Data representation is a fundamental task in machine learning. The representation of data affects the performance of the whole machine learning system. In a long history, the representation of data is done by feature engineering, and researchers aim at designing better features for specific tasks. Recently, the rapid development of deep learning and representation learning has brought new inspiration to various domains. In natural language processing, the most widely used feature representation is the Bag-of-Words model. This model has the data sparsity problem and cannot keep the word order information. Other features such as part-of-speech tagging or more complex syntax features can only fit for specific tasks in most cases. This thesis focuses on word representation and document representation. We compare the existing systems and present our new model. First, for generating word embeddings, we make comprehensive comparisons among existing word embedding models. In terms of theory, we figure out the relationship between the two most important models, i.e., Skip-gram and GloVe. In our experiments, we analyze three key points in generating word embeddings, including the model construction, the training corpus and parameter design. We evaluate word embeddings with three types of tasks, and we argue that they cover the existing use of word embeddings. Through theory and practical experiments, we present some guidelines for how to generate a good word embedding. Second, in Chinese character or word representation. We introduce the joint training of Chinese character and word. ... Third, for document representation, we analyze the existing document representation models, including recursive NNs, recurrent NNs and convolutional NNs. We point out the drawbacks of these models and present our new model, the recurrent convolutional neural networks. ...
With the large volume of new information created every day, determining the validity of information in a knowledge graph and filling in its missing parts are crucial tasks for many researchers and practitioners. To address this challenge, a number of knowledge graph completion methods have been developed using low-dimensional graph embeddings. Although researchers continue to improve these models using an increasingly complex feature space, we show that simple changes in the architecture of the underlying model can outperform state-of-the-art models without the need for complex feature engineering. In this work, we present a shared variable neural network model called ProjE that fills-in missing information in a knowledge graph by learning joint embeddings of the knowledge graph's entities and edges, and through subtle, but important, changes to the standard loss function. In doing so, ProjE has a parameter size that is smaller than 11 out of 15 existing methods while performing $37\%$ better than the current-best method on standard datasets. We also show, via a new fact checking task, that ProjE is capable of accurately determining the veracity of many declarative statements.
Most previous approaches to Chinese word segmentation formalize this problem as a character-based sequence labeling task where only contextual information within fixed sized local windows and simple interactions between adjacent tags can be captured. In this paper, we propose a novel neural framework which thoroughly eliminates context windows and can utilize complete segmentation history. Our model employs a gated combination neural network over characters to produce distributed representations of word candidates, which are then given to a long short-term memory (LSTM) language scoring model. Experiments on the benchmark datasets show that without the help of feature engineering as most existing approaches, our models achieve competitive or better performances with previous state-of-the-art methods.
Online content publishers often use catchy headlines for their articles in order to attract users to their websites. These headlines, popularly known as clickbaits, exploit a user's curiosity gap and lure them to click on links that often disappoint them. Existing methods for automatically detecting clickbaits rely on heavy feature engineering and domain knowledge. Here, we introduce a neural network architecture based on Recurrent Neural Networks for detecting clickbaits. Our model relies on distributed word representations learned from a large unannotated corpora, and character embeddings learned via Convolutional Neural Networks. Experimental results on a dataset of news headlines show that our model outperforms existing techniques for clickbait detection with an accuracy of 0.98 with F1-score of 0.98 and ROC-AUC of 0.99.
Extending the success of deep neural networks to natural language understanding and symbolic reasoning requires complex operations and external memory. Recent neural program induction approaches have attempted to address this problem, but are typically limited to differentiable memory, and consequently cannot scale beyond small synthetic tasks. In this work, we propose the Manager-Programmer-Computer framework, which integrates neural networks with non-differentiable memory to support abstract, scalable and precise operations through a friendly neural computer interface. Specifically, we introduce a Neural Symbolic Machine, which contains a sequence-to-sequence neural "programmer", and a non-differentiable "computer" that is a Lisp interpreter with code assist. To successfully apply REINFORCE for training, we augment it with approximate gold programs found by an iterative maximum likelihood training process. NSM is able to learn a semantic parser from weak supervision over a large knowledge base. It achieves new state-of-the-art performance on WebQuestionsSP, a challenging semantic parsing dataset, with weak supervision. Compared to previous approaches, NSM is end-to-end, therefore does not rely on feature engineering or domain specific knowledge.
