We present a method of discovering governing differential equations from data without the need to specify a priori the terms to appear in the equation. The input to our method is a dataset (or ensemble of datasets) corresponding to a particular solution (or ensemble of particular solutions) of a differential equation. The output is a human-readable differential equation with parameters calibrated to the individual particular solutions provided. The key to our method is to learn differentiable models of the data that subsequently serve as inputs to a genetic programming algorithm in which graphs specify computation over arbitrary compositions of functions, parameters, and (potentially differential) operators on functions. Differential operators are composed and evaluated using recursive application of automatic differentiation, allowing our algorithm to explore arbitrary compositions of operators without the need for human intervention. We also demonstrate an active learning process to identify and remedy deficiencies in the proposed governing equations.
Neural Machine Translation (NMT) models have been proved strong when translating clean texts, but they are very sensitive to noise in the input. Improving NMT models robustness can be seen as a form of "domain" adaption to noise. The recently created Machine Translation on Noisy Text task corpus provides noisy-clean parallel data for a few language pairs, but this data is very limited in size and diversity. The state-of-the-art approaches are heavily dependent on large volumes of back-translated data. This paper has two main contributions: Firstly, we propose new data augmentation methods to extend limited noisy data and further improve NMT robustness to noise while keeping the models small. Secondly, we explore the effect of utilizing noise from external data in the form of speech transcripts and show that it could help robustness.
Word embeddings are an essential component in a wide range of natural language processing applications. However, distributional semantic models are known to struggle when only a small number of context sentences are available. Several methods have been proposed to obtain higher-quality vectors for these words, leveraging both this context information and sometimes the word forms themselves through a hybrid approach. We show that the current tasks do not suffice to evaluate models that use word-form information, as such models can easily leverage word forms in the training data that are related to word forms in the test data. We introduce 3 new tasks, allowing for a more balanced comparison between models. Furthermore, we show that hyperparameters that have largely been ignored in previous work can consistently improve the performance of both baseline and advanced models, achieving a new state of the art on 4 out of 6 tasks.
This paper tackles the problem of video object segmentation. We are specifically concerned with the task of segmenting all pixels of a target object in all frames, given the annotation mask in the first frame. Even when such annotation is available this remains a challenging problem because of the changing appearance and shape of the object over time. In this paper, we tackle this task by formulating it as a meta-learning problem, where the base learner grasping the semantic scene understanding for a general type of objects, and the meta learner quickly adapting the appearance of the target object with a few examples. Our proposed meta-learning method uses a closed form optimizer, the so-called "ridge regression", which has been shown to be conducive for fast and better training convergence. Moreover, we propose a mechanism, named "block splitting", to further speed up the training process as well as to reduce the number of learning parameters. In comparison with the-state-of-the art methods, our proposed framework achieves significant boost up in processing speed, while having very competitive performance compared to the best performing methods on the widely used datasets.
The dependency of the generalization error of neural networks on model and dataset size is of critical importance both in practice and for understanding the theory of neural networks. Nevertheless, the functional form of this dependency remains elusive. In this work, we present a functional form which approximates well the generalization error in practice. Capitalizing on the successful concept of model scaling (e.g., width, depth), we are able to simultaneously construct such a form and specify the exact models which can attain it across model/data scales. Our construction follows insights obtained from observations conducted over a range of model/data scales, in various model types and datasets, in vision and language tasks. We show that the form both fits the observations well across scales, and provides accurate predictions from small- to large-scale models and data.
Uncertainty analysis in the form of probabilistic forecasting can significantly improve decision making processes in the smart power grid when integrating renewable energy sources such as wind. Whereas point forecasting provides a single expected value, probabilistic forecasts provide more information in the form of quantiles, prediction intervals, or full predictive densities. Traditionally quantile regression is applied for such forecasting and recently quantile regression neural networks have become popular for weather and renewable energy forecasting. However, one major shortcoming of composite quantile estimation in neural networks is the quantile crossover problem. This paper analyzes the effectiveness of a novel smoothed loss and penalty function for neural network architectures to prevent the quantile crossover problem. Its efficacy is examined on the wind power forecasting problem. A numerical case study is conducted using publicly available wind data from the Global Energy Forecasting Competition 2014. Multiple quantiles are estimated to form 10\%, to 90\% prediction intervals which are evaluated using a quantile score and reliability measures. Benchmark models such as the persistence and climatology distributions, multiple quantile regression, and support vector quantile regression are used for comparison where results demonstrate the proposed approach leads to improved performance while preventing the problem of overlapping quantile estimates.
