Página 13 dos resultados de 471 itens digitais encontrados em 0.049 segundos

Storing non-uniformly distributed messages in networks of neural cliques

Boguslawski, Bartosz; Gripon, Vincent; Seguin, Fabrice; Heitzmann, Frédéric
Fonte: Universidade Cornell Publicador: Universidade Cornell
Tipo: Artigo de Revista Científica
Publicado em 24/07/2013 Português
Relevância na Pesquisa
37.23192%
Associative memories are data structures that allow retrieval of stored messages from part of their content. They thus behave similarly to human brain that is capable for instance of retrieving the end of a song given its beginning. Among different families of associative memories, sparse ones are known to provide the best efficiency (ratio of the number of bits stored to that of bits used). Nevertheless, it is well known that non-uniformity of the stored messages can lead to dramatic decrease in performance. We introduce several strategies to allow efficient storage of non-uniform messages in recently introduced sparse associative memories. We analyse and discuss the methods introduced. We also present a practical application example.; Comment: 21 pages, 8 figures, submitted to Neurocomputing

Understanding Locally Competitive Networks

Srivastava, Rupesh Kumar; Masci, Jonathan; Gomez, Faustino; Schmidhuber, Jürgen
Fonte: Universidade Cornell Publicador: Universidade Cornell
Tipo: Artigo de Revista Científica
Português
Relevância na Pesquisa
37.23192%
Recently proposed neural network activation functions such as rectified linear, maxout, and local winner-take-all have allowed for faster and more effective training of deep neural architectures on large and complex datasets. The common trait among these functions is that they implement local competition between small groups of computational units within a layer, so that only part of the network is activated for any given input pattern. In this paper, we attempt to visualize and understand this self-modularization, and suggest a unified explanation for the beneficial properties of such networks. We also show how our insights can be directly useful for efficiently performing retrieval over large datasets using neural networks.; Comment: 9 pages + 2 supplementary, Accepted to ICLR 2015 Conference track

INSTRUCT: Space-Efficient Structure for Indexing and Complete Query Management of String Databases

Dutta, Sourav; Bhattacharya, Arnab
Fonte: Universidade Cornell Publicador: Universidade Cornell
Tipo: Artigo de Revista Científica
Português
Relevância na Pesquisa
37.23192%
The tremendous expanse of search engines, dictionary and thesaurus storage, and other text mining applications, combined with the popularity of readily available scanning devices and optical character recognition tools, has necessitated efficient storage, retrieval and management of massive text databases for various modern applications. For such applications, we propose a novel data structure, INSTRUCT, for efficient storage and management of sequence databases. Our structure uses bit vectors for reusing the storage space for common triplets, and hence, has a very low memory requirement. INSTRUCT efficiently handles prefix and suffix search queries in addition to the exact string search operation by iteratively checking the presence of triplets. We also propose an extension of the structure to handle substring search efficiently, albeit with an increase in the space requirements. This extension is important in the context of trie-based solutions which are unable to handle such queries efficiently. We perform several experiments portraying that INSTRUCT outperforms the existing structures by nearly a factor of two in terms of space requirements, while the query times are better. The ability to handle insertion and deletion of strings in addition to supporting all kinds of queries including exact search...

A Time Efficient Indexing Scheme for Complex Spatiotemporal Retrieval

George, Lagogiannis; Nikos, Lorentzos; Spyros, Sioutas; Evaggelos, Theodoridis
Fonte: Universidade Cornell Publicador: Universidade Cornell
Tipo: Artigo de Revista Científica
Publicado em 10/05/2008 Português
Relevância na Pesquisa
37.23192%
The paper is concerned with the time efficient processing of spatiotemporal predicates, i.e. spatial predicates associated with an exact temporal constraint. A set of such predicates forms a buffer query or a Spatio-temporal Pattern (STP) Query with time. In the more general case of an STP query, the temporal dimension is introduced via the relative order of the spatial predicates (STP queries with order). Therefore, the efficient processing of a spatiotemporal predicate is crucial for the efficient implementation of more complex queries of practical interest. We propose an extension of a known approach, suitable for processing spatial predicates, which has been used for the efficient manipulation of STP queries with order. The extended method is supported by efficient indexing structures. We also provide experimental results that show the efficiency of the technique.; Comment: 6 pages, 7 figures, submitted to Sigmod Record

Web-enabling Cache Daemon for Complex Data

Voras, Ivan; Zagar, Mario
Fonte: Universidade Cornell Publicador: Universidade Cornell
Tipo: Artigo de Revista Científica
Publicado em 01/10/2009 Português
Relevância na Pesquisa
37.23192%
One of the most common basic techniques for improving the performance of web applications is caching frequently accessed data in fast data stores, colloquially known as cache daemons. In this paper we present a cache daemon suitable for storing complex data while maintaining fine-grained control over data storage, retrieval and expiry. Data manipulation in this cache daemon is performed via standard SQL statements so we call it SQLcached. It is a practical, usable solution already implemented in several large web sites.

