Query Examples Research Articles

Few-Shot Learning With a Strong Teacher.

Few-shot learning (FSL) aims to generate a classifier using limited labeled examples. Many existing works take the meta-learning approach, constructing a few-shot learner (a meta-model) that can learn from few-shot examples to generate a classifier. Typically, the few-shot learner is constructed or meta-trained by sampling multiple few-shot tasks in turn and optimizing the few-shot learner's performance in generating classifiers for those tasks. The performance is measured by how well the resulting classifiers classify the test (i.e., query) examples of those tasks. In this paper, we point out two potential weaknesses of this approach. First, the sampled query examples may not provide sufficient supervision for meta-training the few-shot learner. Second, the effectiveness of meta-learning diminishes sharply with the increasing number of shots (i.e., the number of training examples per class). To resolve these issues, we propose a novel meta-training objective for the few-shot learner, which is to encourage the few-shot learner to generate classifiers that perform like strong classifiers. Concretely, we associate each sampled few-shot task with a strong classifier, which is trained with ample labeled examples. The strong classifiers can be seen as the target classifiers that we hope the few-shot learner to generate given few-shot examples, and we use the strong classifiers to supervise the few-shot learner. We present an efficient way to construct the strong classifier, making our proposed objective an easily plug-and-play term to existing meta-learning based FSL methods. We validate our approach, (Learning with A Strong Teacher for few-SHOT learning), in combinations with many representative meta-learning methods. On several benchmark datasets including miniImageNet and tieredImageNet, our approach leads to a notable improvement across a variety of tasks. More importantly, with our approach, meta-learning based FSL methods can consistently outperform non-meta-learning based methods at different numbers of shots, even in many-shot settings, greatly strengthening their applicability.

PFEMed: Few-shot medical image classification using prior guided feature enhancement

Deep learning-based methods have recently demonstrated outstanding performance on general image classification tasks. As optimization of these methods is dependent on a large amount of labeled data, their application in medical image classification is limited. To address this issue, we propose PFEMed, a novel few-shot classification method for medical images. To extract general and specific features from medical images, this method employs a dual-encoder structure, that is, one encoder with fixed weights pre-trained on public image classification datasets and another encoder trained on the target medical dataset. In addition, we introduce a novel prior-guided Variational Autoencoder (VAE) module to enhance the robustness of the target feature, which is the concatenation of the general and specific features. Then, we match the target features extracted from both the support and query medical image samples and predict the category attribution of the query examples. Extensive experiments on several publicly available medical image datasets show that our method outperforms current state-of-the-art few-shot methods by a wide margin, particularly outperforming MetaMed on the Pap smear dataset by over 2.63%.

A Temporal JSON Data Model and Its Query Languages

As a ubiquitous form of data in the natural world, time has been widely used in many domains such as the military, finance, medical treatment, and environment. Temporal data models that are used to model the dynamic development process of data are constantly proposed, such as temporal database and temporal XML. With the rapid development and popularization of the Web, the requirement of dealing with diverse data efficiently is becoming more important. As a new generation of data exchange language, JSON (JavaScript Object Notation) has been widely used because it is lightweight, has fast parsing, and has high transmission efficiency. In this paper, we propose a novel temporal JSON data model. Based on the proposed temporal JSON data model, we further propose two temporal JSON query languages, t-JSONPath, and t-JSONiq, which are the temporal extensions to JSONPath and JSONiq, respectively. With the proposed temporal JSON data model, we demonstrate our temporal JSON query languages with typical temporal query examples and evaluate their query response times.

Task-Aware Feature Composition for Few-Shot Relation Classification

Relation classification tends to struggle when training data are limited or when it needs to adapt to unseen categories. In such challenging scenarios, recent approaches employ the metric-learning framework to measure similarities between query and support examples and to determine relation labels of the query sentences based on the similarities. However, these approaches treat each support class independently from one another, never looking at the entire task as a whole. Because of this, they are constrained to using a shared set of features for all meta-tasks, which hinders the ability to compose discriminative features for the task at hand. For example, if two similar relation types occur in a meta-task, the model needs to construct more detailed, task-related features instead of common features shared by all tasks. In this paper, we propose a novel task-aware relation classification model to tackle this issue. We first build a task embedding component to capture task-specific information, after which two mechanisms, such as task-specific gate and gated feature combination methods, are proposed to utilize the task-specific information to guide feature composition dynamically for each meta-task. Experiment results show that our model improves performance considerably over high performing baseline systems on both FewRel 1.0 and FewRel 2.0 benchmarks. Moreover, our proposed methods can be incorporated into metric-learning-based methods and significantly improve their performance.