Chronic Obstructive Pulmonary Disease (COPD) is a leading cause of morbidity and mortality. While COPD diagnosis is based on lung function tests, early stages and progression of different aspects of the disease can be visible and quantitatively assessed on computed tomography (CT) scans. Many studies have been published that quantify imaging biomarkers related to COPD. In this paper we present a convolutional neural network that directly computes visual emphysema scores and predicts the outcome of lung function tests for 195 CT scans from the COPDGene study. Contrary to previous work, the proposed method does not encode any specific prior knowledge about what to quantify, but it is trained end-to-end with a set of 1424 CT scans for which the output parameters were available. The network provided state-of-the-art results for these tasks: Visual emphysema scores are comparable to those assessed by trained human observers; COPD diagnosis from estimated lung function reaches an area under the ROC curve of 0.94, outperforming prior art. The method is easily generalizable to other situations where information from whole scans needs to be summarized in single quantities.
Neural models for abstractive summarization tend to achieve the best performance in the presence of highly specialized, summarization specific modeling add-ons such as pointer-generator, coverage-modeling, and inferencetime heuristics. We show here that pretraining can complement such modeling advancements to yield improved results in both short-form and long-form abstractive summarization using two key concepts: full-network initialization and multi-stage pretraining. Our method allows the model to transitively benefit from multiple pretraining tasks, from generic language tasks to a specialized summarization task to an even more specialized one such as bullet-based summarization. Using this approach, we demonstrate improvements of 1.05 ROUGE-L points on the Gigaword benchmark and 1.78 ROUGE-L points on the CNN/DailyMail benchmark, compared to a randomly-initialized baseline.
Class activation map (CAM) highlights regions of classes based on classification network, which is widely used in weakly supervised tasks. However, it faces the problem that the class activation regions are usually small and local. Although several efforts paid to the second step (the CAM generation step) have partially enhanced the generation, we believe such problem is also caused by the first step (training step), because single classification model trained on the entire classes contains finite discriminate information that limits the object region extraction. To this end, this paper solves CAM generation by using multiple classification models. To form multiple classification networks that carry different discriminative information, we try to capture the semantic relationships between classes to form different semantic levels of classification models. Specifically, hierarchical clustering based on class relationships is used to form hierarchical clustering results, where the clustering levels are treated as semantic levels to form the classification models. Moreover, a new orthogonal module and a two-branch based CAM generation method are proposed to generate class regions that are orthogonal and complementary. We use the PASCAL VOC 2012 dataset to verify the proposed method. Experimental results show that our approach improves the CAM generation.
In tabular case, when the reward and environment dynamics are known, policy evaluation can be written as $\bm{V}_{\bm{\pi}} = (I - \gamma P_{\bm{\pi}})^{-1} \bm{r}_{\bm{\pi}}$, where $P_{\bm{\pi}}$ is the state transition matrix given policy ${\bm{\pi}}$ and $\bm{r}_{\bm{\pi}}$ is the reward signal given ${\bm{\pi}}$. What annoys us is that $P_{\bm{\pi}}$ and $\bm{r}_{\bm{\pi}}$ are both mixed with ${\bm{\pi}}$, which means every time when we update ${\bm{\pi}}$, they will change together. In this paper, we leverage the notation from \cite{wang2007dual} to disentangle ${\bm{\pi}}$ and environment dynamics which makes optimization over policy more straightforward. We show that policy gradient theorem \cite{sutton2018reinforcement} and TRPO \cite{schulman2015trust} can be put into a more general framework and such notation has good potential to be extended to model-based reinforcement learning.
