labels may sound quite similar. In this paper, the similarity relations amongst scenes are correlated with the classification error. A class hierarchy construction method by using classification error is then proposed and integrated into a multitask learning framework. The experiments have shown that the proposed multitask learning method improves the performance of ASC. On the TUT Acoustic Scene 2017 dataset, we obtain the ensemble fine-grained accuracy of 81.4%, which is better than the state-of-the-art. By using multitask learning, the basic Convolutional Neural Network (CNN) model can be improved by about 2.0 to 3.5 percent according to different spectrograms. The coarse category accuracies (for two to six super-classes) range from 77.0% to 96.2% by single models. On the revised version of the LITIS Rouen dataset, we achieve the ensemble fine-grained accuracy of 83.9%. The multitask learning models obtain an improvement of 1.6% to 1.8% compared to their basic models. The coarse category accuracies range from 94.9% to 97.9% for two to six super-classes with single models.

Hierarchical Modified Fast R-CNN for Object Detection

In object detection there is high degree of skewedness for objects' visual separability. It is difficult to distinguish object categories which demand dedicated classification. The deep convolutional neural networks (CNNs) are trained as N-way classifiers. As such considerable work is required towards leveraging hierarchical category structures. We present here Modified Fast region-based CNN (Mod Fast R-CNN) and Hierarchical Modified Fast region-based CNN (HMod Fast R-CNN) with deep CNNs being embedded considering categorical hierarchy. The easy classes are separated through coarse classifiers. The difficult classes are classified by fine classifiers. HMod Fast R-CNN is trained by initial components training which follows fine-tuning globally using multiple group discriminant analysis. The regularization is done using coarse category consistency. For large-scale recognition tasks, scalability is done considering conditional execution of fine category classifiers and layer parameters compression. Using MS-COCO (benchmark) CIFAR100 and VisualQA datasets we obtain good results. We build several different HMod Fast R-CNN versions where standard CNNs top-1 error is reduced significantly. HMod Fast R-CNN’s performance superiority with other object detectors on PASCAL VOC 2007 and VOC 2012 datasets are also highlighted.

Learning and Representation of Hierarchical Concepts in Hippocampus and Prefrontal Cortex.

A key aspect of conceptual knowledge is that it can be flexibly applied at different levels of abstraction, implying a hierarchical organization. It is yet unclear how this hierarchical structure is acquired and represented in the brain. Here we investigate the computations underlying the acquisition and representation of the hierarchical structure of conceptual knowledge in the hippocampal-prefrontal system of 32 human participants (22 females). We assessed the hierarchical nature of learning during a novel tree-like categorization task via computational model comparisons. The winning model allowed to extract and quantify estimates for accumulation and updating of hierarchical compared with single-feature-based concepts from behavior. We find that mPFC tracks accumulation of hierarchical conceptual knowledge over time, and mPFC and hippocampus both support trial-to-trial updating. As a function of those learning parameters, mPFC and hippocampus further show connectivity changes to rostro-lateral PFC, which ultimately represented the hierarchical structure of the concept in the final stages of learning. Our results suggest that mPFC and hippocampus support the integration of accumulated evidence and instantaneous updates into hierarchical concept representations in rostro-lateral PFC.SIGNIFICANCE STATEMENT A hallmark of human cognition is the flexible use of conceptual knowledge at different levels of abstraction, ranging from a coarse category level to a fine-grained subcategory level. While previous work probed the representational geometry of long-term category knowledge, it is unclear how this hierarchical structure inherent to conceptual knowledge is acquired and represented. By combining a novel hierarchical concept learning task with computational modeling of categorization behavior and concurrent fMRI, we differentiate the roles of key concept learning regions in hippocampus and PFC in learning computations and the representation of a hierarchical category structure.

Asymptotic dimension of fuzzy metric spaces

In this paper, we define asymptotic dimension of fuzzy metric spaces in the sense of George and Veeramani. We prove that asymptotic dimension is an invariant in the coarse category of fuzzy metric spaces. We also show several properties of asymptotic dimension in the fuzzy setting which resemble the properties of asymptotic dimension in the metric setting. We finish with calculating asymptotic dimension of a few fuzzy metric spaces.

