Anomaly detection incurs certain sampling and sensing costs and therefore it is of great importance to strike a balance between the detection accuracy and these costs. In this work, we study anomaly detection by considering the detection of threshold crossings in a stochastic time series without the knowledge of its statistics. To reduce the sampling cost in this detection process, we propose the use of hierarchical generative adversarial networks (GANs) to perform nonuniform sampling. In order to improve the detection accuracy and reduce the delay in detection, we introduce a buffer zone in the operation of the proposed GAN-based detector. In the experiments, we analyze the performance of the proposed hierarchical GAN detector considering the metrics of detection delay, miss rates, average cost of error, and sampling ratio. We identify the tradeoffs in the performance as the buffer zone sizes and the number of GAN levels in the hierarchy vary. We also compare the performance with that of a sampling policy that approximately minimizes the sum of average costs of sampling and error given the parameters of the stochastic process. We demonstrate that the proposed GAN-based detector can have significant performance improvements in terms of detection delay and average cost of error with a larger buffer zone but at the cost of increased sampling rates.

Adversarial examples are imperceptible perturbations in the input to a neural model that result in misclassification. Generating adversarial examples for source code poses an additional challenge compared to the domains of images and natural language, because source code perturbations must adhere to strict semantic guidelines so the resulting programs retain the functional meaning of the code. We propose a simple and efficient black-box method for generating state-of-the-art adversarial examples on models of code. Our method generates untargeted and targeted attacks, and empirically outperforms competing gradient-based methods with less information and less computational effort. We also use adversarial training to construct a model robust to these attacks; our attack reduces the F1 score of code2seq by 42%. Adversarial training brings the F1 score on adversarial examples up to 99% of baseline.

Batch normalization (BN) is an important technique commonly incorporated into deep learning models to perform standardization within mini-batches. The merits of BN in improving model's learning efficiency can be further amplified by applying whitening, while its drawbacks in estimating population statistics for inference can be avoided through group normalization (GN). This paper proposes group whitening (GW), which elaborately exploits the advantages of the whitening operation and avoids the disadvantages of normalization within mini-batches. Specifically, GW divides the neurons of a sample into groups for standardization, like GN, and then further decorrelates the groups. In addition, we quantitatively analyze the constraint imposed by normalization, and show how the batch size (group number) affects the performance of batch (group) normalized networks, from the perspective of model's representational capacity. This analysis provides theoretical guidance for applying GW in practice. Finally, we apply the proposed GW to ResNet and ResNeXt architectures and conduct experiments on the ImageNet and COCO benchmarks. Results show that GW consistently improves the performance of different architectures, with absolute gains of $1.02\%$ $\sim$ $1.49\%$ in top-1 accuracy on ImageNet and $1.82\%$ $\sim$ $3.21\%$ in bounding box AP on COCO.

Transfer of pre-trained representations can improve sample efficiency and reduce computational requirements for new tasks. However, representations used for transfer are usually generic, and are not tailored to a particular distribution of downstream tasks. We explore the use of expert representations for transfer with a simple, yet effective, strategy. We train a diverse set of experts by exploiting existing label structures, and use cheap-to-compute performance proxies to select the relevant expert for each target task. This strategy scales the process of transferring to new tasks, since it does not revisit the pre-training data during transfer. Accordingly, it requires little extra compute per target task, and results in a speed-up of 2-3 orders of magnitude compared to competing approaches. Further, we provide an adapter-based architecture able to compress many experts into a single model. We evaluate our approach on two different data sources and demonstrate that it outperforms baselines on over 20 diverse vision tasks in both cases.

Coronary artery disease(CAD) is the most common type of heart disease and the leading cause of death worldwide[1]. A progressive state of this disease marked by plaque rupture and clot formation in the coronary arteries, also known as an acute coronary syndrome (ACS), is a condition of the heart associated with sudden, reduced blood flow caused due to partial or full occlusion of coronary vasculature that normally perfuses the myocardium and nerve bundles, compromising the proper functioning of the heart. Often manifesting with pain or tightness in the chest as the second most common cause of emergency department visits in the United States, it is imperative to detect ACS at the earliest. This is particularly relevant to diabetic patients at home, that may not feel classic chest pain symptoms, and are susceptible to silent myocardial injury. In this study, we developed the RCE- ECG-Detect algorithm, a machine learning model to detect the morphological patterns in significant ST change associated with myocardial ischemia. We developed the RCE- ECG-Detect using data from the LTST database which has a sufficiently large sample set to train a reliable model. We validated the predictive performance of the machine learning model on a holdout test set collected using RCE's ECG wearable. Our deep neural network model, equipped with convolution layers, achieves 90.31% ROC-AUC, 89.34% sensitivity, 87.81% specificity.

