May 6, 2019

3454 words 17 mins read

Paper Group ANR 352

How should we evaluate supervised hashing?. A Survey on Learning to Hash. Defining Concepts of Emotion: From Philosophy to Science. The Projected Power Method: An Efficient Algorithm for Joint Alignment from Pairwise Differences. Fast and reliable stereopsis measurement at multiple distances with iPad. How to Use Temporal-Driven Constrained Cluster …

How should we evaluate supervised hashing?


Title	How should we evaluate supervised hashing?
Authors	Alexandre Sablayrolles, Matthijs Douze, Hervé Jégou, Nicolas Usunier
Abstract	Hashing produces compact representations for documents, to perform tasks like classification or retrieval based on these short codes. When hashing is supervised, the codes are trained using labels on the training data. This paper first shows that the evaluation protocols used in the literature for supervised hashing are not satisfactory: we show that a trivial solution that encodes the output of a classifier significantly outperforms existing supervised or semi-supervised methods, while using much shorter codes. We then propose two alternative protocols for supervised hashing: one based on retrieval on a disjoint set of classes, and another based on transfer learning to new classes. We provide two baseline methods for image-related tasks to assess the performance of (semi-)supervised hashing: without coding and with unsupervised codes. These baselines give a lower- and upper-bound on the performance of a supervised hashing scheme.
Tasks	Transfer Learning
Published	2016-09-21
URL	http://arxiv.org/abs/1609.06753v3
PDF	http://arxiv.org/pdf/1609.06753v3.pdf
PWC	https://paperswithcode.com/paper/how-should-we-evaluate-supervised-hashing
Repo
Framework

A Survey on Learning to Hash


Title	A Survey on Learning to Hash
Authors	Jingdong Wang, Ting Zhang, Jingkuan Song, Nicu Sebe, Heng Tao Shen
Abstract	Nearest neighbor search is a problem of finding the data points from the database such that the distances from them to the query point are the smallest. Learning to hash is one of the major solutions to this problem and has been widely studied recently. In this paper, we present a comprehensive survey of the learning to hash algorithms, categorize them according to the manners of preserving the similarities into: pairwise similarity preserving, multiwise similarity preserving, implicit similarity preserving, as well as quantization, and discuss their relations. We separate quantization from pairwise similarity preserving as the objective function is very different though quantization, as we show, can be derived from preserving the pairwise similarities. In addition, we present the evaluation protocols, and the general performance analysis, and point out that the quantization algorithms perform superiorly in terms of search accuracy, search time cost, and space cost. Finally, we introduce a few emerging topics.
Tasks	Quantization
Published	2016-06-01
URL	http://arxiv.org/abs/1606.00185v2
PDF	http://arxiv.org/pdf/1606.00185v2.pdf
PWC	https://paperswithcode.com/paper/a-survey-on-learning-to-hash
Repo
Framework

Defining Concepts of Emotion: From Philosophy to Science


Title	Defining Concepts of Emotion: From Philosophy to Science
Authors	Changqing Liu
Abstract	This paper is motivated by a series of (related) questions as to whether a computer can have pleasure and pain, what pleasure (and intensity of pleasure) is, and, ultimately, what concepts of emotion are. To determine what an emotion is, is a matter of conceptualization, namely, understanding and explicitly encoding the concept of emotion as people use it in everyday life. This is a notoriously difficult problem (Frijda, 1986, Fehr & Russell, 1984). This paper firstly shows why this is a difficult problem by aligning it with the conceptualization of a few other so called semantic primitives such as “EXIST”, “FORCE”, “BIG” (plus “LIMIT”). The definitions of these thought-to-be-indefinable concepts, given in this paper, show what formal definitions of concepts look like and how concepts are constructed. As a by-product, owing to the explicit account of the meaning of “exist”, the famous dispute between Einstein and Bohr is naturally resolved from linguistic point of view. Secondly, defending Frijda’s view that emotion is action tendency (or Ryle’s behavioral disposition (propensity)), we give a list of emotions defined in terms of action tendency. In particular, the definitions of pleasure and the feeling of beauty are presented. Further, we give a formal definition of “action tendency”, from which the concept of “intensity” of emotions (including pleasure) is naturally derived in a formal fashion. The meanings of “wish”, “wait”, “good”, “hot” are analyzed.
Tasks
Published	2016-02-11
URL	http://arxiv.org/abs/1604.08148v1
PDF	http://arxiv.org/pdf/1604.08148v1.pdf
PWC	https://paperswithcode.com/paper/defining-concepts-of-emotion-from-philosophy
Repo
Framework

