In this paper, we target at recovering the exact routes taken by commuters inside a metro system that arenot captured by an Automated Fare Collection (AFC) system and hence remain unknown. We strategicallypropose two inference tasks to handle the recovering, one to infer the travel time of each travel link thatcontributes to the total duration of any trip inside a metro network and the other to infer the route preferencesbased on historical trip records and the travel time of each travel link inferred in the previous inferencetask. As these two inference tasks have interrelationship, most of existing works perform these two taskssimultaneously. However, our solutionTripDecoderadopts a totally different approach. To the best of ourknowledge,TripDecoderis the first model that points out and fully utilizes the fact that there are some tripsinside a metro system with only one practical route available. It strategically decouples these two inferencetasks by only taking those trip records with only one practical route as the input for the first inference taskof travel time and feeding the inferred travel time to the second inference task as an additional input whichnot only improves the accuracy but also effectively reduces the complexity of both inference tasks. Twocase studies have been performed based on the city-scale real trip records captured by the AFC systems inSingapore and Taipei to compare the accuracy and efficiency ofTripDecoderand its competitors. As expected,TripDecoderhas achieved the best accuracy in both datasets, and it also demonstrates its superior efficiencyand scalability.

The fragility of deep neural networks to adversarially-chosen inputs has motivated the need to revisit deep learning algorithms. Including adversarial examples during training is a popular defense mechanism against adversarial attacks. This mechanism can be formulated as a min-max optimization problem, where the adversary seeks to maximize the loss function using an iterative first-order algorithm while the learner attempts to minimize it. However, finding adversarial examples in this way causes excessive computational overhead during training. By interpreting the min-max problem as an optimal control problem, it has recently been shown that one can exploit the compositional structure of neural networks in the optimization problem to improve the training time significantly. In this paper, we provide the first convergence analysis of this adversarial training algorithm by combining techniques from robust optimal control and inexact oracle methods in optimization. Our analysis sheds light on how the hyperparameters of the algorithm affect the its stability and convergence. We support our insights with experiments on a robust classification problem.

Data sparsity is a common issue to train machine learning tools such as neural networks for engineering and scientific applications, where experiments and simulations are expensive. Recently physics-constrained neural networks (PCNNs) were developed to reduce the required amount of training data. However, the weights of different losses from data and physical constraints are adjusted empirically in PCNNs. In this paper, a new physics-constrained neural network with the minimax architecture (PCNNMM) is proposed so that the weights of different losses can be adjusted systematically. The training of the PCNN-MM is searching the high-order saddle points of the objective function. A novel saddle point search algorithm called Dual-Dimer method is developed. It is demonstrated that the Dual-Dimer method is computationally more efficient than the gradient descent ascent method for nonconvex-nonconcave functions and provides additional eigenvalue information to verify search results. A heat transfer example also shows that the convergence of PCNN-MMs is faster than that of traditional PCNNs.

Multi-label image classification has generated significant interest in recent years and the performance of such systems often suffers from the not so infrequent occurrence of incorrect or missing labels in the training data. In this paper, we extend the state-of the-art of training classifiers to jointly deal with both forms of errorful data. We accomplish this by modeling noisy and missing labels in multi-label images with a new Noise Modeling Network (NMN) that follows our convolutional neural network (CNN), integrates with it, forming an end-to-end deep learning system, which can jointly learn the noise distribution and CNN parameters. The NMN learns the distribution of noise patterns directly from the noisy data without the need for any clean training data. The NMN can model label noise that depends only on the true label or is also dependent on the image features. We show that the integrated NMN/CNN learning system consistently improves the classification performance, for different levels of label noise, on the MSR-COCO dataset and MSR-VTT dataset. We also show that noise performance improvements are obtained when multiple instance learning methods are used.

This paper proposes a simple method to extract from a set of multiple related time series a compressed representation for each time series based on statistics for the entire set of all time series. This is achieved by a hierarchical algorithm that first generates an alphabet of shapelets based on the segmentation of centroids for clustered data, before labels of these shapelets are assigned to the segmentation of each single time series via nearest neighbor search using unconstrained dynamic time warping as distance measure to deal with non-uniform time series lenghts. Thereby, a sequence of labels is assigned for each time series. Completion of the last label sequence permits prediction of individual time series. Proposed method is evaluated on two global COVID-19 datasets, first, for the number of daily net cases (daily new infections minus daily recoveries), and, second, for the number of daily deaths attributed to COVID-19 as of April 27, 2020. The first dataset involves 249 time series for different countries, each of length 96. The second dataset involves 264 time series, each of length 96. Based on detected anomalies in available data a decentralized exit strategy from lockdowns is advocated.