In this paper, we study the problem of author identification under double-blind review setting, which is to identify potential authors given information of an anonymized paper. Different from existing approaches that rely heavily on feature engineering, we propose to use network embedding approach to address the problem, which can automatically represent nodes into lower dimensional feature vectors. However, there are two major limitations in recent studies on network embedding: (1) they are usually general-purpose embedding methods, which are independent of the specific tasks; and (2) most of these approaches can only deal with homogeneous networks, where the heterogeneity of the network is ignored. Hence, challenges faced here are two folds: (1) how to embed the network under the guidance of the author identification task, and (2) how to select the best type of information due to the heterogeneity of the network. To address the challenges, we propose a task-guided and path-augmented heterogeneous network embedding model. In our model, nodes are first embedded as vectors in latent feature space. Embeddings are then shared and jointly trained according to task-specific and network-general objectives. We extend the existing unsupervised network embedding to incorporate meta paths in heterogeneous networks, and select paths according to the specific task. The guidance from author identification task for network embedding is provided both explicitly in joint training and implicitly during meta path selection. Our experiments demonstrate that by using path-augmented network embedding with task guidance, our model can obtain significantly better accuracy at identifying the true authors comparing to existing methods.
In this paper, a Wide Learning architecture is proposed that attempts to automate the feature engineering portion of the machine learning (ML) pipeline. Feature engineering is widely considered as the most time consuming and expert knowledge demanding portion of any ML task. The proposed feature recommendation approach is tested on 3 healthcare datasets: a) PhysioNet Challenge 2016 dataset of phonocardiogram (PCG) signals, b) MIMIC II blood pressure classification dataset of photoplethysmogram (PPG) signals and c) an emotion classification dataset of PPG signals. While the proposed method beats the state of the art techniques for 2nd and 3rd dataset, it reaches 94.38% of the accuracy level of the winner of PhysioNet Challenge 2016. In all cases, the effort to reach a satisfactory performance was drastically less (a few days) than manual feature engineering.
Automatic profiling of social media users is an important task for supporting a multitude of downstream applications. While a number of studies have used social media content to extract and study collective social attributes, there is a lack of substantial research that addresses the detection of a user's industry. We frame this task as classification using both feature engineering and ensemble learning. Our industry-detection system uses both posted content and profile information to detect a user's industry with 64.3% accuracy, significantly outperforming the majority baseline in a taxonomy of fourteen industry classes. Our qualitative analysis suggests that a person's industry not only affects the words used and their perceived meanings, but also the number and type of emotions being expressed.
We introduce Deep-HiTS, a rotation invariant convolutional neural network (CNN) model for classifying images of transients candidates into artifacts or real sources for the High cadence Transient Survey (HiTS). CNNs have the advantage of learning the features automatically from the data while achieving high performance. We compare our CNN model against a feature engineering approach using random forests (RF). We show that our CNN significantly outperforms the RF model reducing the error by almost half. Furthermore, for a fixed number of approximately 2,000 allowed false transient candidates per night we are able to reduce the miss-classified real transients by approximately 1/5. To the best of our knowledge, this is the first time CNNs have been used to detect astronomical transient events. Our approach will be very useful when processing images from next generation instruments such as the Large Synoptic Survey Telescope (LSST). We have made all our code and data available to the community for the sake of allowing further developments and comparisons at https://github.com/guille-c/Deep-HiTS.
Ranking functions used in information retrieval are primarily used in the search engines and they are often adopted for various language processing applications. However, features used in the construction of ranking functions should be analyzed before applying it on a data set. This paper gives guidelines on construction of generalized ranking functions with application-dependent features. The paper prescribes a specific case of a generalized function for recommendation system using feature engineering guidelines on the given data set. The behavior of both generalized and specific functions are studied and implemented on the unstructured textual data. The proximity feature based ranking function has outperformed by 52% from regular BM25.