Toward Fine-Grained Sketch-Based 3D Shape Retrieval.

In this paper we study, for the first time, the problem of fine-grained sketch-based 3D shape retrieval. We advocate the use of sketches as a fine-grained input modality to retrieve 3D shapes at instance-level - e.g., given a sketch of a chair, we set out to retrieve a specific chair from a gallery of all chairs. Fine-grained sketch-based 3D shape retrieval (FG-SBSR) has not been possible till now due to a lack of datasets that exhibit one-to-one sketch-3D correspondences. The first key contribution of this paper is two new datasets, consisting a total of 4,680 sketch-3D pairings from two object categories. Even with the datasets, FG-SBSR is still highly challenging because (i) the inherent domain gap between 2D sketch and 3D shape is large, and (ii) retrieval needs to be conducted at the instance level instead of the coarse category level matching as in traditional SBSR. Thus, the second contribution of the paper is the first cross-modal deep embedding model for FG-SBSR, which specifically tackles the unique challenges presented by this new problem. Core to the deep embedding model is a novel cross-modal view attention module which automatically computes the optimal combination of 2D projections of a 3D shape given a query sketch.

Coarse cohomology with twisted coefficients

Abstract To a coarse structure we associate a Grothendieck topology which is determined by coarse covers. A coarse map between coarse spaces gives rise to a morphism of Grothendieck topologies. This way we define sheaves and sheaf cohomology on coarse spaces. We obtain that sheaf cohomology is a functor on the coarse category: if two coarse maps are close they induce the same map in cohomology. There is a coarse version of a Mayer-Vietoris sequence and for every inclusion of coarse spaces there is a coarse version of relative cohomology. Cohomology with constant coefficients can be computed using the number of ends of a coarse space.

Characterization of pecan shells and their effect on the physical properties and acceptability of smoked chicken breast.

Pecan shells were used as a source of smoke for cooked chicken breast and compared with samples smoked with hickory, mesquite, and apple tree wood. The wood moisture content ranged from 7.94% to 11.43%, whereas ash ranged from 0.97% to 3.11%. The particle size varied among the smoke sources, with chopped pecan shells having more pieces in the "coarse" category. Several properties related to quality of the cooked meat were measured including moisture content, water activity, color, cook loss, expressible moisture, and maximum shear force. Moisture content ranged from 70.13% to 71.72%, whereas pH ranged from 6.39 to 6.43. Chicken breasts smoked with pecan shells were darker (L* = 72.86 compared to 74.94 to 76.57) and slightly redder (a* = 5.71; b* = 25.70) than other samples. Color development was localized to the surface of the meat. Cooking loss did not differ significantly among the samples (26.13% to 27.03%) nor did expressible moisture (9.68% to 11.87%). Meat tenderness was assessed by shear values, but did not differ among the samples. Consumer panels showed that all samples were well liked. Samples smoked with mesquite had slightly lower scores for flavor and overall likability, whereas the scores for samples smoked with hickory, apple, or pecan shells were no different. PRACTICAL APPLICATION: Many meat products are cooked in the presence of smoke produced from burning hardwoods. This work shows that smoke can be produced from waste pecan shells, resulting in chicken breast with similar yield, texture, moisture retention, and acceptability as that smoked with hickory, mesquite, or apple tree wood.

Comparative study of growth and yield attributes of different paddy cultivars under Tarai region of Uttar Pradesh