The Projected Power Method: An Efficient Algorithm for Joint Alignment from Pairwise Differences


Title	The Projected Power Method: An Efficient Algorithm for Joint Alignment from Pairwise Differences
Authors	Yuxin Chen, Emmanuel Candes
Abstract	Various applications involve assigning discrete label values to a collection of objects based on some pairwise noisy data. Due to the discrete—and hence nonconvex—structure of the problem, computing the optimal assignment (e.g.~maximum likelihood assignment) becomes intractable at first sight. This paper makes progress towards efficient computation by focusing on a concrete joint alignment problem—that is, the problem of recovering $n$ discrete variables $x_i \in {1,\cdots, m}$, $1\leq i\leq n$ given noisy observations of their modulo differences ${x_i - x_j~\mathsf{mod}~m}$. We propose a low-complexity and model-free procedure, which operates in a lifted space by representing distinct label values in orthogonal directions, and which attempts to optimize quadratic functions over hypercubes. Starting with a first guess computed via a spectral method, the algorithm successively refines the iterates via projected power iterations. We prove that for a broad class of statistical models, the proposed projected power method makes no error—and hence converges to the maximum likelihood estimate—in a suitable regime. Numerical experiments have been carried out on both synthetic and real data to demonstrate the practicality of our algorithm. We expect this algorithmic framework to be effective for a broad range of discrete assignment problems.
Tasks
Published	2016-09-19
URL	http://arxiv.org/abs/1609.05820v3
PDF	http://arxiv.org/pdf/1609.05820v3.pdf
PWC	https://paperswithcode.com/paper/the-projected-power-method-an-efficient
Repo
Framework

Fast and reliable stereopsis measurement at multiple distances with iPad


Title	Fast and reliable stereopsis measurement at multiple distances with iPad
Authors	Manuel Rodriguez-Vallejo, Clara Llorens-Quintana, Diego Montagud, Walter D. Furlan, Juan A. Monsoriu
Abstract	Purpose: To present a new fast and reliable application for iPad (ST) for screening stereopsis at multiple distances. Methods: A new iPad application (app) based on a random dot stereogram was designed for screening stereopsis at multiple distances. Sixty-five subjects with no ocular diseases and wearing their habitual correction were tested at two different distances: 3 m and at 0.4 m. Results were compared with other commercial tests: TNO (at near) and Howard Dolman (at distance) Subjects were cited one week later in order to repeat the same procedures for assessing reproducibility of the tests. Results: Stereopsis at near was better with ST (40 arcsec) than with TNO (60 arcsec), but not significantly (p = 0.36). The agreement was good (k = 0.604) and the reproducibility was better with ST (k = 0.801) than with TNO (k = 0.715), in fact median difference between days was significant only with TNO (p = 0.02). On the other hand, poor agreement was obtained between HD and ST at far distance (k=0.04), obtaining significant differences in medians (p = 0.001) and poorer reliability with HD (k = 0.374) than with ST (k = 0.502). Conclusions: Screening stereopsis at near with a new iPad app demonstrated to be a fast and realiable. Results were in a good agreement with conventional tests as TNO, but it could not be compared at far vision with HD due to the limited resolution of the iPad.
Tasks
Published	2016-09-21
URL	http://arxiv.org/abs/1609.06669v1
PDF	http://arxiv.org/pdf/1609.06669v1.pdf
PWC	https://paperswithcode.com/paper/fast-and-reliable-stereopsis-measurement-at
Repo
Framework

