Time-varying parameter (TVP) regressions commonly assume that time-variation in the coefficients is determined by a simple stochastic process such as a random walk. While such models are capable of capturing a wide range of dynamic patterns, the true nature of time variation might stem from other sources, or arise from different laws of motion. In this paper, we propose a flexible TVP VAR that assumes the TVPs to depend on a panel of partially latent covariates. The latent part of these covariates differ in their state dynamics and thus capture smoothly evolving or abruptly changing coefficients. To determine which of these covariates are important, and thus to decide on the appropriate state evolution, we introduce Bayesian shrinkage priors to perform model selection. As an empirical application, we forecast the US term structure of interest rates and show that our approach performs well relative to a set of competing models. We then show how the model can be used to explain structural breaks in coefficients related to the US yield curve.
We introduce a new test for a two-sided hypothesis involving a subset of the structural parameter vector in the linear instrumental variables (IVs) model. Guggenberger et al. (2019), GKM19 from now on, introduce a subvector Anderson-Rubin (AR) test with data-dependent critical values that has asymptotic size equal to nominal size for a parameter space that allows for arbitrary strength or weakness of the IVs and has uniformly nonsmaller power than the projected AR test studied in Guggenberger et al. (2012). However, GKM19 imposes the restrictive assumption of conditional homoskedasticity. The main contribution here is to robustify the procedure in GKM19 to arbitrary forms of conditional heteroskedasticity. We first adapt the method in GKM19 to a setup where a certain covariance matrix has an approximate Kronecker product (AKP) structure which nests conditional homoskedasticity. The new test equals this adaption when the data is consistent with AKP structure as decided by a model selection procedure. Otherwise the test equals the AR/AR test in Andrews (2017) that is fully robust to conditional heteroskedasticity but less powerful than the adapted method. We show theoretically that the new test has asymptotic size bounded by the nominal size and document improved power relative to the AR/AR test in a wide array of Monte Carlo simulations when the covariance matrix is not too far from AKP.
Factor and sparse models are two widely used methods to impose a low-dimensional structure in high-dimensions. However, they are seemingly mutually exclusive. We propose a lifting method that combines the merits of these two models in a supervised learning methodology that allows for efficiently exploring all the information in high-dimensional datasets. The method is based on a flexible model for high-dimensional panel data, called factor-augmented regression model with observable and/or latent common factors, as well as idiosyncratic components. This model not only includes both principal component regression and sparse regression as specific models but also significantly weakens the cross-sectional dependence and facilitates model selection and interpretability. The method consists of several steps and a novel test for (partial) covariance structure in high dimensions to infer the remaining cross-section dependence at each step. We develop the theory for the model and demonstrate the validity of the multiplier bootstrap for testing a high-dimensional (partial) covariance structure. The theory is supported by a simulation study and applications.
Many studies have shown that there are regularities in the way human beings make decisions. However, our ability to obtain models that capture such regularities and can accurately predict unobserved decisions is still limited. We tackle this problem in the context of individuals who are given information relative to the evolution of market prices and asked to guess the direction of the market. We use a networks inference approach with stochastic block models (SBM) to find the model and network representation that is most predictive of unobserved decisions. Our results suggest that users mostly use recent information (about the market and about their previous decisions) to guess. Furthermore, the analysis of SBM groups reveals a set of strategies used by players to process information and make decisions that is analogous to behaviors observed in other contexts. Our study provides and example on how to quantitatively explore human behavior strategies by representing decisions as networks and using rigorous inference and model-selection approaches.
We survey the state-of-the-art in offboard diagnostics for vehicles, their occupants, and environments, with particular focus on vibroacoustic approaches. We identify promising application areas including data-driven management for shared mobility and automated fleets, usage-based insurance, and vehicle, occupant, and environmental state and condition monitoring. We close by exploring the particular application of vibroacoustic monitoring to vehicle diagnostics and prognostics and propose the introduction of automated vehicle- and context-specific model selection as a means of improving algorithm performance, e.g. to enable smartphone-resident diagnostics. Towards this vision, four strong-performing, interdependent classifiers are presented as a proof-of-concept for identifying vehicle configuration from acoustic signatures. The described approach may serve as the first step in developing "universal diagnostics," with applicability extending beyond the automotive domain.
Transfer learning across heterogeneous data distributions (a.k.a. domains) and distinct tasks is a more general and challenging problem than conventional transfer learning, where either domains or tasks are assumed to be the same. While neural network based feature transfer is widely used in transfer learning applications, finding the optimal transfer strategy still requires time-consuming experiments and domain knowledge. We propose a transferability metric called Optimal Transport based Conditional Entropy (OTCE), to analytically predict the transfer performance for supervised classification tasks in such cross-domain and cross-task feature transfer settings. Our OTCE score characterizes transferability as a combination of domain difference and task difference, and explicitly evaluates them from data in a unified framework. Specifically, we use optimal transport to estimate domain difference and the optimal coupling between source and target distributions, which is then used to derive the conditional entropy of the target task (task difference). Experiments on the largest cross-domain dataset DomainNet and Office31 demonstrate that OTCE shows an average of 21% gain in the correlation with the ground truth transfer accuracy compared to state-of-the-art methods. We also investigate two applications of the OTCE score including source model selection and multi-source feature fusion.