The Efficient Market Hypothesis has been a staple of economics research for decades. In particular, weak-form market efficiency -- the notion that past prices cannot predict future performance -- is strongly supported by econometric evidence. In contrast, machine learning algorithms implemented to predict stock price have been touted, to varying degrees, as successful. Moreover, some data scientists boast the ability to garner above-market returns using price data alone. This study endeavors to connect existing econometric research on weak-form efficient markets with data science innovations in algorithmic trading. First, a traditional exploration of stationarity in stock index prices over the past decade is conducted with Augmented Dickey-Fuller and Variance Ratio tests. Then, an algorithmic trading platform is implemented with the use of five machine learning algorithms. Econometric findings identify potential stationarity, hinting technical evaluation may be possible, though algorithmic trading results find little predictive power in any machine learning model, even when using trend-specific metrics. Accounting for transaction costs and risk, no system achieved above-market returns consistently. Our findings reinforce the validity of weak-form market efficiency.
We shed new insights on the two commonly used updates for the online $k$-PCA problem, namely, Krasulina's and Oja's updates. We show that Krasulina's update corresponds to a projected gradient descent step on the Stiefel manifold of the orthonormal $k$-frames, while Oja's update amounts to a gradient descent step using the unprojected gradient. Following these observations, we derive a more \emph{implicit} form of Krasulina's $k$-PCA update, i.e. a version that uses the information of the future gradient as much as possible. Most interestingly, our implicit Krasulina update avoids the costly QR-decomposition step by bypassing the orthonormality constraint. We show that the new update in fact corresponds to an online EM step applied to a probabilistic $k$-PCA model. The probabilistic view of the updates allows us to combine multiple models in a distributed setting. We show experimentally that the implicit Krasulina update yields superior convergence while being significantly faster. We also give strong evidence that the new update can benefit from parallelism and is more stable w.r.t. tuning of the learning rate.
Text-to-speech systems are typically evaluated on single sentences. When long-form content, such as data consisting of full paragraphs or dialogues is considered, evaluating sentences in isolation is not always appropriate as the context in which the sentences are synthesized is missing. In this paper, we investigate three different ways of evaluating the naturalness of long-form text-to-speech synthesis. We compare the results obtained from evaluating sentences in isolation, evaluating whole paragraphs of speech, and presenting a selection of speech or text as context and evaluating the subsequent speech. We find that, even though these three evaluations are based upon the same material, the outcomes differ per setting, and moreover that these outcomes do not necessarily correlate with each other. We show that our findings are consistent between a single speaker setting of read paragraphs and a two-speaker dialogue scenario. We conclude that to evaluate the quality of long-form speech, the traditional way of evaluating sentences in isolation does not suffice, and that multiple evaluations are required.
The paper presents a new method for approximating Strong Stackelberg Equilibrium in general-sum sequential games with imperfect information and perfect recall. The proposed approach is generic as it does not rely on any specific properties of a particular game model. The method is based on iterative interleaving of the two following phases: (1) guided Monte Carlo Tree Search sampling of the Follower's strategy space and (2) building the Leader's behavior strategy tree for which the sampled Follower's strategy is an optimal response. The above solution scheme is evaluated with respect to expected Leader's utility and time requirements on three sets of interception games with variable characteristics, played on graphs. A comparison with three state-of-the-art MILP/LP-based methods shows that in vast majority of test cases proposed simulation-based approach leads to optimal Leader's strategies, while excelling the competitive methods in terms of better time scalability and lower memory requirements.
Automatic, template-free extraction of information from form images is challenging due to the variety of form layouts. This is even more challenging for historical forms due to noise and degradation. A crucial part of the extraction process is associating input text with pre-printed labels. We present a learned, template-free solution to detecting pre-printed text and input text/handwriting and predicting pair-wise relationships between them. While previous approaches to this problem have been focused on clean images and clear layouts, we show our approach is effective in the domain of noisy, degraded, and varied form images. We introduce a new dataset of historical form images (late 1800s, early 1900s) for training and validating our approach. Our method uses a convolutional network to detect pre-printed text and input text lines. We pool features from the detection network to classify possible relationships in a language-agnostic way. We show that our proposed pairing method outperforms heuristic rules and that visual features are critical to obtaining high accuracy.