How to Use Temporal-Driven Constrained Clustering to Detect Typical Evolutions


Title	How to Use Temporal-Driven Constrained Clustering to Detect Typical Evolutions
Authors	Marian-Andrei Rizoiu, Julien Velcin, Stéphane Lallich
Abstract	In this paper, we propose a new time-aware dissimilarity measure that takes into account the temporal dimension. Observations that are close in the description space, but distant in time are considered as dissimilar. We also propose a method to enforce the segmentation contiguity, by introducing, in the objective function, a penalty term inspired from the Normal Distribution Function. We combine the two propositions into a novel time-driven constrained clustering algorithm, called TDCK-Means, which creates a partition of coherent clusters, both in the multidimensional space and in the temporal space. This algorithm uses soft semi-supervised constraints, to encourage adjacent observations belonging to the same entity to be assigned to the same cluster. We apply our algorithm to a Political Studies dataset in order to detect typical evolution phases. We adapt the Shannon entropy in order to measure the entity contiguity, and we show that our proposition consistently improves temporal cohesion of clusters, without any significant loss in the multidimensional variance.
Tasks
Published	2016-01-11
URL	http://arxiv.org/abs/1601.02603v1
PDF	http://arxiv.org/pdf/1601.02603v1.pdf
PWC	https://paperswithcode.com/paper/how-to-use-temporal-driven-constrained
Repo
Framework

Reflections on Shannon Information: In search of a natural information-entropy for images


Title	Reflections on Shannon Information: In search of a natural information-entropy for images
Authors	Kieran G. Larkin
Abstract	It is not obvious how to extend Shannon’s original information entropy to higher dimensions, and many different approaches have been tried. We replace the English text symbol sequence originally used to illustrate the theory by a discrete, bandlimited signal. Using Shannon’s later theory of sampling we derive a new and symmetric version of the second order entropy in 1D. The new theory then naturally extends to 2D and higher dimensions, where by naturally we mean simple, symmetric, isotropic and parsimonious. Simplicity arises from the direct application of Shannon’s joint entropy equalities and inequalities to the gradient (del) vector field image embodying the second order relations of the scalar image. Parsimony is guaranteed by halving of the vector data rate using Papoulis’ generalized sampling expansion. The new 2D entropy measure, which we dub delentropy, is underpinned by a computable probability density function we call deldensity. The deldensity captures the underlying spatial image structure and pixel co-occurrence. It achieves this because each scalar image pixel value is nonlocally related to the entire gradient vector field. Both deldensity and delentropy are highly tractable and yield many interesting connections and useful inequalities. The new measure explicitly defines a realizable encoding algorithm and a corresponding reconstruction. Initial tests show that delentropy compares favourably with the conventional intensity-based histogram entropy and the compressed data rates of lossless image encoders (GIF, PNG, WEBP, JP2K-LS and JPG-LS) for a selection of images. The symmetric approach may have applications to higher dimensions and problems concerning image complexity measures.
Tasks
Published	2016-09-05
URL	http://arxiv.org/abs/1609.01117v1
PDF	http://arxiv.org/pdf/1609.01117v1.pdf
PWC	https://paperswithcode.com/paper/reflections-on-shannon-information-in-search
Repo
Framework

From narrative descriptions to MedDRA: automagically encoding adverse drug reactions


