In Natural Language Processing (NLP), it is important to detect the relationship between two sequences or to generate a sequence of tokens given another observed sequence. We call the type of problems on modelling sequence pairs as sequence to sequence (seq2seq) mapping problems. A lot of research has been devoted to finding ways of tackling these problems, with traditional approaches relying on a combination of hand-crafted features, alignment models, segmentation heuristics, and external linguistic resources. Although great progress has been made, these traditional approaches suffer from various drawbacks, such as complicated pipeline, laborious feature engineering, and the difficulty for domain adaptation. Recently, neural networks emerged as a promising solution to many problems in NLP, speech recognition, and computer vision. Neural models are powerful because they can be trained end to end, generalise well to unseen examples, and the same framework can be easily adapted to a new domain. The aim of this thesis is to advance the state-of-the-art in seq2seq mapping problems with neural networks. We explore solutions from three major aspects: investigating neural models for representing sequences, modelling interactions between sequences, and using unpaired data to boost the performance of neural models. For each aspect, we propose novel models and evaluate their efficacy on various tasks of seq2seq mapping.
Bioresorbable scaffolds have become a popular choice for treatment of coronary heart disease, replacing traditional metal stents. Often, intravascular optical coherence tomography is used to assess potential malapposition after implantation and for follow-up examinations later on. Typically, the scaffold is manually reviewed by an expert, analyzing each of the hundreds of image slices. As this is time consuming, automatic stent detection and visualization approaches have been proposed, mostly for metal stent detection based on classic image processing. As bioresorbable scaffolds are harder to detect, recent approaches have used feature extraction and machine learning methods for automatic detection. However, these methods require detailed, pixel-level labels in each image slice and extensive feature engineering for the particular stent type which might limit the approaches' generalization capabilities. Therefore, we propose a deep learning-based method for bioresorbable scaffold visualization using only image-level labels. A convolutional neural network is trained to predict whether an image slice contains a metal stent, a bioresorbable scaffold, or no device. Then, we derive local stent strut information by employing weakly supervised localization using saliency maps with guided backpropagation. As saliency maps are generally diffuse and noisy, we propose a novel patch-based method with image shifting which allows for high resolution stent visualization. Our convolutional neural network model achieves a classification accuracy of 99.0 % for image-level stent classification which can be used for both high quality in-slice stent visualization and 3D rendering of the stent structure.
The convolutional neural network (CNN) architecture is increasingly being applied to new domains, such as malware detection, where it is able to learn malicious behavior from raw bytes extracted from executables. These architectures reach impressive performance with no feature engineering effort involved, but their robustness against active attackers is yet to be understood. Such malware detectors could face a new attack vector in the form of adversarial interference with the classification model. Existing evasion attacks intended to cause misclassification on test-time instances, which have been extensively studied for image classifiers, are not applicable because of the input semantics that prevents arbitrary changes to the binaries. This paper explores the area of adversarial examples for malware detection. By training an existing model on a production-scale dataset, we show that some previous attacks are less effective than initially reported, while simultaneously highlighting architectural weaknesses that facilitate new attack strategies for malware classification. Finally, we explore how generalizable different attack strategies are, the trade-offs when aiming to increase their effectiveness, and the transferability of single-step attacks.
The advent of representation learning methods enabled large performance gains on various language tasks, alleviating the need for manual feature engineering. While engineered representations are usually based on some linguistic understanding and are therefore more interpretable, learned representations are harder to interpret. Empirically studying the complementarity of both approaches can provide more linguistic insights that would help reach a better compromise between interpretability and performance. We present INFODENS, a framework for studying learned and engineered representations of text in the context of text classification tasks. It is designed to simplify the tasks of feature engineering as well as provide the groundwork for extracting learned features and combining both approaches. INFODENS is flexible, extensible, with a short learning curve, and is easy to integrate with many of the available and widely used natural language processing tools.