Mining Associated Text and Images with Dual-Wing Harmoniums

Xing, Eric P.; Yan, Rong; Hauptmann, Alexander G.
Fonte: Universidade Cornell Publicador: Universidade Cornell
Tipo: Artigo de Revista Científica
Publicado em 04/07/2012 Português
Relevância na Pesquisa
37.23192%
We propose a multi-wing harmonium model for mining multimedia data that extends and improves on earlier models based on two-layer random fields, which capture bidirectional dependencies between hidden topic aspects and observed inputs. This model can be viewed as an undirected counterpart of the two-layer directed models such as LDA for similar tasks, but bears significant difference in inference/learning cost tradeoffs, latent topic representations, and topic mixing mechanisms. In particular, our model facilitates efficient inference and robust topic mixing, and potentially provides high flexibilities in modeling the latent topic spaces. A contrastive divergence and a variational algorithm are derived for learning. We specialized our model to a dual-wing harmonium for captioned images, incorporating a multivariate Poisson for word-counts and a multivariate Gaussian for color histogram. We present empirical results on the applications of this model to classification, retrieval and image annotation on news video collections, and we report an extensive comparison with various extant models.; Comment: Appears in Proceedings of the Twenty-First Conference on Uncertainty in Artificial Intelligence (UAI2005)

Comparative Results: Group Search Optimizer and Central Force Optimization

Formato, Richard A.
Fonte: Universidade Cornell Publicador: Universidade Cornell
Tipo: Artigo de Revista Científica
Português
Relevância na Pesquisa
37.23192%
This note compares the performance of two multidimensional search and optimization algorithms: Group Search Optimizer and Central Force Optimization. GSO is a new state-of-the-art algorithm that has gained some notoriety, consequently providing an excellent yardstick for measuring the performance of other algorithms. CFO is a novel deterministic metaheuristic that has performed well against GSO in previous tests. The CFO implementation reported here includes architectural improvements in errant probe retrieval and decision space adaptation that result in even better performance. Detailed results are provided for the twenty-three function benchmark suite used to evaluate GSO. CFO performs better than or essentially as well as GSO on twenty functions and nearly as well on one of the remaining three. Includes update 24 February 2010.; Comment: Includes detailed numerical results and source code in appendices. Update 02-24-10: Replaces Fig. A2(b) for improved visualization; corrects minor typos (note that trajectory plots were removed to meet file size restrictions - see Ver. 1 for complete set)

Riemannian Dictionary Learning and Sparse Coding for Positive Definite Matrices

Cherian, Anoop; Sra, Suvrit
Fonte: Universidade Cornell Publicador: Universidade Cornell
Tipo: Artigo de Revista Científica
Português
Relevância na Pesquisa
37.25806%
Data encoded as symmetric positive definite (SPD) matrices frequently arise in many areas of computer vision and machine learning. While these matrices form an open subset of the Euclidean space of symmetric matrices, viewing them through the lens of non-Euclidean Riemannian geometry often turns out to be better suited in capturing several desirable data properties. However, formulating classical machine learning algorithms within such a geometry is often non-trivial and computationally expensive. Inspired by the great success of dictionary learning and sparse coding for vector-valued data, our goal in this paper is to represent data in the form of SPD matrices as sparse conic combinations of SPD atoms from a learned dictionary via a Riemannian geometric approach. To that end, we formulate a novel Riemannian optimization objective for dictionary learning and sparse coding in which the representation loss is characterized via the affine invariant Riemannian metric. We also present a computationally simple algorithm for optimizing our model. Experiments on several computer vision datasets demonstrate superior classification and retrieval performance using our approach when compared to sparse coding via alternative non-Riemannian formulations.