Title	From narrative descriptions to MedDRA: automagically encoding adverse drug reactions
Authors	Carlo Combi, Margherita Zorzi, Gabriele Pozzani, Ugo Moretti
Abstract	The collection of narrative spontaneous reports is an irreplaceable source for the prompt detection of suspected adverse drug reactions (ADRs): qualified domain experts manually revise a huge amount of narrative descriptions and then encode texts according to MedDRA standard terminology. The manual annotation of narrative documents with medical terminology is a subtle and expensive task, since the number of reports is growing up day-by-day. MagiCoder, a Natural Language Processing algorithm, is proposed for the automatic encoding of free-text descriptions into MedDRA terms. MagiCoder procedure is efficient in terms of computational complexity (in particular, it is linear in the size of the narrative input and the terminology). We tested it on a large dataset of about 4500 manually revised reports, by performing an automated comparison between human and MagiCoder revisions. For the current base version of MagiCoder, we measured: on short descriptions, an average recall of $86%$ and an average precision of $88%$; on medium-long descriptions (up to 255 characters), an average recall of $64%$ and an average precision of $63%$. From a practical point of view, MagiCoder reduces the time required for encoding ADR reports. Pharmacologists have simply to review and validate the MagiCoder terms proposed by the application, instead of choosing the right terms among the 70K low level terms of MedDRA. Such improvement in the efficiency of pharmacologists’ work has a relevant impact also on the quality of the subsequent data analysis. We developed MagiCoder for the Italian pharmacovigilance language. However, our proposal is based on a general approach, not depending on the considered language nor the term dictionary.
Tasks
Published	2016-12-12
URL	http://arxiv.org/abs/1612.03762v1
PDF	http://arxiv.org/pdf/1612.03762v1.pdf
PWC	https://paperswithcode.com/paper/from-narrative-descriptions-to-meddra
Repo
Framework

Families in the Wild (FIW): Large-Scale Kinship Image Database and Benchmarks


Title	Families in the Wild (FIW): Large-Scale Kinship Image Database and Benchmarks
Authors	Joseph P. Robinson, Ming Shao, Yue Wu, Yun Fu
Abstract	We present the largest kinship recognition dataset to date, Families in the Wild (FIW). Motivated by the lack of a single, unified dataset for kinship recognition, we aim to provide a dataset that captivates the interest of the research community. With only a small team, we were able to collect, organize, and label over 10,000 family photos of 1,000 families with our annotation tool designed to mark complex hierarchical relationships and local label information in a quick and efficient manner. We include several benchmarks for two image-based tasks, kinship verification and family recognition. For this, we incorporate several visual features and metric learning methods as baselines. Also, we demonstrate that a pre-trained Convolutional Neural Network (CNN) as an off-the-shelf feature extractor outperforms the other feature types. Then, results were further boosted by fine-tuning two deep CNNs on FIW data: (1) for kinship verification, a triplet loss function was learned on top of the network of pre-trained weights; (2) for family recognition, a family-specific softmax classifier was added to the network.
Tasks	Metric Learning
Published	2016-04-07
URL	http://arxiv.org/abs/1604.02182v2
PDF	http://arxiv.org/pdf/1604.02182v2.pdf
PWC	https://paperswithcode.com/paper/families-in-the-wild-fiw-large-scale-kinship
Repo
Framework

A Label Semantics Approach to Linguistic Hedges


Title	A Label Semantics Approach to Linguistic Hedges
Authors	Martha Lewis, Jonathan Lawry
Abstract	We introduce a model for the linguistic hedges `very' and` quite’ within the label semantics framework, and combined with the prototype and conceptual spaces theories of concepts. The proposed model emerges naturally from the representational framework we use and as such, has a clear semantic grounding. We give generalisations of these hedge models and show that they can be composed with themselves and with other functions, going on to examine their behaviour in the limit of composition.
Tasks
Published	2016-01-25
URL	http://arxiv.org/abs/1601.06738v1
PDF	http://arxiv.org/pdf/1601.06738v1.pdf
PWC	https://paperswithcode.com/paper/a-label-semantics-approach-to-linguistic
Repo
Framework