Few-shot image classification with composite rotation based self-supervised auxiliary task

Many real-life problem settings have classes of data with very few examples for training. Deep learning networks do not perform well for such few-shot classes. In order to perform well in this setting, the networks should learn to extract highly discriminative and generic features. In this paper, we propose to use a composite rotation based self-supervised auxiliary task to improve the representation learning of the network so that it can extract such discriminative features. Our proposed composite rotation based auxiliary task rotates the image at two levels, i.e., it rotates patches inside the image (inner rotation) and also rotates the whole image (outer rotation), and assigns one out of 16 rotation classes to the transformed image. We jointly train the network on the main image classification task and the composite rotation based auxiliary task. This helps the network to learn to extract more generic and discriminative features, which in turn helps to improve its few-shot classification performance. Additionally, during the few-shot testing phase, we consolidate the predictions for images obtained by applying different composite rotation transformations to the same query image to improve the final prediction further. We perform experiments on several few-shot benchmark datasets and empirically show the efficacy of our method for various problem settings. We experimentally show that our method improves the performance of deep learning models on the few-shot classification, fine-grained few-shot classification, cross-domain few-shot classification, and transductive few-shot classification settings. We also experimentally show that models trained using our approach perform better than the baseline even when the query examples in the episode are not aligned with the support examples in the episode. We perform extensive ablation experiments to validate the different components of our approach. We also analyze the effect of our approach on the ability of the network to focus on the discriminative regions of the image.

Dynamic Boundary Time Warping for sub-sequence matching with few examples

The paper presents a novel method of finding a fragment in a long temporal sequence similar to the set of shorter sequences. We are the first to propose an algorithm for such a search that does not rely on computing the average sequence from query examples. Instead, we use query examples as is, utilizing all of them simultaneously. The introduced method based on the Dynamic Time Warping (DTW) technique is suited explicitly for few-shot query-by-example retrieval tasks. We evaluate it on two different few-shot problems from the field of Natural Language Processing. The results show it either outperforms baselines and previous approaches or achieves comparable results when a low number of examples is available.

Revisiting metric learning for few-shot image classification

The goal of few-shot learning is to recognize new visual concepts with just a few amount of labeled samples in each class. Recent effective metric-based few-shot approaches employ neural networks to learn a feature similarity comparison between query and support examples. However, the importance of feature embedding, i.e., exploring the relationship among training samples, is neglected. In this work, we present a simple yet powerful baseline for few-shot classification by emphasizing the importance of feature embedding. Specifically, we revisit the classical triplet network from deep metric learning, and extend it into a deep K-tuplet network for few-shot learning, utilizing the relationship among the input samples to learn a general representation learning via episode-training. Once trained, our network is able to extract discriminative features for unseen novel categories and can be seamlessly incorporated with a non-linear distance metric function to facilitate the few-shot classification. Our result on the miniImageNet benchmark outperforms other metric-based few-shot classification methods. More importantly, when evaluated on completely different datasets (Caltech-101, CUB-200, Stanford Dogs and Cars) using the model trained with miniImageNet, our method significantly outperforms prior methods, demonstrating its superior capability to generalize to unseen classes.

Approximate Query Answering Based on Topological Neighborhood and Semantic Similarity in OpenStreetMap

In this paper we focus on a pictorial query language, referred to as Geographical Pictorial Query Language (GeoPQL), and we revise its formal semantics by considering the polygon-polyline, polyline-polyline, and polygon-polygon topological relationships. This work proposes the Approximate Answering Engine (AAE) within a Distributed System, referred to as GeoPQLJSON (GeoPQLJ). The AAE provides approximate answers to query with empty results by following two directions: the Operator Conceptual Neighborhood (OCN) graph, and the OpenStreetMap (OSM) attribute hierarchy, giving maximum flexibility to the user choices. According to the former, the geo-operators of the queries can be replaced with the ones labeling the adjacent nodes of the OCN graph. By following the latter, the system evaluates the OSM attribute semantic similarity according to the information content approach, and proposes possible attribute replacements to the user. Note that the presence of OSM attributes allows the quick and direct access to large amount of geographical data, without requiring in our case the use of the topological elements. The functionalities of the Distributed GeoPQLJ System are illustrated by several query examples.