Regularization improves generalization of supervised models to out-of-sample data. Prior works have shown that prediction in the causal direction (effect from cause) results in lower testing error than the anti-causal direction. However, existing regularization methods are agnostic of causality. We introduce Causal Structure Learning (CASTLE) regularization and propose to regularize a neural network by jointly learning the causal relationships between variables. CASTLE learns the causal directed acyclical graph (DAG) as an adjacency matrix embedded in the neural network's input layers, thereby facilitating the discovery of optimal predictors. Furthermore, CASTLE efficiently reconstructs only the features in the causal DAG that have a causal neighbor, whereas reconstruction-based regularizers suboptimally reconstruct all input features. We provide a theoretical generalization bound for our approach and conduct experiments on a plethora of synthetic and real publicly available datasets demonstrating that CASTLE consistently leads to better out-of-sample predictions as compared to other popular benchmark regularizers.

This paper is an attempt to bridge the gap between deep learning and grammatical inference. Indeed, it provides an algorithm to extract a (stochastic) formal language from any recurrent neural network trained for language modelling. In detail, the algorithm uses the already trained network as an oracle -- and thus does not require the access to the inner representation of the black-box -- and applies a spectral approach to infer a weighted automaton. As weighted automata compute linear functions, they are computationally more efficient than neural networks and thus the nature of the approach is the one of knowledge distillation. We detail experiments on 62 data sets (both synthetic and from real-world applications) that allow an in-depth study of the abilities of the proposed algorithm. The results show the WA we extract are good approximations of the RNN, validating the approach. Moreover, we show how the process provides interesting insights toward the behavior of RNN learned on data, enlarging the scope of this work to the one of explainability of deep learning models.

Standardized DNN models that have been proved to perform well on machine learning tasks are widely used and often adopted as-is to solve downstream tasks, forming the transfer learning paradigm. However, when serving multiple instances of such DNN models from a cluster of GPU servers, existing techniques to improve GPU utilization such as batching are inapplicable because models often do not share weights due to fine-tuning. We propose NetFuse, a technique of merging multiple DNN models that share the same architecture but have different weights and different inputs. NetFuse is made possible by replacing operations with more general counterparts that allow a set of weights to be associated with only a certain set of inputs. Experiments on ResNet-50, ResNeXt-50, BERT, and XLNet show that NetFuse can speed up DNN inference time up to 3.6x on a NVIDIA V100 GPU, and up to 3.0x on a TITAN Xp GPU when merging 32 model instances, while only using up a small additional amount of GPU memory.

We propose Parametric UMAP, a parametric variation of the UMAP (Uniform Manifold Approximation and Projection) algorithm. UMAP is a nonparametric graph-based dimensionality reduction algorithm using applied Riemannian geometry and algebraic topology to find low-dimensional embeddings of structured data. The UMAP algorithm consists of two steps: (1) Compute a graphical representation of a dataset (fuzzy simplicial complex), and (2) Through stochastic gradient descent, optimize a low-dimensional embedding of the graph. Here, we replace the second step of UMAP with a deep neural network that learns a parametric relationship between data and embedding. We demonstrate that our method performs similarly to its nonparametric counterpart while conferring the benefit of a learned parametric mapping (e.g. We then show that UMAP loss can be extended to arbitrary deep learning applications, for example constraining the latent distribution of autoencoders, and improving classifier accuracy for semi-supervised learning by capturing structure in unlabeled data. Current nonlinear dimensionality reduction algorithms can be divided broadly into nonparametric algorithms which rely on the efficient computation of probabilistic relationships from neighborhood graphs to extract structure in large datasets (e.g. UMAP (McInnes et al., 2018), t-SNE (van der Maaten & Hinton, 2008), LargeVis (Tang et al., 2016)), and parametric algorithms, which, driven by advances in deep-learning, optimize an objective function related to capturing structure in a dataset over neural network weights (e.g. The goal of this paper is to wed those two classes of methods: learning a structured graphical representation of the data and using a deep neural network to embed that graph. Over the past decade several varients of the t-SNE algorithm have proposed parameterized forms of t-SNE (Van Der Maaten, 2009; Gisbrecht et al., 2015; Bunte et al., 2012; Gisbrecht et al., 2012).

We consider the problem of representation learning for temporal interaction graphs where a network of entities with complex interactions over an extended period of time is modeled as a graph with a rich set of node and edge attributes. In particular, an edge between a node-pair within the graph corresponds to a multi-dimensional time-series. To fully capture and model the dynamics of the network, we propose GTEA, a framework of representation learning for temporal interaction graphs with per-edge time-based aggregation. Under GTEA, a Graph Neural Network (GNN) is integrated with a state-of-the-art sequence model, such as LSTM, Transformer and their time-aware variants. The sequence model generates edge embeddings to encode temporal interaction patterns between each pair of nodes, while the GNN-based backbone learns the topological dependencies and relationships among different nodes. GTEA also incorporates a sparsity-inducing self-attention mechanism to distinguish and focus on the more important neighbors of each node during the aggregation process. By capturing temporal interactive dynamics together with multi-dimensional node and edge attributes in a network, GTEA can learn fine-grained representations for a temporal interaction graph to enable or facilitate other downstream data analytic tasks. Experimental results show that GTEA outperforms state-of-the-art schemes including GraphSAGE, APPNP, and TGAT by delivering higher accuracy (100.00%, 98.51%, 98.05% ,79.90%) and macro-F1 score (100.00%, 98.51%, 96.68% ,79.90%) over four large-scale real-world datasets for binary/ multi-class node classification.