Spontaneous Analogy by Piggybacking on a Perceptual System

Pickett, Marc; Aha, David W.
Fonte: Universidade Cornell Publicador: Universidade Cornell
Tipo: Artigo de Revista Científica
Publicado em 10/10/2013 Português
Relevância na Pesquisa
37.310671%
Most computational models of analogy assume they are given a delineated source domain and often a specified target domain. These systems do not address how analogs can be isolated from large domains and spontaneously retrieved from long-term memory, a process we call spontaneous analogy. We present a system that represents relational structures as feature bags. Using this representation, our system leverages perceptual algorithms to automatically create an ontology of relational structures and to efficiently retrieve analogs for new relational structures from long-term memory. We provide a demonstration of our approach that takes a set of unsegmented stories, constructs an ontology of analogical schemas (corresponding to plot devices), and uses this ontology to efficiently find analogs within new stories, yielding significant time-savings over linear analog retrieval at a small accuracy cost.; Comment: Proceedings of the 35th Meeting of the Cognitive Science Society, 2013

Optimizing the performance of Lattice Gauge Theory simulations with Streaming SIMD extensions

Srinivasan, Shyam
Fonte: Universidade Cornell Publicador: Universidade Cornell
Tipo: Artigo de Revista Científica
Publicado em 02/09/2013 Português
Relevância na Pesquisa
37.310671%
Two factors, which affect simulation quality are the amount of computing power and implementation. The Streaming SIMD (single instruction multiple data) extensions (SSE) present a technique for influencing both by exploiting the processor's parallel functionalism. In this paper, we show how SSE improves performance of lattice gauge theory simulations. We identified two significant trends through an analysis of data from various runs. The speed-ups were higher for single precision than double precision floating point numbers. Notably, though the use of SSE significantly improved simulation time, it did not deliver the theoretical maximum. There are a number of reasons for this: architectural constraints imposed by the FSB speed, the spatial and temporal patterns of data retrieval, ratio of computational to non-computational instructions, and the need to interleave miscellaneous instructions with computational instructions. We present a model for analyzing the SSE performance, which could help factor in the bottlenecks or weaknesses in the implementation, the computing architecture, and the mapping of software to the computing substrate while evaluating the improvement in efficiency. The model or framework would be useful in evaluating the use of other computational frameworks...

A Compact Graph Model of Handwritten Images: Integration into Authentification and Recognition

Popel, Denis V.
Fonte: Universidade Cornell Publicador: Universidade Cornell
Tipo: Artigo de Revista Científica
Publicado em 04/07/2002 Português
Relevância na Pesquisa
37.341172%
A novel algorithm for creating a mathematical model of curved shapes is introduced. The core of the algorithm is based on building a graph representation of the contoured image, which occupies less storage space than produced by raster compression techniques. Different advanced applications of the mathematical model are discussed: recognition of handwritten characters and verification of handwritten text and signatures for authentification purposes. Reducing the storage requirements due to the efficient mathematical model results in faster retrieval and processing times. The experimental outcomes in compression of contoured images and recognition of handwritten numerals are given.; Comment: 9 pages, 6 figures, 1 tables, SSPR'02

Partitioning Graph Databases - A Quantitative Evaluation

Averbuch, Alex; Neumann, Martin
Fonte: Universidade Cornell Publicador: Universidade Cornell
Tipo: Artigo de Revista Científica
Publicado em 22/01/2013 Português
Relevância na Pesquisa
37.341172%
Electronic data is growing at increasing rates, in both size and connectivity: the increasing presence of, and interest in, relationships between data. An example is the Twitter social network graph. Due to this growth demand is increasing for technologies that can process such data. Currently relational databases are the predominant technology, but they are poorly suited to processing connected data as they are optimized for index-intensive operations. Conversely, graph databases are optimized for graph computation. They link records by direct references, avoiding index lookups, and enabling retrieval of adjacent elements in constant time, regardless of graph size. However, as data volume increases these databases outgrow the resources of one computer and data partitioning becomes necessary. We evaluate the viability of using graph partitioning algorithms to partition graph databases. A prototype partitioned database was developed. Three partitioning algorithms explored and one implemented. Three graph datasets were used: two real and one synthetically generated. These were partitioned in various ways and the impact on database performance measured. We defined one synthetic access pattern per dataset and executed each on the partitioned datasets. Evaluation took place in a simulation environment...