Optimal performance is critical for decision-making tasks from medicine to autonomous driving, however common performance measures may be too general or too specific. For binary classifiers, diagnostic tests or prognosis at a timepoint, measures such as the area under the receiver operating characteristic curve, or the area under the precision recall curve, are too general because they include unrealistic decision thresholds. On the other hand, measures such as accuracy, sensitivity or the F1 score are measures at a single threshold that reflect an individual single probability or predicted risk, rather than a range of individuals or risk. We propose a method in between, deep ROC analysis, that examines groups of probabilities or predicted risks for more insightful analysis. We translate esoteric measures into familiar terms: AUC and the normalized concordant partial AUC are balanced average accuracy (a new finding); the normalized partial AUC is average sensitivity; and the normalized horizontal partial AUC is average specificity. Along with post-test measures, we provide a method that can improve model selection in some cases and provide interpretation and assurance for patients in each risk group. We demonstrate deep ROC analysis in two case studies and provide a toolkit in Python.
Thermal issue is of great importance during layout design of heat source components in systems engineering, especially for high functional-density products. Thermal analysis generally needs complex simulation, which leads to an unaffordable computational burden to layout optimization as it iteratively evaluates different schemes. Surrogate modeling is an effective way to alleviate computation complexity. However, temperature field prediction (TFP) with complex heat source layout (HSL) input is an ultra-high dimensional nonlinear regression problem, which brings great difficulty to traditional regression models. The Deep neural network (DNN) regression method is a feasible way for its good approximation performance. However, it faces great challenges in both data preparation for sample diversity and uniformity in the layout space with physical constraints, and proper DNN model selection and training for good generality, which necessitates efforts of both layout designer and DNN experts. To advance this cross-domain research, this paper proposes a DNN based HSL-TFP surrogate modeling task benchmark. With consideration for engineering applicability, sample generation, dataset evaluation, DNN model, and surrogate performance metrics, are thoroughly studied. Experiments are conducted with ten representative state-of-the-art DNN models. Detailed discussion on baseline results is provided and future prospects are analyzed for DNN based HSL-TFP tasks.
Learning curves provide insight into the dependence of a learner's generalization performance on the training set size. This important tool can be used for model selection, to predict the effect of more training data, and to reduce the computational complexity of model training and hyperparameter tuning. This review recounts the origins of the term, provides a formal definition of the learning curve, and briefly covers basics such as its estimation. Our main contribution is a comprehensive overview of the literature regarding the shape of learning curves. We discuss empirical and theoretical evidence that supports well-behaved curves that often have the shape of a power law or an exponential. We consider the learning curves of Gaussian processes, the complex shapes they can display, and the factors influencing them. We draw specific attention to examples of learning curves that are ill-behaved, showing worse learning performance with more training data. To wrap up, we point out various open problems that warrant deeper empirical and theoretical investigation. All in all, our review underscores that learning curves are surprisingly diverse and no universal model can be identified.
Distributionally robust optimization (DRO) provides a framework for training machine learning models that are able to perform well on a collection of related data distributions (the "uncertainty set"). This is done by solving a min-max game: the model is trained to minimize its maximum expected loss among all distributions in the uncertainty set. While careful design of the uncertainty set is critical to the success of the DRO procedure, previous work has been limited to relatively simple alternatives that keep the min-max optimization problem exactly tractable, such as $f$-divergence balls. In this paper, we argue instead for the use of neural generative models to characterize the worst-case distribution, allowing for more flexible and problem-specific selection of the uncertainty set. However, while simple conceptually, this approach poses a number of implementation and optimization challenges. To circumvent these issues, we propose a relaxation of the KL-constrained inner maximization objective that makes the DRO problem more amenable to gradient-based optimization of large scale generative models, and develop model selection heuristics to guide hyper-parameter search. On both toy settings and realistic NLP tasks, we find that the proposed approach yields models that are more robust than comparable baselines.