One major obstacle that precludes the success of reinforcement learning in real-world applications is the lack of robustness, either to model uncertainties or external disturbances, of the trained policies. Robustness is critical when the policies are trained in simulations instead of real world environment. In this work, we propose a risk-aware algorithm to learn robust policies in order to bridge the gap between simulation training and real-world implementation. Our algorithm is based on recently discovered distributional RL framework. We incorporate CVaR risk measure in sample based distributional policy gradients (SDPG) for learning risk-averse policies to achieve robustness against a range of system disturbances.

Ensembling is a simple and popular technique for boosting evaluation performance by training multiple models (e.g., with different initializations) and aggregating their predictions. This approach is commonly reserved for the largest models, as it is commonly held that increasing the model size provides a more substantial reduction in error than ensembling smaller models. However, we show results from experiments on CIFAR-10 and ImageNet that ensembles can outperform single models with both higher accuracy and requiring fewer total FLOPs to compute, even when those individual models' weights and hyperparameters are highly optimized. Furthermore, this gap in improvement widens as models become large. This presents an interesting observation that output diversity in ensembling can often be more efficient than training larger models, especially when the models approach the size of what their dataset can foster. Instead of using the common practice of tuning a single large model, one can use ensembles as a more flexible trade-off between a model's inference speed and accuracy. This also potentially eases hardware design, e.g., an easier way to parallelize the model across multiple workers for real-time or distributed inference.

Many measurements and searches for new phenomena performed by the experiments at the Large Hadron Collider (LHC) use a classification algorithm, such as Boosted Decision Trees or Neural Networks, to discriminate the physics process of interest (signal) from other physics processes with similar signature (background). The algorithms are optimized using supervised training on detailed simulated Monte Carlo (MC) data sets, labeled as signal or background. The resulting classifier is applied to unlabeled data to separate signal and background, and measure the statistical significance of the signal or its strength, assuming that the simulated and the real data sets are identically distributed. However, differences between real and simulated data sets always exist and the learner may pick up a discriminating feature which differs between the data sets, introducing a bias to the sample used for training. This problem is similar to that of visual recognition where training is performed on simulated pictures, the so-called source domain and applied to real photographs, the target domain. In order to avoid training specific to the source domain, algorithms of domain adaptation have been developed. In this paper, we apply the method of domain adaptation to high energy physics data. In this paper we present a Domain Adversarial Neural Network (DANN) to classify events in the search for the t tH(H b b) process at the LHC, which is very rare and hard to separate from the t t jets background [1].

While typical named entity recognition (NER) models require the training set to be annotated with all target types, each available datasets may only cover a part of them. Instead of relying on fully-typed NER datasets, many efforts have been made to leverage multiple partially-typed ones for training and allow the resulting model to cover a full type set. However, there is neither guarantee on the quality of integrated datasets, nor guidance on the design of training algorithms. Here, we conduct a systematic analysis and comparison between partially-typed NER datasets and fully-typed ones, in both theoretical and empirical manner. Firstly, we derive a bound to establish that models trained with partially-typed annotations can reach a similar performance with the ones trained with fully-typed annotations, which also provides guidance on the algorithm design. Moreover, we conduct controlled experiments, which shows partially-typed datasets leads to similar performance with the model trained with the same amount of fully-typed annotations.

In recent years, the concept of automated machine learning has become very popular. Automated Machine Learning (AutoML) mainly refers to the automated methods for model selection and hyper-parameter optimization of various algorithms such as random forests, gradient boosting, neural networks, etc. In this paper, we introduce a new package i.e. DriveML for automated machine learning. DriveML helps in implementing some of the pillars of an automated machine learning pipeline such as automated data preparation, feature engineering, model building and model explanation by running the function instead of writing lengthy R codes. The DriveML package is available in CRAN. We compare the DriveML package with other relevant packages in CRAN/Github and find that DriveML performs the best across different parameters. We also provide an illustration by applying the DriveML package with default configuration on a real world dataset. Overall, the main benefits of DriveML are in development time savings, reduce developer's errors, optimal tuning of machine learning models and reproducibility.