The effectiveness of recurrent neural networks can be largely influenced by their ability to store into their dynamical memory information extracted from input sequences at different frequencies and timescales. Such a feature can be introduced into a neural architecture by an appropriate modularization of the dynamic memory. In this paper we propose a novel incrementally trained recurrent architecture targeting explicitly multi-scale learning. First, we show how to extend the architecture of a simple RNN by separating its hidden state into different modules, each subsampling the network hidden activations at different frequencies. Then, we discuss a training algorithm where new modules are iteratively added to the model to learn progressively longer dependencies. Each new module works at a slower frequency than the previous ones and it is initialized to encode the subsampled sequence of hidden activations. Experimental results on synthetic and real-world datasets on speech recognition and handwritten characters show that the modular architecture and the incremental training algorithm improve the ability of recurrent neural networks to capture long-term dependencies.

Interest in stochastic zeroth-order (SZO) methods has recently been revived in black-box optimization scenarios such as adversarial black-box attacks to deep neural networks. SZO methods only require the ability to evaluate the objective function at random input points, however, their weakness is the dependency of their convergence speed on the dimensionality of the function to be evaluated. We present a sparse SZO optimization method that reduces this factor to the expected dimensionality of the random perturbation during learning. We give a proof that justifies this reduction for sparse SZO optimization for non-convex functions without making any assumptions on sparsity of objective function or gradient. Furthermore, we present experimental results for neural networks on MNIST and CIFAR that show faster convergence in training loss and test accuracy, and a smaller distance of the gradient approximation to the true gradient in sparse SZO compared to dense SZO.

This paper describes the hotel price aggregation system - PriceAggregator, deployed at Agoda, a global online travel agency for hotels, vacation rentals, flights and airport transfer. Agoda aggregates non-direct suppliers' hotel rooms to ensure that Agoda's customers always have the widest selection of hotels, room types and packages. As of today, Agoda aggregates millions of hotels. The major challenge is that each supplier only allows Agoda to fetch for the hotel price with a limited amount of Queries Per Second (QPS). Due to the sheer volume of Agoda's user search traffic, this limited amount of QPS is never enough to cover all user searches. Inevitably, many user searches have to be ignored. Hence, booking lost. To overcome the challenge, we built PriceAggregator. PriceAggregator intelligently determines when, how and what to send to the suppliers to fetch for price. In this paper, we not only prove PriceAggregator is optimal theoretically but also demonstrate that PriceAggregator performs well in practice. PriceAggregator has been deployed in Agoda. Extensive online A/B experimentation have shown that PriceAggregator increases Agoda's bookings significantly.

Deep learning models achieve high predictive accuracy across a broad spectrum of tasks, but rigorously quantifying their predictive uncertainty remains challenging. Usable estimates of predictive uncertainty should (1) cover the true prediction targets with high probability, and (2) discriminate between high- and low-confidence prediction instances. Existing methods for uncertainty quantification are based predominantly on Bayesian neural networks; these may fall short of (1) and (2) -- i.e., Bayesian credible intervals do not guarantee frequentist coverage, and approximate posterior inference undermines discriminative accuracy. In this paper, we develop the discriminative jackknife (DJ), a frequentist procedure that utilizes influence functions of a model's loss functional to construct a jackknife (or leave-one-out) estimator of predictive confidence intervals. The DJ satisfies (1) and (2), is applicable to a wide range of deep learning models, is easy to implement, and can be applied in a post-hoc fashion without interfering with model training or compromising its accuracy. Experiments demonstrate that DJ performs competitively compared to existing Bayesian and non-Bayesian regression baselines.

Elastography ultrasound (EUS) provides additional bio-mechanical information about lesion for B-mode ultrasound (BUS) in the diagnosis of breast cancers. However, joint utilization of both BUS and EUS is not popular due to the lack of EUS devices in rural hospitals, which arouses a novel modality imbalance problem in computer-aided diagnosis (CAD) for breast cancers. Current transfer learning (TL) pay little attention to this special issue of clinical modality imbalance, that is, the source domain (EUS modality) has fewer labeled samples than those in the target domain (BUS modality). Moreover, these TL methods cannot fully use the label information to explore the intrinsic relation between two modalities and then guide the promoted knowledge transfer. To this end, we propose a novel doubly supervised TL network (DDSTN) that integrates the Learning Using Privileged Information (LUPI) paradigm and the Maximum Mean Discrepancy (MMD) criterion into a unified deep TL framework. The proposed algorithm can not only make full use of the shared labels to effectively guide knowledge transfer by LUPI paradigm, but also perform additional supervised transfer between unpaired data. We further introduce the MMD criterion to enhance the knowledge transfer. The experimental results on the breast ultrasound dataset indicate that the proposed DDSTN outperforms all the compared state-of-the-art algorithms for the BUSbased CAD. Keywords: Ultrasound imaging, Breast cancer, Deep doubly supervised transfer learning, Support vector machine plus, Maximum mean discrepancy.