Validation of Tsallis Entropy In Inter-Modality Neuroimage Registration


Title	Validation of Tsallis Entropy In Inter-Modality Neuroimage Registration
Authors	Henrique Tomaz Amaral-Silva, Luiz Otavio Murta-Jr, Paulo Mazzoncini de Azevedo-Marques, Lauro Wichert-Ana, V. B. Surya Prasath, Colin Studholme
Abstract	Medical image registration plays an important role in determining topographic and morphological changes for functional diagnostic and therapeutic purposes. Manual alignment and semi-automated software still have been used; however they are subjective and make specialists spend precious time. Fully automated methods are faster and user-independent, but the critical point is registration reliability. Similarity measurement using Mutual Information (MI) with Shannon entropy (MIS) is the most common automated method that is being currently applied in medical images, although more reliable algorithms have been proposed over the last decade, suggesting improvements and different entropies; such as Studholme et al, (1999), who demonstrated that the normalization of Mutual Information (NMI) provides an invariant entropy measure for 3D medical image registration. In this paper, we described a set of experiments to evaluate the applicability of Tsallis entropy in the Mutual Information (MIT) and in the Normalized Mutual Information (NMIT) as cost functions for Magnetic Resonance Imaging (MRI), Positron Emission Tomography (PET) and Computed Tomography (CT) exams registration. The effect of changing overlap in a simple image model and clinical experiments on current entropies (Entropy Correlation Coefficient - ECC, MIS and NMI) and the proposed ones (MIT and NMT) showed NMI and NMIT with Tsallis parameter close to 1 as the best options (confidence and accuracy) for CT to MRI and PET to MRI automatic neuroimaging registration.
Tasks	Computed Tomography (CT), Image Registration, Medical Image Registration
Published	2016-11-06
URL	http://arxiv.org/abs/1611.01730v1
PDF	http://arxiv.org/pdf/1611.01730v1.pdf
PWC	https://paperswithcode.com/paper/validation-of-tsallis-entropy-in-inter
Repo
Framework

Learning Hidden Unit Contributions for Unsupervised Acoustic Model Adaptation


Title	Learning Hidden Unit Contributions for Unsupervised Acoustic Model Adaptation
Authors	Pawel Swietojanski, Jinyu Li, Steve Renals
Abstract	This work presents a broad study on the adaptation of neural network acoustic models by means of learning hidden unit contributions (LHUC) – a method that linearly re-combines hidden units in a speaker- or environment-dependent manner using small amounts of unsupervised adaptation data. We also extend LHUC to a speaker adaptive training (SAT) framework that leads to a more adaptable DNN acoustic model, working both in a speaker-dependent and a speaker-independent manner, without the requirements to maintain auxiliary speaker-dependent feature extractors or to introduce significant speaker-dependent changes to the DNN structure. Through a series of experiments on four different speech recognition benchmarks (TED talks, Switchboard, AMI meetings, and Aurora4) comprising 270 test speakers, we show that LHUC in both its test-only and SAT variants results in consistent word error rate reductions ranging from 5% to 23% relative depending on the task and the degree of mismatch between training and test data. In addition, we have investigated the effect of the amount of adaptation data per speaker, the quality of unsupervised adaptation targets, the complementarity to other adaptation techniques, one-shot adaptation, and an extension to adapting DNNs trained in a sequence discriminative manner.
Tasks	Speech Recognition
Published	2016-01-12
URL	http://arxiv.org/abs/1601.02828v2
PDF	http://arxiv.org/pdf/1601.02828v2.pdf
PWC	https://paperswithcode.com/paper/learning-hidden-unit-contributions-for
Repo
Framework