Embedded adaptive cross-modulation neural network for few-shot learning

Although deep neural networks have made great success in several scenarios of machine learning, they face persistent challenges in small training datasets learning scenarios. Few-shot learning aims to learn from a few labeled examples. However, the limited training samples and weakly distinguishable embedding vectors in a metric space often lead to unsatisfactory test results and directly calculating the distance between tensors can cause ambiguity. This paper proposes an embedded adaptive cross-modulation (EACM) method for few-shot learning which combines the information between support and query examples. Specifically, the inter-class categorizability between the support set prototype representations is enhanced by the adaptive cosine metric module to improve the accuracy of the few-shot recognition result. The learning is performed by using the cross-modulation module at many levels of abstraction layers along the prediction pipeline. The support set and query set feature cross-enhance, which improves the generalization ability and robustness of image recognition. Afterward, we further combine above two methods by a weight balance scalar to determine the task-related metric space and construct a joint loss function. Theoretical analysis demonstrates the generalization ability of EACM. We conduct comprehensive experiments on mini-ImageNet and CUB datasets. Experimental results show that our approach is the state-of-the-art approach by significant margins.

The neXtProt knowledgebase in 2020: data, tools and usability improvements.

The neXtProt knowledgebase (https://www.nextprot.org) is an integrative resource providing both data on human protein and the tools to explore these. In order to provide comprehensive and up-to-date data, we evaluate and add new data sets. We describe the incorporation of three new data sets that provide expression, function, protein-protein binary interaction, post-translational modifications (PTM) and variant information. New SPARQL query examples illustrating uses of the new data were added. neXtProt has continued to develop tools for proteomics. We have improved the peptide uniqueness checker and have implemented a new protein digestion tool. Together, these tools make it possible to determine which proteases can be used to identify trypsin-resistant proteins by mass spectrometry. In terms of usability, we have finished revamping our web interface and completely rewritten our API. Our SPARQL endpoint now supports federated queries. All the neXtProt data are available via our user interface, API, SPARQL endpoint and FTP site, including the new PEFF 1.0 format files. Finally, the data on our FTP site is now CC BY 4.0 to promote its reuse.

A Method of Emergent Event Evolution Reasoning Based on Ontology Cluster and Bayesian Network

Comprehensive real-time event information is critical to policymakers during emergency response and decision-making process. However, the development process of the emergent events has great uncertainty, and situational evolutions of emergencies are often difficult to use the fixed reasoning mode to attain. For this reason, this paper proposes a new method based on the ontology cluster for the evolution reasoning of emergency scenarios and extends the sematic web rule language to realize the scenario deduction, which can apply the Bayesian network to perform the conditional probability reasoning. A counterpart modeling and modifying of the Bayesian network optimization process is introduced. Besides, the probabilistic interpretation rules of atom components in context evolution are described with detailed query examples of emergency situation deducting and reasoning. The experimental results show that this approach is efficient in describing and capable of calculating the occurrence possibilities of the emergent events.

Sparse Subspace Modeling for Query by Example Spoken Term Detection

This paper focuses on the problem of query by example spoken term detection (QbE-STD) in zero-resource scenario. Current state-of-the-art approaches to tackle this problem rely on dynamic programming based template matching techniques using phone posterior features extracted at the output of a deep neural network. Previously, it has been shown that the space of phone posteriors is highly structured, as a union of low-dimensional subspaces. To exploit the temporal and sparse structure of the speech data, we investigate here three different QbE-STD systems based on sparse model recovery. More specifically, we use query examples to model the query subspace using dictionary for sparse coding. Reconstruction errors calculated using sparse representation of feature vectors are then used to characterize the underlying subspaces. The first approach uses these reconstruction errors in a dynamic programming framework to detect the spoken query, resulting in a much faster search compared to standard template matching. The other two methods aim at merging template matching and sparsity-based approaches to further improve the performance. The first one proposes to regularize the template matching local distances using sparse reconstruction errors. The second approach aims at using the sparse reconstruction errors to rescore (improve) the template matching likelihood. Experiments on two different databases (AMI and MediaEval) show that the proposed hybrid systems perform better than a highly competitive QbE-STD baseline system.