This paper provides a report on our solution including model selection, tuning strategy and results obtained for Global Road Damage Detection Challenge. This Big Data Cup Challenge was held as a part of IEEE International Conference on Big Data 2020. We assess single and multi-stage network architectures for object detection and provide a benchmark using popular state-of-the-art open-source PyTorch frameworks like Detectron2 and Yolov5. Data preparation for provided Road Damage training dataset, captured using smartphone camera from Czech, India and Japan is discussed. We studied the effect of training on a per country basis with respect to a single generalizable model. We briefly describe the tuning strategy for the experiments conducted on two-stage Faster R-CNN with Deep Residual Network (Resnet) and Feature Pyramid Network (FPN) backbone. Additionally, we compare this to a one-stage Yolov5 model with Cross Stage Partial Network (CSPNet) backbone. We show a mean F1 score of 0.542 on Test2 and 0.536 on Test1 datasets using a multi-stage Faster R-CNN model, with Resnet-50 and Resnet-101 backbones respectively. This shows the generalizability of the Resnet-50 model when compared to its more complex counterparts. Experiments were conducted using Google Colab having K80 and a Linux PC with 1080Ti, NVIDIA consumer grade GPU. A PyTorch based Detectron2 code to preprocess, train, test and submit the Avg F1 score to is made available at https://github.com/vishwakarmarhl/rdd2020
Time series data have grown at an explosive rate in numerous domains and have stimulated a surge of time series modeling research. A comprehensive comparison of different time series models, for a considered data analytics task, provides useful guidance on model selection for data analytics practitioners. Data scarcity is a universal issue that occurs in a vast range of data analytics problems, due to the high costs associated with collecting, generating, and labeling data as well as some data quality issues such as missing data. In this paper, we focus on the temporal classification/regression problem that attempts to build a mathematical mapping from multivariate time series inputs to a discrete class label or a real-valued response variable. For this specific problem, we identify two types of scarce data: scarce data with small samples and scarce data with sparsely and irregularly observed time series covariates. Observing that all existing works are incapable of utilizing the sparse time series inputs for proper modeling building, we propose a model called sparse functional multilayer perceptron (SFMLP) for handling the sparsity in the time series covariates. The effectiveness of the proposed SFMLP under each of the two types of data scarcity, in comparison with the conventional deep sequential learning models (e.g., Recurrent Neural Network, and Long Short-Term Memory), is investigated through mathematical arguments and numerical experiments.
We study the problem of word-level confidence estimation in subword-based end-to-end (E2E) models for automatic speech recognition (ASR). Although prior works have proposed training auxiliary confidence models for ASR systems, they do not extend naturally to systems that operate on word-pieces (WP) as their vocabulary. In particular, ground truth WP correctness labels are needed for training confidence models, but the non-unique tokenization from word to WP causes inaccurate labels to be generated. This paper proposes and studies two confidence models of increasing complexity to solve this problem. The final model uses self-attention to directly learn word-level confidence without needing subword tokenization, and exploits full context features from multiple hypotheses to improve confidence accuracy. Experiments on Voice Search and long-tail test sets show standard metrics (e.g., NCE, AUC, RMSE) improving substantially. The proposed confidence module also enables a model selection approach to combine an on-device E2E model with a hybrid model on the server to address the rare word recognition problem for the E2E model.
In comparison to classical shallow representation learning techniques, deep neural networks have achieved superior performance in nearly every application benchmark. But despite their clear empirical advantages, it is still not well understood what makes them so effective. To approach this question, we introduce deep frame approximation: a unifying framework for constrained representation learning with structured overcomplete frames. While exact inference requires iterative optimization, it may be approximated by the operations of a feed-forward deep neural network. We indirectly analyze how model capacity relates to frame structures induced by architectural hyperparameters such as depth, width, and skip connections. We quantify these structural differences with the deep frame potential, a data-independent measure of coherence linked to representation uniqueness and stability. As a criterion for model selection, we show correlation with generalization error on a variety of common deep network architectures and datasets. We also demonstrate how recurrent networks implementing iterative optimization algorithms can achieve performance comparable to their feed-forward approximations while improving adversarial robustness. This connection to the established theory of overcomplete representations suggests promising new directions for principled deep network architecture design with less reliance on ad-hoc engineering.
Model complexity is a fundamental problem in deep learning. In this paper we conduct a systematic overview of the latest studies on model complexity in deep learning. Model complexity of deep learning can be categorized into expressive capacity and effective model complexity. We review the existing studies on those two categories along four important factors, including model framework, model size, optimization process and data complexity. We also discuss the applications of deep learning model complexity including understanding model generalization, model optimization, and model selection and design. We conclude by proposing several interesting future directions.