Diabetes is a major public health problem in the United States, affecting roughly 30 million people. Diabetes complications, along with the mental health comorbidities that often co-occur with them, are major drivers of high healthcare costs, poor outcomes, and reduced treatment adherence in diabetes. Here, we evaluate in a large state-wide population whether we can use artificial intelligence (AI) techniques to identify clusters of patient trajectories within the broader diabetes population in order to create cost-effective, narrowly-focused case management intervention strategies to reduce development of complications. This approach combined data from: 1) claims, 2) case management notes, and 3) social determinants of health from ~300,000 real patients between 2014 and 2016. We categorized complications as five types: Cardiovascular, Neuropathy, Opthalmic, Renal, and Other. Modeling was performed combining a variety of machine learning algorithms, including supervised classification, unsupervised clustering, natural language processing of unstructured care notes, and feature engineering. The results showed that we can predict development of diabetes complications roughly 83.5% of the time using claims data or social determinants of health data. They also showed we can reveal meaningful clusters in the patient population related to complications and mental health that can be used to cost-effective screening program, reducing the number of patients to be screened down by 85%. This study outlines creation of an AI framework to develop protocols to better address mental health comorbidities that lead to complications development in the diabetes population. Future work is described that outlines potential lines of research and the need for better addressing the 'people side' of the equation.
This paper presents an extensive comparative study of four neural network models, including feed-forward networks, convolutional networks, recurrent networks and long short-term memory networks, on two sentence classification datasets of English and Vietnamese text. We show that on the English dataset, the convolutional network models without any feature engineering outperform some competitive sentence classifiers with rich hand-crafted linguistic features. We demonstrate that the GloVe word embeddings are consistently better than both Skip-gram word embeddings and word count vectors. We also show the superiority of convolutional neural network models on a Vietnamese newspaper sentence dataset over strong baseline models. Our experimental results suggest some good practices for applying neural network models in sentence classification.
Entity resolution (ER) is one of the fundamental problems in data integration, where machine learning (ML) based classifiers often provide the state-of-the-art results. Considerable human effort goes into feature engineering and training data creation. In this paper, we investigate a new problem: Given a dataset D_T for ER with limited or no training data, is it possible to train a good ML classifier on D_T by reusing and adapting the training data of dataset D_S from same or related domain? Our major contributions include (1) a distributed representation based approach to encode each tuple from diverse datasets into a standard feature space; (2) identification of common scenarios where the reuse of training data can be beneficial; and (3) five algorithms for handling each of the aforementioned scenarios. We have performed comprehensive experiments on 12 datasets from 5 different domains (publications, movies, songs, restaurants, and books). Our experiments show that our algorithms provide significant benefits such as providing superior performance for a fixed training data size.
Automatically predicting the level of non-native English speakers given their written essays is an interesting machine learning problem. In this work I present the system "balikasg" that achieved the state-of-the-art performance in the CAp 2018 data science challenge among 14 systems. I detail the feature extraction, feature engineering and model selection steps and I evaluate how these decisions impact the system's performance. The paper concludes with remarks for future work.
We propose the genetic algorithm for time window optimization, which is an embedded genetic algorithm (GA), to optimize the time window (TW) of the attributes using feature selection and support vector machine. This GA is evolved using the results of a trading simulation, and it determines the best TW for each technical indicator. An appropriate evaluation was conducted using a walk-forward trading simulation, and the trained model was verified to be generalizable for forecasting other stock data. The results show that using the GA to determine the TW can improve the rate of return, leading to better prediction models than those resulting from using the default TW.
We investigate a Deep Learning based system for malware detection. In the investigation, we experiment with different combination of Deep Learning architectures including Auto-Encoders, and Deep Neural Networks with varying layers over Malicia malware dataset on which earlier studies have obtained an accuracy of (98%) with an acceptable False Positive Rates (1.07%). But these results were done using extensive man-made custom domain features and investing corresponding feature engineering and design efforts. In our proposed approach, besides improving the previous best results (99.21% accuracy and a False Positive Rate of 0.19%) indicates that Deep Learning based systems could deliver an effective defense against malware. Since it is good in automatically extracting higher conceptual features from the data, Deep Learning based systems could provide an effective, general and scalable mechanism for detection of existing and unknown malware.