Automatic cell segmentation in microscopy images works well with the support of deep neural networks trained with full supervision. Collecting and annotating images, though, is not a sustainable solution for every new microscopy database and cell type. Instead, we assume that we can access a plethora of annotated image data sets from different domains (sources) and a limited number of annotated image data sets from the domain of interest (target), where each domain denotes not only different image appearance but also a different type of cell segmentation problem. We pose this problem as meta-learning where the goal is to learn a generic and adaptable few-shot learning model from the available source domain data sets and cell segmentation tasks. The model can be afterwards fine-tuned on the few annotated images of the target domain that contains different image appearance and different cell type. In our meta-learning training, we propose the combination of three objective functions to segment the cells, move the segmentation results away from the classification boundary using cross-domain tasks, and learn an invariant representation between tasks of the source domains. Our experiments on five public databases show promising results from 1- to 10-shot meta-learning using standard segmentation neural network architectures.

Gradient-only and probabilistic line searches have recently reintroduced the ability to adaptively determine learning rates in dynamic mini-batch sub-sampled neural network training. However, stochastic line searches are still in their infancy and thus call for an ongoing investigation. We study the application of the Gradient-Only Line Search that is Inexact (GOLS-I) to automatically determine the learning rate schedule for a selection of popular neural network training algorithms, including NAG, Adagrad, Adadelta, Adam and LBFGS, with numerous shallow, deep and convolutional neural network architectures trained on different datasets with various loss functions. We find that GOLS-I's learning rate schedules are competitive with manually tuned learning rates, over seven optimization algorithms, three types of neural network architecture, 23 datasets and two loss functions. We demonstrate that algorithms, which include dominant momentum characteristics, are not well suited to be used with GOLS-I. However, we find GOLS-I to be effective in automatically determining learning rate schedules over 15 orders of magnitude, for most popular neural network training algorithms, effectively removing the need to tune the sensitive hyperparameters of learning rate schedules in neural network training.

How much knowledge do pretrained language models hold? Recent research observed that pretrained transformers are adept at modeling semantics but it is unclear to what degree they grasp human knowledge, or how to ensure they do so. In this paper we incorporate knowledge-awareness in language model pretraining without changing the transformer architecture, inserting explicit knowledge layers, or adding external storage of semantic information. Rather, we simply signal the existence of entities to the input of the transformer in pretraining, with an entity-extended tokenizer; and at the output, with an additional entity prediction task. Our experiments show that solely by adding these entity signals in pretraining, significantly more knowledge is packed into the transformer parameters: we observe improved language modeling accuracy, factual correctness in LAMA knowledge probing tasks, and semantics in the hidden representations through edge probing.We also show that our knowledge-aware language model (KALM) can serve as a drop-in replacement for GPT-2 models, significantly improving downstream tasks like zero-shot question-answering with no task-related training.

In this paper, we propose DiFF-RF, an ensemble approach composed of random partitioning binary trees to detect point-wise and collective (as well as contextual) anomalies. Thanks to a distance-based paradigm used at the leaves of the trees, this semi-supervised approach solves a drawback that has been identified in the isolation forest (IF) algorithm. Moreover, taking into account the frequencies of visits in the leaves of the random trees allows to significantly improve the performance of DiFF-RF when considering the presence of collective anomalies. DiFF-RF is fairly easy to train, and excellent performance can be obtained by using a simple semi-supervised procedure to setup the extra hyper-parameter that is introduced. We first evaluate DiFF-RF on a synthetic data set to i) verify that the limitation of the IF algorithm is overcome, ii) demonstrate how collective anomalies are actually detected and iii) to analyze the effect of the meta-parameters it involves. We assess the DiFF-RF algorithm on a large set of datasets from the UCI repository, as well as two benchmarks related to intrusion detection applications. Our experiments show that DiFF-RF almost systematically outperforms the IF algorithm, but also challenges the one-class SVM baseline and a deep learning variational auto-encoder architecture. Furthermore, our experience shows that DiFF-RF can work well in the presence of small-scale learning data, which is conversely difficult for deep neural architectures. Finally, DiFF-RF is computationally efficient and can be easily parallelized on multi-core architectures.

The COVID-19 pandemic has created unprecedented challenges worldwide. Strained healthcare providers make difficult decisions on patient triage, treatment and care management on a daily basis. Policy makers have imposed social distancing measures to slow the disease, at a steep economic price. We design analytical tools to support these decisions and combat the pandemic. Specifically, we propose a comprehensive data-driven approach to understand the clinical characteristics of COVID-19, predict its mortality, forecast its evolution, and ultimately alleviate its impact. By leveraging cohort-level clinical data, patient-level hospital data, and census-level epidemiological data, we develop an integrated four-step approach, combining descriptive, predictive and prescriptive analytics. First, we aggregate hundreds of clinical studies into the most comprehensive database on COVID-19 to paint a new macroscopic picture of the disease. Second, we build personalized calculators to predict the risk of infection and mortality as a function of demographics, symptoms, comorbidities, and lab values. Third, we develop a novel epidemiological model to project the pandemic's spread and inform social distancing policies. Fourth, we propose an optimization model to re-allocate ventilators and alleviate shortages. Our results have been used at the clinical level by several hospitals to triage patients, guide care management, plan ICU capacity, and re-distribute ventilators. At the policy level, they are currently supporting safe back-to-work policies at a major institution and equitable vaccine distribution planning at a major pharmaceutical company, and have been integrated into the US Center for Disease Control's pandemic forecast.