On the Geometry of Message Passing Algorithms for Gaussian Reciprocal Processes


Title	On the Geometry of Message Passing Algorithms for Gaussian Reciprocal Processes
Authors	Francesca Paola Carli
Abstract	Reciprocal processes are acausal generalizations of Markov processes introduced by Bernstein in 1932. In the literature, a significant amount of attention has been focused on developing dynamical models for reciprocal processes. Recently, probabilistic graphical models for reciprocal processes have been provided. This opens the way to the application of efficient inference algorithms in the machine learning literature to solve the smoothing problem for reciprocal processes. Such algorithms are known to converge if the underlying graph is a tree. This is not the case for a reciprocal process, whose associated graphical model is a single loop network. The contribution of this paper is twofold. First, we introduce belief propagation for Gaussian reciprocal processes. Second, we establish a link between convergence analysis of belief propagation for Gaussian reciprocal processes and stability theory for differentially positive systems.
Tasks
Published	2016-03-30
URL	http://arxiv.org/abs/1603.09279v2
PDF	http://arxiv.org/pdf/1603.09279v2.pdf
PWC	https://paperswithcode.com/paper/on-the-geometry-of-message-passing-algorithms
Repo
Framework

Making Contextual Decisions with Low Technical Debt


Title	Making Contextual Decisions with Low Technical Debt
Authors	Alekh Agarwal, Sarah Bird, Markus Cozowicz, Luong Hoang, John Langford, Stephen Lee, Jiaji Li, Dan Melamed, Gal Oshri, Oswaldo Ribas, Siddhartha Sen, Alex Slivkins
Abstract	Applications and systems are constantly faced with decisions that require picking from a set of actions based on contextual information. Reinforcement-based learning algorithms such as contextual bandits can be very effective in these settings, but applying them in practice is fraught with technical debt, and no general system exists that supports them completely. We address this and create the first general system for contextual learning, called the Decision Service. Existing systems often suffer from technical debt that arises from issues like incorrect data collection and weak debuggability, issues we systematically address through our ML methodology and system abstractions. The Decision Service enables all aspects of contextual bandit learning using four system abstractions which connect together in a loop: explore (the decision space), log, learn, and deploy. Notably, our new explore and log abstractions ensure the system produces correct, unbiased data, which our learner uses for online learning and to enable real-time safeguards, all in a fully reproducible manner. The Decision Service has a simple user interface and works with a variety of applications: we present two live production deployments for content recommendation that achieved click-through improvements of 25-30%, another with 18% revenue lift in the landing page, and ongoing applications in tech support and machine failure handling. The service makes real-time decisions and learns continuously and scalably, while significantly lowering technical debt.
Tasks	Multi-Armed Bandits
Published	2016-06-13
URL	http://arxiv.org/abs/1606.03966v2
PDF	http://arxiv.org/pdf/1606.03966v2.pdf
PWC	https://paperswithcode.com/paper/making-contextual-decisions-with-low
Repo
Framework

Scalable Models for Computing Hierarchies in Information Networks


Title	Scalable Models for Computing Hierarchies in Information Networks
Authors	Baoxu Shi, Tim Weninger
Abstract	Information hierarchies are organizational structures that often used to organize and present large and complex information as well as provide a mechanism for effective human navigation. Fortunately, many statistical and computational models exist that automatically generate hierarchies; however, the existing approaches do not consider linkages in information {\em networks} that are increasingly common in real-world scenarios. Current approaches also tend to present topics as an abstract probably distribution over words, etc rather than as tangible nodes from the original network. Furthermore, the statistical techniques present in many previous works are not yet capable of processing data at Web-scale. In this paper we present the Hierarchical Document Topic Model (HDTM), which uses a distributed vertex-programming process to calculate a nonparametric Bayesian generative model. Experiments on three medium size data sets and the entire Wikipedia dataset show that HDTM can infer accurate hierarchies even over large information networks.
Tasks
Published	2016-01-04
URL	http://arxiv.org/abs/1601.00626v1
PDF	http://arxiv.org/pdf/1601.00626v1.pdf
PWC	https://paperswithcode.com/paper/scalable-models-for-computing-hierarchies-in
Repo
Framework