Paddy is a predominant kharif crop of Tarai region of Uttar Pradesh which plays a major role in augmenting the income of small and marginal farmers of tarai region of Uttar Pradesh. One of the major constraints of low productivity of paddy is non-adoption of improved package of practices. In this view, a study was carried out at Krishi Vigyan Kendra, Kushinagar farm during the kharif season 2015-16 with ten high yielding varieties of paddy under three categories, scented fine category viz. HUBR-10-9(V3), HUBR-2-1(V1), HUR-4-3(V7), P-1121(V10), P-1612(V6), fine category viz. BPT-5204(V8), HUR-36(V5), HUR-3022(V2) and coarse category viz. P-44(V9), Saryu-52(V4). The experiment was laid out in Randomized Complete Block Design with three replications. The observation were recorded on growth attributing traits viz. plant height (cm), number of tillers per sq.mt., leaf area index and dry matter of plant (g per sqm.) and yield attributing traits viz. Number of panicles per plant, panicle length (cm), number of grains per panicle, number of fertile grains per panicle, grain yield (q/ha), straw yield (q/ha) and grain: straw ratio. Good agricultural practices (GAP) were applied to raise healthy paddy crop. Maximum grain yield was recorded under V4 (59.5 q/ha) which was statistically at par with V3 (57.5 q/ha), V7 (55 q/ha), V9 (59.33 q/ha) and significantly superior than rest other varieties, while under scented group V3 recorded highest yield (57.5 q/ha) and under fine rice group V7 recorded highest yield (55 q/ha) both V3 and V7 are at par with V4. Maximum straw yield was recorded under V9 (70.5 q/ha) which was statistically at par with V3 (70.2 q/ha), V4 (70 q/ha), V7 (65 q/ha) and significantly superior than rest other varieties. Grain: straw ratio was maximum recorded under V4, V7 and V9 (0.85) which was statistically at par with rest other varieties. Although Saryu-52 is predominant paddy variety in this region and its yield was found maximum but it is closely related to P-44 (under coarse grain category). So it can be replaced by P-44. Under fine category BPT-5204 is predominant paddy variety in this region but its yield is poor and badly affected by draught and diseases, so under fine category it could be replace by HUR-3022 and under scented group HUBR-10-9 could be promoted in this region.

Adversarial training for fast arbitrary style transfer

We adversarially train a fast feed-forward generator network for arbitrary style transfer. The trained generator can output stylized images given unseen content and style image pairs as input during inference time. Our generator network is an encoder–decoder network that stylized the content by shifting the statistics of deep features, which has a novel mask module that can automatically and spatially decide the stylization level. We introduce an interface to manually and spatially control the stylization levels, and combine multiple styles in the generator. Our training objective is composed of traditional perceptual loss for content, traditional statistical loss for style, and adversarial loss that learns the intrinsic property of image styles from large-scale multi-domain artistic images. We use a conditional discriminator based on the coarse category of styles to tackle the challenging adversarial training for arbitrary style transfer, as both the input and output of our generator are diverse multi-domain images. We observe the mask module can stabilize adversarial training and help avoiding mode collapse in practice. As a side effect, our trained discriminator can be applied to rank and select representative stylized images, and the stylized images can significantly improve the efficiency of optimization-based style transfer. We qualitatively and quantitatively evaluate the proposed method, and compare with recent style transfer methods. We perform systematic study on arbitrary style transfer on multi-domain style images, and release our code and model at https://github.com/nightldj/behance_release.

Tree-CNN: from generalization to specialization

Traditional convolutional neural networks (CNNs) classify all categories by a single network, which passes all kinds of samples through totally the same network flow. In fact, it is quite challengeable to distinguish schooner with ketch and chair by a single network. To address it, we propose a new image classification architecture composed of a cluster algorithm and the Tree-CNN. The cluster algorithm devotes to classifying similar fine categories into a coarse category. The Tree-CNN is comprised of a Trunk-CNN for coarse classification of all categories and Branch-CNNs to treat different groups of similar categories differently. Branch-CNNs are fine-tuning based on the Trunk-CNN, which extracts the special feature map of image and divides it into fine categories. But Branch-CNNs bring extra computation and are hard to train. To address it, we introduce adaptive algorithm to balance the heavy computation and accuracy. We have tested Tree-CNNs based on CaffeNet, VGG16, and GoogLeNet in Caltech101 and Caltech256 for image classification. Experiment results show the superiority of the proposed Tree-CNN.