Multimodal sparse representation learning and applications

Cha, Miriam; Gwon, Youngjune; Kung, H. T.
Fonte: Universidade Cornell Publicador: Universidade Cornell
Tipo: Artigo de Revista Científica
Publicado em 19/11/2015 Português
Relevância na Pesquisa
37.341172%
Unsupervised methods have proven effective for discriminative tasks in a single-modality scenario. In this paper, we present a multimodal framework for learning sparse representations that can capture semantic correlation between modalities. The framework can model relationships at a higher level by forcing the shared sparse representation. In particular, we propose the use of joint dictionary learning technique for sparse coding and formulate the joint representation for concision, cross-modal representations (in case of a missing modality), and union of the cross-modal representations. Given the accelerated growth of multimodal data posted on the Web such as YouTube, Wikipedia, and Twitter, learning good multimodal features is becoming increasingly important. We show that the shared representations enabled by our framework substantially improve the classification performance under both unimodal and multimodal settings. We further show how deep architectures built on the proposed framework are effective for the case of highly nonlinear correlations between modalities. The effectiveness of our approach is demonstrated experimentally in image denoising, multimedia event detection and retrieval on the TRECVID dataset (audio-video), category classification on the Wikipedia dataset (image-text)...

Multimodal diffusion geometry by joint diagonalization of Laplacians

Eynard, Davide; Glashoff, Klaus; Bronstein, Michael M.; Bronstein, Alexander M.
Fonte: Universidade Cornell Publicador: Universidade Cornell
Tipo: Artigo de Revista Científica
Português
Relevância na Pesquisa
37.341172%
We construct an extension of diffusion geometry to multiple modalities through joint approximate diagonalization of Laplacian matrices. This naturally extends classical data analysis tools based on spectral geometry, such as diffusion maps and spectral clustering. We provide several synthetic and real examples of manifold learning, retrieval, and clustering demonstrating that the joint diffusion geometry frequently better captures the inherent structure of multi-modal data. We also show that many previous attempts to construct multimodal spectral clustering can be seen as particular cases of joint approximate diagonalization of the Laplacians.

From Visual Attributes to Adjectives through Decompositional Distributional Semantics

Lazaridou, Angeliki; Dinu, Georgiana; Liska, Adam; Baroni, Marco
Fonte: Universidade Cornell Publicador: Universidade Cornell
Tipo: Artigo de Revista Científica
Português
Relevância na Pesquisa
37.341172%
As automated image analysis progresses, there is increasing interest in richer linguistic annotation of pictures, with attributes of objects (e.g., furry, brown...) attracting most attention. By building on the recent "zero-shot learning" approach, and paying attention to the linguistic nature of attributes as noun modifiers, and specifically adjectives, we show that it is possible to tag images with attribute-denoting adjectives even when no training data containing the relevant annotation are available. Our approach relies on two key observations. First, objects can be seen as bundles of attributes, typically expressed as adjectival modifiers (a dog is something furry, brown, etc.), and thus a function trained to map visual representations of objects to nominal labels can implicitly learn to map attributes to adjectives. Second, objects and attributes come together in pictures (the same thing is a dog and it is brown). We can thus achieve better attribute (and object) label retrieval by treating images as "visual phrases", and decomposing their linguistic representation into an attribute-denoting adjective and an object-denoting noun. Our approach performs comparably to a method exploiting manual attribute annotation, it outperforms various competitive alternatives in both attribute and object annotation...

Learning hashing with affinity-based loss functions using auxiliary coordinates

Raziperchikolaei, Ramin; Carreira-Perpiñán, Miguel Á.
Fonte: Universidade Cornell Publicador: Universidade Cornell
Tipo: Artigo de Revista Científica
Publicado em 21/01/2015 Português
Relevância na Pesquisa
37.341172%
In binary hashing, one wants to learn a function that maps a high-dimensional feature vector to a vector of binary codes, for application to fast image retrieval. This typically results in a difficult optimization problem, nonconvex and nonsmooth, because of the discrete variables involved. Much work has simply relaxed the problem during training, solving a continuous optimization, and truncating the codes a posteriori. This gives reasonable results but is suboptimal. Recent work has applied alternating optimization to the objective over the binary codes and achieved better results, but the hash function was still learned a posteriori, which remains suboptimal. We propose a general framework for learning hash functions using affinity-based loss functions that closes the loop and optimizes jointly over the hash functions and the binary codes. The resulting algorithm can be seen as a corrected, iterated version of the procedure of optimizing first over the codes and then learning the hash function. Compared to this, our optimization is guaranteed to obtain better hash functions while being not much slower, as demonstrated experimentally in various supervised and unsupervised datasets. In addition, the framework facilitates the design of optimization algorithms for arbitrary types of loss and hash functions.; Comment: 18 pages...

