Monoaural audio source separation is a challenging research area in machine learning. In this area, a mixture containing multiple audio sources is given, and a model is expected to disentangle the mixture into isolated atomic sources. In this paper, we first introduce a challenging new dataset for monoaural source separation called WildMix. WildMix is designed with the goal of extending the boundaries of source separation beyond what previous datasets in this area would allow. It contains diverse in-the-wild recordings from 25 different sound classes, combined with each other using arbitrary composition policies. Source separation often requires modeling long-range dependencies in both temporal and spectral domains. To this end, we introduce a novel trasnformer-based model called Spectro-Temporal Transformer (STT). STT utilizes a specialized encoder, called Spectro-Temporal Encoder (STE). STE highlights temporal and spectral components of sources within a mixture, using a self-attention mechanism. It subsequently disentangles them in a hierarchical manner. In our experiments, STT swiftly outperforms various previous baselines for monoaural source separation on the challenging WildMix dataset.

A BSTRACT Performing controlled experiments on noisy data is essential in thoroughly understanding deep learning across a spectrum of noise levels. Due to the lack of suitable datasets, previous research have only examined deep learning on controlled synthetic noise, and real-world noise has never been systematically studied in a controlled setting. To this end, this paper establishes a benchmark of real-world noisy labels at 10 controlled noise levels. As real-world noise possesses unique properties, to understand the difference, we conduct a large-scale study across a variety of noise levels and types, architectures, methods, and training settings. Our study shows that: (1) Deep Neural Networks (DNNs) generalize much better on real-world noise. We hope our benchmark, as well as our findings, will facilitate deep learning research on noisy data. 1 I NTRODUCTION Y ou take the blue pill you wake up in your bed and believe whatever you want to believe. Y ou take the red pill and I show you how deep the rabbit hole goes. Remember, all I'm offering is the truth. Morpheus (The Matrix 1999) Deep Neural Networks (DNNs) trained on noisy data demonstrate intriguing properties. For example, DNNs are capable of memorizing completely random training labels but generalize poorly on clean test data Zhang et al. (2017). When trained with stochastic gradient descent, DNNs learn patterns first before memorizing the label noise Arpit et al. (2017). These findings inspired recent research on noisy data. As training data are usually noisy, the fact that DNNs are able to memorize the noisy labels highlights the importance of deep learning research on noisy data. To study DNNs on noisy data, previous work often performs controlled experiments by injecting a series of synthetic noises into a well-annotated dataset. The noise level p may vary in the range of 0%- 100%, where p 0% is the clean dataset whereas p 100% represents the dataset of zero correct labels.

Membership inference attacks seek to infer the membership of individual training instances of a privately trained model. This paper presents a membership privacy analysis and evaluation system, called MPLens, with three unique contributions. First, through MPLens, we demonstrate how membership inference attack methods can be leveraged in adversarial machine learning. Second, through MPLens, we highlight how the vulnerability of pre-trained models under membership inference attack is not uniform across all classes, particularly when the training data itself is skewed. We show that risk from membership inference attacks is routinely increased when models use skewed training data. Finally, we investigate the effectiveness of differential privacy as a mitigation technique against membership inference attacks. We discuss the trade-offs of implementing such a mitigation strategy with respect to the model complexity, the learning task complexity, the dataset complexity and the privacy parameter settings. Our empirical results reveal that (1) minority groups within skewed datasets display increased risk for membership inference and (2) differential privacy presents many challenging trade-offs as a mitigation technique to membership inference risk.

Batch Normalization (BN) is a highly successful and widely used batch dependent training method. Its use of mini-batch statistics to normalize the activations introduces dependence between samples, which can hurt the training if the mini-batch size is too small, or if the samples are correlated. Several alternatives, such as Batch Renormalization and Group Normalization (GN), have been proposed to address these issues. However, they either do not match the performance of BN for large batches, or still exhibit degradation in performance for smaller batches, or introduce artificial constraints on the model architecture. In this paper we propose the Filter Response Normalization (FRN) layer, a novel combination of a normalization and an activation function, that can be used as a drop-in replacement for other normalizations and activations. Our method operates on each activation map of each batch sample independently, eliminating the dependency on other batch samples or channels of the same sample. Our method outperforms BN and all alternatives in a variety of settings for all batch sizes. FRN layer performs $\approx 0.7-1.0\%$ better on top-1 validation accuracy than BN with large mini-batch sizes on Imagenet classification on InceptionV3 and ResnetV2-50 architectures. Further, it performs $>1\%$ better than GN on the same problem in the small mini-batch size regime. For object detection problem on COCO dataset, FRN layer outperforms all other methods by at least $0.3-0.5\%$ in all batch size regimes.