As high-throughput biological sequencing becomes faster and cheaper, the need to extract useful information from sequencing becomes ever more paramount, often limited by low-throughput experimental characterizations. For proteins, accurate prediction of their functions directly from their primary amino-acid sequences has been a long standing challenge. Here, machine learning using artificial recurrent neural networks (RNN) was applied towards classification of protein function directly from primary sequence without sequence alignment, heuristic scoring or feature engineering. The RNN models containing long-short-term-memory (LSTM) units trained on public, annotated datasets from UniProt achieved high performance for in-class prediction of four important protein functions tested, particularly compared to other machine learning algorithms using sequence-derived protein features. RNN models were used also for out-of-class predictions of phylogenetically distinct protein families with similar functions, including proteins of the CRISPR-associated nuclease, ferritin-like iron storage and cytochrome P450 families. Applying the trained RNN models on the partially unannotated UniRef100 database predicted not only candidates validated by existing annotations but also currently unannotated sequences. Some RNN predictions for the ferritin-like iron sequestering function were experimentally validated, even though their sequences differ significantly from known, characterized proteins and from each other and cannot be easily predicted using popular bioinformatics methods. As sequencing and experimental characterization data increases rapidly, the machine-learning approach based on RNN could be useful for discovery and prediction of homologues for a wide range of protein functions.
Machine learning models, such as neural networks, decision trees, random forests, and gradient boosting machines, accept a feature vector, and provide a prediction. These models learn in a supervised fashion where we provide feature vectors mapped to the expected output. It is common practice to engineer new features from the provided feature set. Such engineered features will either augment or replace portions of the existing feature vector. These engineered features are essentially calculated fields based on the values of the other features. Engineering such features is primarily a manual, time-consuming task. Additionally, each type of model will respond differently to different kinds of engineered features. This paper reports empirical research to demonstrate what kinds of engineered features are best suited to various machine learning model types. We provide this recommendation by generating several datasets that we designed to benefit from a particular type of engineered feature. The experiment demonstrates to what degree the machine learning model can synthesize the needed feature on its own. If a model can synthesize a planned feature, it is not necessary to provide that feature. The research demonstrated that the studied models do indeed perform differently with various types of engineered features.
The application of Deep Neural Networks for ranking in search engines may obviate the need for the extensive feature engineering common to current learning-to-rank methods. However, we show that combining simple relevance matching features like BM25 with existing Deep Neural Net models often substantially improves the accuracy of these models, indicating that they do not capture essential local relevance matching signals. We describe a novel deep Recurrent Neural Net-based model that we call Match-Tensor. The architecture of the Match-Tensor model simultaneously accounts for both local relevance matching and global topicality signals allowing for a rich interplay between them when computing the relevance of a document to a query. On a large held-out test set consisting of social media documents, we demonstrate not only that Match-Tensor outperforms BM25 and other classes of DNNs but also that it largely subsumes signals present in these models.
Spatiotemporal human representation based on 3D visual perception data is a rapidly growing research area. Based on the information sources, these representations can be broadly categorized into two groups based on RGB-D information or 3D skeleton data. Recently, skeleton-based human representations have been intensively studied and kept attracting an increasing attention, due to their robustness to variations of viewpoint, human body scale and motion speed as well as the realtime, online performance. This paper presents a comprehensive survey of existing space-time representations of people based on 3D skeletal data, and provides an informative categorization and analysis of these methods from the perspectives, including information modality, representation encoding, structure and transition, and feature engineering. We also provide a brief overview of skeleton acquisition devices and construction methods, enlist a number of public benchmark datasets with skeleton data, and discuss potential future research directions.
In this work, we present a novel background subtraction system that uses a deep Convolutional Neural Network (CNN) to perform the segmentation. With this approach, feature engineering and parameter tuning become unnecessary since the network parameters can be learned from data by training a single CNN that can handle various video scenes. Additionally, we propose a new approach to estimate background model from video. For the training of the CNN, we employed randomly 5 percent video frames and their ground truth segmentations taken from the Change Detection challenge 2014(CDnet 2014). We also utilized spatial-median filtering as the post-processing of the network outputs. Our method is evaluated with different data-sets, and the network outperforms the existing algorithms with respect to the average ranking over different evaluation metrics. Furthermore, due to the network architecture, our CNN is capable of real time processing.