Improving the multimodal probabilistic semantic model by ELM classifiers

The multi-modal retrieval is considered as performing information retrieval among different modalities of multimedia information. Nowadays, it becomes increasingly important in the information science field. However, it is so difficult to bridge the meanings of different multimedia modalities that the performance of multimodal retrieval is deteriorated now. In this paper, we propose a new mechanism to build the relationship between visual and textual modalities and to verify the multimodal retrieval. Specifically, this mechanism depends on the multimodal binary classifiers based on the Extreme Learning Machine (ELM) to verify whether the answers are related to the query examples. Firstly, we propose the multimodal probabilistic semantic model to rank the answers according to their generative probabilities. Furthermore, we build the multimodal binary classifiers to filter out unrelated answers. The multimodal binary classifiers are called the word classifiers. It can improve the performance of the multimodal probabilistic semantic model. The experimental results show that the multimodal probabilistic semantic model and the word classifiers are effective and efficient. Also they demonstrate that the word classifiers based on ELM not only can improve the performance of the probabilistic semantic model but also can be easily applied to other probabilistic semantic models.

Modeling Topological Relationships between Fuzzy Spatio-Temporal Objects

Topological relationships between spatio-temporal objects are the most fundamental elements in spatio-temporal database systems, GIS, and image database systems. The research issue of modeling topological relationships has increasingly attracted attention, especially for querying of spatio-temporal objects and reasoning of topological relationships. Currently, topological relationship operating on spatio-temporal objects with precisely defined boundaries has been well studied. However, in the real world, spatio-temporal objects are not always crisp but with the nature of fuzziness and imprecision. Therefore, how to model topological relationship between fuzzy spatio-temporal objects is a significant topic and needs more investigations. This paper presents a study on modeling topological relationships between fuzzy spatio-temporal objects. Firstly, we give a model of fuzzy spatio-temporal objects in three-dimensional space and define those objects as moving fuzzy points , moving fuzzy lines , and moving fuzzy regions . On this basis, we propose a model for identifying basic topological relations between fuzzy spatio-temporal objects. Furthermore, in order to describe the evolution of basic topological relations over time, we give a model of complex topological relationships which are the sequences of basic relationships. The benefit of this model is that the complex topological relationships can be used as fuzzy spatio-temporal query operators in query languages. Finally, we provide some query examples to demonstrate fuzzy spatio-temporal queries in spatio-temporal database.

Data Management for Heterogeneous Genomic Datasets.

Next Generation Sequencing (NGS), a family of technologies for reading DNA and RNA, is changing biological research, and will soon change medical practice, by quickly providing sequencing data and high-level features of numerous individual genomes in different biological and clinical conditions. The availability of millions of whole genome sequences may soon become the biggest and most important "big data" problem of mankind. In this exciting framework, we recently proposed a new paradigm to raise the level of abstraction in NGS data management, by introducing a GenoMetric Query Language (GMQL) and demonstrating its usefulness through several biological query examples. Leveraging on that effort, here we motivate and formalize GMQL operations, especially focusing on the most characteristic and domain-specific ones. Furthermore, we address their efficient implementation and illustrate the architecture of the new software system that we have developed for their execution on big genomic data in a cloud computing environment, providing the evaluation of its performance. The new system implementation is available for download at the GMQL website (http://www.bioinformatics.deib.polimi.it/GMQL/); GMQL can also be tested through a set of predefined queries on ENCODE and Roadmap Epigenomics data at http://www.bioinformatics.deib.polimi.it/GMQL/queries/.

Time modelling in digital humanities. Challenges posed by the development of a database of medieval charters

Abstract Information management systems aimed at fully exploiting imperfect temporal data require advanced data modelling and data querying facilities. In this paper, we study how novel computational intelligence techniques can help to develop such facilities. More specifically, we describe how possibility theory can be used to model and handle the uncertainty that is inherent to vague, linguistic time specifications in a mathematically sound way. These mathematical time representations can then be stored in a database and advanced query evaluation techniques, for adequately coping with the represented uncertainty, can be used. Medieval charters are a typical example of documents on which related temporal information is often imperfect. To show the applicability of our technology, we illustrate how the development of the Diplomata Belgica database, a database describing medieval charters, can benefit from this novel technology. The added-value of the technology for future research in the humanities, in our case medieval history, is demonstrated by means of several illustrative query examples.