The rapid development of remote sensing techniques provides rich, large-coverage, and high-temporal information of the ground, which can be coupled with the emerging deep learning approaches that enable latent features and hidden geographical patterns to be extracted. This study marks the first attempt to cross-compare performances of popular state-of-the-art deep learning models in estimating population distribution from remote sensing images, investigate the contribution of neighboring effect, and explore the potential systematic population estimation biases. We conduct an end-to-end training of four popular deep learning architectures, i.e., VGG, ResNet, Xception, and DenseNet, by establishing a mapping between Sentinel-2 image patches and their corresponding population count from the LandScan population grid. The results reveal that DenseNet outperforms the other three models, while VGG has the worst performances in all evaluating metrics under all selected neighboring scenarios. As for the neighboring effect, contradicting existing studies, our results suggest that the increase of neighboring sizes leads to reduced population estimation performance, which is found universal for all four selected models in all evaluating metrics. In addition, there exists a notable, universal bias that all selected deep learning models tend to overestimate sparsely populated image patches and underestimate densely populated image patches, regardless of neighboring sizes. The methodological, experimental, and contextual knowledge this study provides is expected to benefit a wide range of future studies that estimate population distribution via remote sensing imagery.
Bayesian priors offer a compact yet general means of incorporating domain knowledge into many learning tasks. The correctness of the Bayesian analysis and inference, however, largely depends on accuracy and correctness of these priors. PAC-Bayesian methods overcome this problem by providing bounds that hold regardless of the correctness of the prior distribution. This paper introduces the first PAC-Bayesian bound for the batch reinforcement learning problem with function approximation. We show how this bound can be used to perform model-selection in a transfer learning scenario. Our empirical results confirm that PAC-Bayesian policy evaluation is able to leverage prior distributions when they are informative and, unlike standard Bayesian RL approaches, ignore them when they are misleading.
Time series subject to change in regime have attracted much interest in domains such as econometry, finance or meteorology. For discrete-valued regimes, some models such as the popular Hidden Markov Chain (HMC) describe time series whose state process is unknown at all time-steps. Sometimes, time series are firstly labelled thanks to some annotation function. Thus, another category of models handles the case with regimes observed at all time-steps. We present a novel model which addresses the intermediate case: (i) state processes associated to such time series are modelled by Partially Hidden Markov Chains (PHMCs); (ii) a linear autoregressive (LAR) model drives the dynamics of the time series, within each regime. We describe a variant of the expection maximization (EM) algorithm devoted to PHMC-LAR model learning. We propose a hidden state inference procedure and a forecasting function that take into account the observed states when existing. We assess inference and prediction performances, and analyze EM convergence times for the new model, using simulated data. We show the benefits of using partially observed states to decrease EM convergence times. A fully labelled scheme with unreliable labels also speeds up EM. This offers promising prospects to enhance PHMC-LAR model selection. We also point out the robustness of PHMC-LAR to labelling errors in inference task, when large training datasets and moderate labelling error rates are considered. Finally, we highlight the remarkable robustness to error labelling in the prediction task, over the whole range of error rates.
A large number of time series forecasting models including traditional statistical models, machine learning models and more recently deep learning have been proposed in the literature. However, choosing the right model along with good parameter values that performs well on a given data is still challenging. Automatically providing a good set of models to users for a given dataset saves both time and effort from using trial-and-error approaches with a wide variety of available models along with parameter optimization. We present AutoAI for Time Series Forecasting (AutoAI-TS) that provides users with a zero configuration (zero-conf ) system to efficiently train, optimize and choose best forecasting model among various classes of models for the given dataset. With its flexible zero-conf design, AutoAI-TS automatically performs all the data preparation, model creation, parameter optimization, training and model selection for users and provides a trained model that is ready to use. For given data, AutoAI-TS utilizes a wide variety of models including classical statistical models, Machine Learning (ML) models, statistical-ML hybrid models and deep learning models along with various transformations to create forecasting pipelines. It then evaluates and ranks pipelines using the proposed T-Daub mechanism to choose the best pipeline. The paper describe in detail all the technical aspects of AutoAI-TS along with extensive benchmarking on a variety of real world data sets for various use-cases. Benchmark results show that AutoAI-TS, with no manual configuration from the user, automatically trains and selects pipelines that on average outperform existing state-of-the-art time series forecasting toolkits.
We categorize meta-learning evaluation into two settings: $\textit{in-distribution}$ [ID], in which the train and test tasks are sampled $\textit{iid}$ from the same underlying task distribution, and $\textit{out-of-distribution}$ [OOD], in which they are not. While most meta-learning theory and some FSL applications follow the ID setting, we identify that most existing few-shot classification benchmarks instead reflect OOD evaluation, as they use disjoint sets of train (base) and test (novel) classes for task generation. This discrepancy is problematic because -- as we show on numerous benchmarks -- meta-learning methods that perform better on existing OOD datasets may perform significantly worse in the ID setting. In addition, in the OOD setting, even though current FSL benchmarks seem befitting, our study highlights concerns in 1) reliably performing model selection for a given meta-learning method, and 2) consistently comparing the performance of different methods. To address these concerns, we provide suggestions on how to construct FSL benchmarks to allow for ID evaluation as well as more reliable OOD evaluation. Our work aims to inform the meta-learning community about the importance and distinction of ID vs. OOD evaluation, as well as the subtleties of OOD evaluation with current benchmarks.