Deep neural networks (DNN) excel at extracting patterns. Through representation learning and automated feature engineering on large datasets, such models have been highly successful in computer vision and natural language applications. Designing optimal network architectures from a principled or rational approach however has been less than successful, with the best successful approaches utilizing an additional machine learning algorithm to tune the network hyperparameters. However, in many technical fields, there exist established domain knowledge and understanding about the subject matter. In this work, we develop a novel furcated neural network architecture that utilizes domain knowledge as high-level design principles of the network. We demonstrate proof-of-concept by developing IL-Net, a furcated network for predicting the properties of ionic liquids, which is a class of complex multi-chemicals entities. Compared to existing state-of-the-art approaches, we show that furcated networks can improve model accuracy by approximately 20-35%, without using additional labeled data. Lastly, we distill two key design principles for furcated networks that can be adapted to other domains.
The real estate market is exposed to many fluctuations in prices because of existing correlations with many variables, some of which cannot be controlled or might even be unknown. Housing prices can increase rapidly (or in some cases, also drop very fast), yet the numerous listings available online where houses are sold or rented are not likely to be updated that often. In some cases, individuals interested in selling a house (or apartment) might include it in some online listing, and forget about updating the price. In other cases, some individuals might be interested in deliberately setting a price below the market price in order to sell the home faster, for various reasons. In this paper, we aim at developing a machine learning application that identifies opportunities in the real estate market in real time, i.e., houses that are listed with a price substantially below the market price. This program can be useful for investors interested in the housing market. We have focused in a use case considering real estate assets located in the Salamanca district in Madrid (Spain) and listed in the most relevant Spanish online site for home sales and rentals. The application is formally implemented as a regression problem that tries to estimate the market price of a house given features retrieved from public online listings. For building this application, we have performed a feature engineering stage in order to discover relevant features that allows for attaining a high predictive performance. Several machine learning algorithms have been tested, including regression trees, k-nearest neighbors, support vector machines and neural networks, identifying advantages and handicaps of each of them.
Detecting anomalous activity in human mobility data has a number of applications including road hazard sensing, telematic based insurance, and fraud detection in taxi services and ride sharing. In this paper we address two challenges that arise in the study of anomalous human trajectories: 1) a lack of ground truth data on what defines an anomaly and 2) the dependence of existing methods on significant pre-processing and feature engineering. While generative adversarial networks seem like a natural fit for addressing these challenges, we find that existing GAN based anomaly detection algorithms perform poorly due to their inability to handle multimodal patterns. For this purpose we introduce an infinite Gaussian mixture model coupled with (bi-directional) generative adversarial networks, IGMM-GAN, that is able to generate synthetic, yet realistic, human mobility data and simultaneously facilitates multimodal anomaly detection. Through estimation of a generative probability density on the space of human trajectories, we are able to generate realistic synthetic datasets that can be used to benchmark existing anomaly detection methods. The estimated multimodal density also allows for a natural definition of outlier that we use for detecting anomalous trajectories. We illustrate our methodology and its improvement over existing GAN anomaly detection on several human mobility datasets, along with MNIST.
Deep neural networks are gaining increasing popularity for the classic text classification task, due to their strong expressive power and less requirement for feature engineering. Despite such attractiveness, neural text classification models suffer from the lack of training data in many real-world applications. Although many semi-supervised and weakly-supervised text classification models exist, they cannot be easily applied to deep neural models and meanwhile support limited supervision types. In this paper, we propose a weakly-supervised method that addresses the lack of training data in neural text classification. Our method consists of two modules: (1) a pseudo-document generator that leverages seed information to generate pseudo-labeled documents for model pre-training, and (2) a self-training module that bootstraps on real unlabeled data for model refinement. Our method has the flexibility to handle different types of weak supervision and can be easily integrated into existing deep neural models for text classification. We have performed extensive experiments on three real-world datasets from different domains. The results demonstrate that our proposed method achieves inspiring performance without requiring excessive training data and outperforms baseline methods significantly.