Learning to rank code examples for code search engines

Source code examples are used by developers to implement unfamiliar tasks by learning from existing solutions. To better support developers in finding existing solutions, code search engines are designed to locate and rank code examples relevant to user's queries. Essentially, a code search engine provides a ranking schema, which combines a set of ranking features to calculate the relevance between a query and candidate code examples. Consequently, the ranking schema places relevant code examples at the top of the result list. However, it is difficult to determine the configurations of the ranking schemas subjectively. In this paper, we propose a code example search approach that applies a machine learning technique to automatically train a ranking schema. We use the trained ranking schema to rank candidate code examples for new queries at run-time. We evaluate the ranking performance of our approach using a corpus of over 360,000 code snippets crawled from 586 open-source Android projects. The performance evaluation study shows that the learning-to-rank approach can effectively rank code examples, and outperform the existing ranking schemas by about 35.65 % and 48.42 % in terms of normalized discounted cumulative gain (NDCG) and expected reciprocal rank (ERR) measures respectively.

A New Approach for Representing and Querying Textual Attributes in Databases

In this paper, a new approach to deal with textual attributes in databases is presented. The basic idea is to give a new representation for these attributes, so that it allows us to treat them together with and in the same way than the other database attributes in querying, data warehousing, data mining processes, etc. Additionally, the transformation process of the textual attribute implies the obtaining of a global representation, which includes a great part of the attribute meaning. Moreover, this representation can be improved by using an ontology during the querying process, which enables the semantic queries. The formal representation model is presented in the paper, as well as the mathematical bases for dealing with it. The whole implementation process is also described by using a medical database to show the experimental results. Finally, a set of semantic query examples are offered to explain the advantages of this new approach.

The Composite Data Model: A Unified Approach for Combining and Querying Multiple Data Models

In this paper, we combine the characteristics of three fundamental data models in order to represent their semantics in a common framework. These fundamental data models include the familiar concepts of modeling (1) object classes (or entities), their properties (attributes) and relationships between them, (2) multidimensional objects and attributes that can be summarized over the dimensions, and (3) hierarchical structures. This model, called the Composite Data Model, facilitates combinations of these three model structures to be represented jointly in a single schema, thus providing more expressive and natural queries over them. The main advantage of the composite data model (CDM), and a composite query language (CQL) over it, is that any combination of the three fundamental models can be represented jointly based on explicit semantics of each of the fundamental data models. This is unlike existing data models that represent each data model individually or obscure the semantics of additional features being modeled. In order to develop a query language over the combined schemas, we introduce a new concept, referred to as anchor, which is an object class that acts as the focus of the query. We provide in the query language path structures relative to the anchor that facilitate data navigation and data manipulation. We develop the syntax and semantics of the proposed language, and illustrate its expressive power through numerous query examples, and comparisons to three other query languages: OQL, SPARQL, and XQuery.

GenoMetric Query Language: a novel approach to large-scale genomic data management.

Improvement of sequencing technologies and data processing pipelines is rapidly providing sequencing data, with associated high-level features, of many individual genomes in multiple biological and clinical conditions. They allow for data-driven genomic, transcriptomic and epigenomic characterizations, but require state-of-the-art 'big data' computing strategies, with abstraction levels beyond available tool capabilities. We propose a high-level, declarative GenoMetric Query Language (GMQL) and a toolkit for its use. GMQL operates downstream of raw data preprocessing pipelines and supports queries over thousands of heterogeneous datasets and samples; as such it is key to genomic 'big data' analysis. GMQL leverages a simple data model that provides both abstractions of genomic region data and associated experimental, biological and clinical metadata and interoperability between many data formats. Based on Hadoop framework and Apache Pig platform, GMQL ensures high scalability, expressivity, flexibility and simplicity of use, as demonstrated by several biological query examples on ENCODE and TCGA datasets. The GMQL toolkit is freely available for non-commercial use at http://www.bioinformatics.deib.polimi.it/GMQL/.