In this paper, we developed a deep neural network (DNN) that learns to solve simultaneously the three tasks of the cQA challenge proposed by the SemEval-2016 Task 3, i.e., question-comment similarity, question-question similarity and new question-comment similarity. The latter is the main task, which can exploit the previous two for achieving better results. Our DNN is trained jointly on all the three cQA tasks and learns to encode questions and comments into a single vector representation shared across the multiple tasks. The results on the official challenge test set show that our approach produces higher accuracy and faster convergence rates than the individual neural networks. Additionally, our method, which does not use any manual feature engineering, approaches the state of the art established with methods that make heavy use of it.
Agglutinative languages such as Turkish, Finnish and Hungarian require morphological disambiguation before further processing due to the complex morphology of words. A morphological disambiguator is used to select the correct morphological analysis of a word. Morphological disambiguation is important because it generally is one of the first steps of natural language processing and its performance affects subsequent analyses. In this paper, we propose a system that uses deep learning techniques for morphological disambiguation. Many of the state-of-the-art results in computer vision, speech recognition and natural language processing have been obtained through deep learning models. However, applying deep learning techniques to morphologically rich languages is not well studied. In this work, while we focus on Turkish morphological disambiguation we also present results for French and German in order to show that the proposed architecture achieves high accuracy with no language-specific feature engineering or additional resource. In the experiments, we achieve 84.12, 88.35 and 93.78 morphological disambiguation accuracy among the ambiguous words for Turkish, German and French respectively.
In this paper we explore whether or not deep neural architectures can learn to classify Boolean satisfiability (SAT). We devote considerable time to discussing the theoretical properties of SAT. Then, we define a graph representation for Boolean formulas in conjunctive normal form, and train neural classifiers over general graph structures called Graph Neural Networks, or GNNs, to recognize features of satisfiability. To the best of our knowledge this has never been tried before. Our preliminary findings are potentially profound. In a weakly-supervised setting, that is, without problem specific feature engineering, Graph Neural Networks can learn features of satisfiability.
Bin Packing problems have been widely studied because of their broad applications in different domains. Known as a set of NP-hard problems, they have different vari- ations and many heuristics have been proposed for obtaining approximate solutions. Specifically, for the 1D variable sized bin packing problem, the two key sets of optimization heuristics are the bin assignment and the bin allocation. Usually the performance of a single static optimization heuristic can not beat that of a dynamic one which is tailored for each bin packing instance. Building such an adaptive system requires modeling the relationship between bin features and packing perform profiles. The primary drawbacks of traditional AI machine learnings for this task are the natural limitations of feature engineering, such as the curse of dimensionality and feature selection quality. We introduce a deep learning approach to overcome the drawbacks by applying a large training data set, auto feature selection and fast, accurate labeling. We show in this paper how to build such a system by both theoretical formulation and engineering practices. Our prediction system achieves up to 89% training accuracy and 72% validation accuracy to select the best heuristic that can generate a better quality bin packing solution.
Convective storm nowcasting has attracted substantial attention in various fields. Existing methods under a deep learning framework rely primarily on radar data. Although they perform nowcast storm advection well, it is still challenging to nowcast storm initiation and growth, due to the limitations of the radar observations. This paper describes the first attempt to nowcast storm initiation, growth, and advection simultaneously under a deep learning framework using multi-source meteorological data. To this end, we present a multi-channel 3D-cube successive convolution network (3D-SCN). As real-time re-analysis meteorological data can now provide valuable atmospheric boundary layer thermal dynamic information, which is essential to predict storm initiation and growth, both raw 3D radar and re-analysis data are used directly without any handcraft feature engineering. These data are formulated as multi-channel 3D cubes, to be fed into our network, which are convolved by cross-channel 3D convolutions. By stacking successive convolutional layers without pooling, we build an end-to-end trainable model for nowcasting. Experimental results show that deep learning methods achieve better performance than traditional extrapolation methods. The qualitative analyses of 3D-SCN show encouraging results of nowcasting of storm initiation, growth, and advection.