The thermal subsystem of the Mars Express (MEX) spacecraft keeps the on-board equipment within its pre-defined operating temperatures range. To plan and optimize the scientific operations of MEX, its operators need to estimate in advance, as accurately as possible, the power consumption of the thermal subsystem. The remaining power can then be allocated for scientific purposes. We present a machine learning pipeline for efficiently constructing accurate predictive models for predicting the power of the thermal subsystem on board MEX. In particular, we employ state-of-the-art feature engineering approaches for transforming raw telemetry data, in turn used for constructing accurate models with different state-of-the-art machine learning methods. We show that the proposed pipeline considerably improve our previous (competition-winning) work in terms of time efficiency and predictive performance. Moreover, while achieving superior predictive performance, the constructed models also provide important insight into the spacecraft's behavior, allowing for further analyses and optimal planning of MEX's operation.
We design and build the first neural temporal dependency parser. It utilizes a neural ranking model with minimal feature engineering, and parses time expressions and events in a text into a temporal dependency tree structure. We evaluate our parser on two domains: news reports and narrative stories. In a parsing-only evaluation setup where gold time expressions and events are provided, our parser reaches 0.81 and 0.70 f-score on unlabeled and labeled parsing respectively, a result that is very competitive against alternative approaches. In an end-to-end evaluation setup where time expressions and events are automatically recognized, our parser beats two strong baselines on both data domains. Our experimental results and discussions shed light on the nature of temporal dependency structures in different domains and provide insights that we believe will be valuable to future research in this area.
Translating characters instead of words or word-fragments has the potential to simplify the processing pipeline for neural machine translation (NMT), and improve results by eliminating hyper-parameters and manual feature engineering. However, it results in longer sequences in which each symbol contains less information, creating both modeling and computational challenges. In this paper, we show that the modeling problem can be solved by standard sequence-to-sequence architectures of sufficient depth, and that deep models operating at the character level outperform identical models operating over word fragments. This result implies that alternative architectures for handling character input are better viewed as methods for reducing computation time than as improved ways of modeling longer sequences. From this perspective, we evaluate several techniques for character-level NMT, verify that they do not match the performance of our deep character baseline model, and evaluate the performance versus computation time tradeoffs they offer. Within this framework, we also perform the first evaluation for NMT of conditional computation over time, in which the model learns which timesteps can be skipped, rather than having them be dictated by a fixed schedule specified before training begins.
Text segmentation plays an important role in various Natural Language Processing (NLP) tasks like summarization, context understanding, document indexing and document noise removal. Previous methods for this task require manual feature engineering, huge memory requirements and large execution times. To the best of our knowledge, this paper is the first one to present a novel supervised neural approach for text segmentation. Specifically, we propose an attention-based bidirectional LSTM model where sentence embeddings are learned using CNNs and the segments are predicted based on contextual information. This model can automatically handle variable sized context information. Compared to the existing competitive baselines, the proposed model shows a performance improvement of ~7% in WinDiff score on three benchmark datasets.
With recent progress in algorithms and the availability of massive amounts of computation power, application of machine learning techniques is becoming a hot topic in the oil and gas industry. One of the most promising aspects to apply machine learning to the upstream field is the rock facies classification in reservoir characterization, which is crucial in determining the net pay thickness of reservoirs, thus a definitive factor in drilling decision making process. For complex machine learning tasks like facies classification, feature engineering is often critical. This paper shows the inclusion of physics-motivated feature interaction in feature augmentation can further improve the capability of machine learning in rock facies classification. We demonstrate this approach with the SEG 2016 machine learning contest dataset and the top winning algorithms. The improvement is roboust and can be $\sim5\%$ better than current existing best F-1 score, where F-1 is an evaluation metric used to quantify average prediction accuracy.
In recent years, the natural language processing community has moved away from task-specific feature engineering, i.e., researchers discovering ad-hoc feature representations for various tasks, in favor of general-purpose methods that learn the input representation by themselves. However, state-of-the-art approaches to disfluency detection in spontaneous speech transcripts currently still depend on an array of hand-crafted features, and other representations derived from the output of pre-existing systems such as language models or dependency parsers. As an alternative, this paper proposes a simple yet effective model for automatic disfluency detection, called an auto-correlational neural network (ACNN). The model uses a convolutional neural network (CNN) and augments it with a new auto-correlation operator at the lowest layer that can capture the kinds of "rough copy" dependencies that are characteristic of repair disfluencies in speech. In experiments, the ACNN model outperforms the baseline CNN on a disfluency detection task with a 5% increase in f-score, which is close to the previous best result on this task.