Research on graph representation learning has received a lot of attention in recent years since many data in real-world applications come in form of graphs. High-dimensional graph data are often in irregular form, which makes them more difficult to analyze than image/video/audio data defined on regular lattices. Various graph embedding techniques have been developed to convert the raw graph data into a low-dimensional vector representation while preserving the intrinsic graph properties. In this review, we first explain the graph embedding task and its challenges. Next, we review a wide range of graph embedding techniques with insights. Then, we evaluate several state-of-the-art methods against small and large datasets and compare their performance. Finally, potential applications and future directions are presented.
Coarse correlation models strategic interactions of rational agents complemented by a correlation device, that is a mediator that can recommend behavior but not enforce it. Despite being a classical concept in the theory of normal-form games for more than forty years, not much is known about the merits of coarse correlation in extensive-form settings. In this paper, we consider two instantiations of the idea of coarse correlation in extensive-form games: normal-form coarse-correlated equilibrium (NFCCE), already defined in the literature, and extensive-form coarse-correlated equilibrium (EFCCE), which we introduce for the first time. We show that EFCCE is a subset of NFCCE and a superset of the related extensive-form correlated equilibrium. We also show that, in two-player extensive-form games, social-welfare-maximizing EFCCEs and NFCEEs are bilinear saddle points, and give new efficient algorithms for the special case of games with no chance moves. In our experiments, our proposed algorithm for NFCCE is two to four orders of magnitude faster than the prior state of the art.
We propose Deep Closed-Form Subspace Clustering (DCFSC), a new embarrassingly simple model for subspace clustering with learning non-linear mapping. Compared with the previous deep subspace clustering (DSC) techniques, our DCFSC does not have any parameters at all for the self-expressive layer. Instead, DCFSC utilizes the implicit data-driven self-expressive layer derived from closed-form shallow auto-encoder. Moreover, DCFSC also has no complicated optimization scheme, unlike the other subspace clustering methods. With its extreme simplicity, DCFSC has significant memory-related benefits over the existing DSC method, especially on the large dataset. Several experiments showed that our DCFSC model had enough potential to be a new reference model for subspace clustering on large-scale high-dimensional dataset.
We propose a novel classifier accuracy metric: the Bayesian Area Under the Receiver Operating Characteristic Curve (CBAUC). The method estimates the area under the ROC curve and is related to the recently proposed Bayesian Error Estimator. The metric can assess the quality of a classifier using only the training dataset without the need for computationally expensive cross-validation. We derive a closed-form solution of the proposed accuracy metric for any linear binary classifier under the Gaussianity assumption, and study the accuracy of the proposed estimator using simulated and real-world data. These experiments confirm that the closed-form CBAUC is both faster and more accurate than conventional AUC estimators.
Recent work on vector-based compositional natural language semantics has proposed the use of density matrices to model lexical ambiguity and (graded) entailment (e.g. Piedeleu et al 2015, Bankova et al 2019, Sadrzadeh et al 2018). Ambiguous word meanings, in this work, are represented as mixed states, and the compositional interpretation of phrases out of their constituent parts takes the form of a strongly monoidal functor sending the derivational morphisms of a pregroup syntax to linear maps in FdHilb. Our aims in this paper are threefold. Firstly, we replace the pregroup front end by a Lambek categorial grammar with directional implications expressing a word's selectional requirements. By the Curry-Howard correspondence, the derivations of the grammar's type logic are associated with terms of the (ordered) linear lambda calculus; these terms can be read as programs for compositional meaning assembly with density matrices as the target semantic spaces. Secondly, we extend on the existing literature and introduce a symmetric, nondegenerate bilinear form called a "metric" that defines a canonical isomorphism between a vector space and its dual, allowing us to keep a distinction between left and right implication. Thirdly, we use this metric to define density matrix spaces in a directional form, modeling the ubiquitous derivational ambiguity of natural language syntax, and show how this alows an integrated treatment of lexical and derivational forms of ambiguity controlled at the level of the interpretation.