Monotone-light factorizations in coarse geometry

We introduce large scale analogues of topological monotone and light maps, which we call coarsely monotone and coarsely light maps respectively. We show that these two classes of maps constitute a factorization system on the coarse category. We also show how coarsely monotone maps arise from a reflection in a similar way to classically monotone maps, and prove that coarsely monotone maps are stable under those pullbacks which exist in the coarse category. For the case of maps between proper metric spaces, we exhibit some connections between the coarse and classical notions of monotone and light using the Higson corona. Finally, we look at some coarse properties which are preserved by coarsely light maps such as finite asymptotic dimension and exactness, and make some remarks on the situation for groups and group homomorphisms.

Asymptotic dimension of coarse spaces via maps to simplicial complexes

It is well-known that a paracompact space X is of covering dimension at most n if and only if any map f:X→K from X to a simplicial complex K can be pushed into its n-skeleton K(n). We use the same idea to characterize asymptotic dimension in the coarse category of arbitrary coarse spaces. Continuity of the map f is replaced by variation of f on elements of a uniformly bounded cover. In the same way, one can generalize Property A of G. Yu to arbitrary coarse spaces.

Heterogeneous Redistribution of Facial Subcategory Information Within and Outside the Face-Selective Domain in Primate Inferior Temporal Cortex.

The inferior temporal cortex (ITC) contains neurons selective to multiple levels of visual categories. However, the mechanisms by which these neurons collectively construct hierarchical category percepts remain unclear. By comparing decoding accuracy with simultaneously acquired electrocorticogram (ECoG), local field potentials (LFPs), and multi-unit activity in the macaque ITC, we show that low-frequency LFPs/ECoG in the early evoked visual response phase contain sufficient coarse category (e.g., face) information, which is homogeneous and enhanced by spatial summation of up to several millimeters. Late-induced high-frequency LFPs additionally carry spike-coupled finer category (e.g., species, view, and identity of the face) information, which is heterogeneous and reduced by spatial summation. Face-encoding neural activity forms a cluster in similar cortical locations regardless of whether it is defined by early evoked low-frequency signals or late-induced high-gamma signals. By contrast, facial subcategory-encoding activity is distributed, not confined to the face cluster, and dynamically increases its heterogeneity from the early evoked to late-induced phases. These findings support a view that, in contrast to the homogeneous and static coarse category-encoding neural cluster, finer category-encoding clusters are heterogeneously distributed even outside their parent category cluster and dynamically increase heterogeneity along with the local cortical processing in the ITC.

Categorical working memory representations are used in delayed estimation of continuous colors.

In the last decade, major strides have been made in understanding visual working memory through mathematical modeling of color production responses. In the delayed color estimation task (Wilken & Ma, 2004), participants are given a set of colored squares to remember, and a few seconds later asked to reproduce those colors by clicking on a color wheel. The degree of error in these responses is characterized with mathematical models that estimate working memory precision and the proportion of items remembered by participants. A standard mathematical model of color memory assumes that items maintained in memory are remembered through memory for precise details about the particular studied shade of color. We contend that this model is incomplete in its present form because no mechanism is provided for remembering the coarse category of a studied color. In the present work, we remedy this omission and present a model of visual working memory that includes both continuous and categorical memory representations. In 2 experiments, we show that our new model outperforms this standard modeling approach, which demonstrates that categorical representations should be accounted for by mathematical models of visual working memory. (PsycINFO Database Record

On spaces with connected Higson coronas

In this paper, we characterise metric spaces which have topologically connected Higson coronas. The characterisation is given by a natural categorical condition applied in the coarse category. We also give a characterisation in terms of coarse cohomology, and consider the special case of finitely generated groups and the more general case of abstract coarse spaces. Along the way, we exhibit some connections to the notion of ω-excisive decomposition introduced by Higson, Roe and Yu, and give a categorical characterisation of such decompositions.

Coarse property C and decomposition complexity

The coarse category was established by Roe [20] to distill the salient features of the large-scale approach to metric spaces and groups that was started by Gromov [13]. In this paper, we use the language of coarse spaces to define coarse versions of asymptotic property C [6] and decomposition complexity [16]. We prove that coarse property C implies coarse property A; we also show that these coarse versions share many of the features of their metric analogs such as preservation by products or unions.