Pairwise Constraint Propagation on Multi-View Data

Lu, Zhiwu; Wang, Liwei
Fonte: Universidade Cornell Publicador: Universidade Cornell
Tipo: Artigo de Revista Científica
Publicado em 18/01/2015 Português
Relevância na Pesquisa
37.341172%
This paper presents a graph-based learning approach to pairwise constraint propagation on multi-view data. Although pairwise constraint propagation has been studied extensively, pairwise constraints are usually defined over pairs of data points from a single view, i.e., only intra-view constraint propagation is considered for multi-view tasks. In fact, very little attention has been paid to inter-view constraint propagation, which is more challenging since pairwise constraints are now defined over pairs of data points from different views. In this paper, we propose to decompose the challenging inter-view constraint propagation problem into semi-supervised learning subproblems so that they can be efficiently solved based on graph-based label propagation. To the best of our knowledge, this is the first attempt to give an efficient solution to inter-view constraint propagation from a semi-supervised learning viewpoint. Moreover, since graph-based label propagation has been adopted for basic optimization, we develop two constrained graph construction methods for interview constraint propagation, which only differ in how the intra-view pairwise constraints are exploited. The experimental results in cross-view retrieval have shown the promising performance of our inter-view constraint propagation.

A Community Contribution Framework for Sharing Materials Data with Materials Project

Huck, Patrick; Jain, Anubhav; Gunter, Dan; Winston, Donald; Persson, Kristin
Fonte: Universidade Cornell Publicador: Universidade Cornell
Tipo: Artigo de Revista Científica
Publicado em 16/10/2015 Português
Relevância na Pesquisa
37.310671%
As scientific discovery becomes increasingly data-driven, software platforms are needed to efficiently organize and disseminate data from disparate sources. This is certainly the case in the field of materials science. For example, Materials Project has generated computational data on over 60,000 chemical compounds and has made that data available through a web portal and REST interface. However, such portals must seek to incorporate community submissions to expand the scope of scientific data sharing. In this paper, we describe MPContribs, a computing/software infrastructure to integrate and organize contributions of simulated or measured materials data from users. Our solution supports complex submissions and provides interfaces that allow contributors to share analyses and graphs. A RESTful API exposes mechanisms for book-keeping, retrieval and aggregation of submitted entries, as well as persistent URIs or DOIs that can be used to reference the data in publications. Our approach isolates contributed data from a host project's quality-controlled core data and yet enables analyses across the entire dataset, programmatically or through customized web apps. We expect the developed framework to enhance collaborative determination of material properties and to maximize the impact of each contributor's dataset. In the long-term...

Analysis of Gait Pattern to Recognize the Human Activities

Gupta, Jay Prakash; Dixit, Pushkar; Singh, Nishant; Semwal, Vijay Bhaskar
Fonte: Universidade Cornell Publicador: Universidade Cornell
Tipo: Artigo de Revista Científica
Português
Relevância na Pesquisa
37.25806%
Human activity recognition based on the computer vision is the process of labelling image sequences with action labels. Accurate systems for this problem are applied in areas such as visual surveillance, human computer interaction and video retrieval.; Comment: This paper has been withdrawn by the author due to a crucial sign error in equation 3

A Study of the Xerox XNS Filing Protocol as Implemented on Several Heterogenous Systems

Flint, Edward
Fonte: Rochester Instituto de Tecnologia Publicador: Rochester Instituto de Tecnologia
Tipo: Tese de Doutorado
Português
Relevância na Pesquisa
37.341172%
The Xerox Network System is composed of heterogeneous processors connected across a variety of transmission media A series of protocols is defined to describe the communication mechanisms between system elements. One of these protocols, the Filing Protocol, defines a general purpose file management system. Current implementations of the protocol, although derived from the Xerox specification, fall short of providing the interconnectivity between elements desired in a heterogeneous network system. The definition of an easily implemented protocol subset that provides the common file system functions of retrieval, storage, enumeration/location and deletion is derived from experiences with several implementations. This definition and an accompanying implementation document provide a mechanism to guide future implementations toward increased interconnectivity.