Abstract--Event-based cameras are bio-inspired novel sensors that asynchronously record changes in illumination in the form of events, thus resulting in significant advantages over conventional cameras in terms of low power utilization, high dynamic range, and no motion blur. Moreover, such cameras, by design, encode only the relative motion between the scene and the sensor (and not the static background) to yield a very sparse data structure, which can be utilized for various motion analytics tasks. We propose to model the motion dynamics in the event domain with dual discriminator conditional Generative adversarial Network (cGAN) built on state-of-the-art architectures. T o adapt event data for using as input to cGAN, we also put forward a deep learning solution to learn a novel representation of event data, which retains the sparsity of the data as well as encode the temporal information readily available from these sensors. Since there is no existing dataset for anomaly detection in event domain, we also provide an anomaly detection event dataset with an exhaustive set of anomalies. Index Terms --Neuromorphic Camera, Event data, Anomaly Detection, Generative Adversarial Network.null 1 I NTRODUCTION This paper focusses on anomaly detection using bio-inspired event-based cameras that register pixel-wise changes in brightness asynchronously in an efficient manner, which is radically different from how a conventional camera works. The asynchronous principle of operation endows event cameras [9] [10] [36] [41] to capture high-speed motions (with temporal resolution in the order of µs), high dynamic range ( 140 db) and sparse data. These low latency sensors have paved way to develop agile robotic applications [1], which was not feasible with conventional cameras.

We develop a communication-efficient distributed learning algorithm that is robust against Byzantine worker machines. We propose and analyze a distributed gradient-descent algorithm that performs a simple thresholding based on gradient norms to mitigate Byzantine failures. We show the (statistical) error-rate of our algorithm matches that of [YCKB18], which uses more complicated schemes (like coordinate-wise median or trimmed mean) and thus optimal. Furthermore, for communication efficiency, we consider a generic class of {\delta}-approximate compressors from [KRSJ19] that encompasses sign-based compressors and top-k sparsification. Our algorithm uses compressed gradients and gradient norms for aggregation and Byzantine removal respectively. We establish the statistical error rate of the algorithm for arbitrary (convex or non-convex) smooth loss function. We show that, in the regime when the compression factor {\delta} is constant and the dimension of the parameter space is fixed, the rate of convergence is not affected by the compression operation, and hence we effectively get the compression for free. Moreover, we extend the compressed gradient descent algorithm with error feedback proposed in [KRSJ19] for the distributed setting. We have experimentally validated our results and shown good performance in convergence for convex (least-square regression) and non-convex (neural network training) problems.

We study a generalized setup for learning from demonstration to build an agent that can manipulate novel objects in unseen scenarios by looking at only a single video of human demonstration from a third-person perspective. To accomplish this goal, our agent should not only learn to understand the intent of the demonstrated third-person video in its context but also perform the intended task in its environment configuration. Our central insight is to enforce this structure explicitly during learning by decoupling what to achieve (intended task) from how to perform it (controller). We propose a hierarchical setup where a high-level module learns to generate a series of first-person sub-goals conditioned on the third-person video demonstration, and a low-level controller predicts the actions to achieve those sub-goals. Our agent acts from raw image observations without any access to the full state information. We show results on a real robotic platform using Baxter for the manipulation tasks of pouring and placing objects in a box. Project video and code are at https://pathak22.github.io/hierarchical-imitation/

REGULARIZING NEURAL NETWORKS BY STOCHASTICALL Y TRAINING LA YER ENSEMBLES Alex Labach and Shahrokh V alaee University of Toronto Department of Electrical and Computer Engineering Toronto, Canada ABSTRACT Dropout and similar stochastic neural network regularization methods are often interpreted as implicitly averaging over a large ensemble of models. We propose STE (stochastically trained ensemble) layers, which enhance the averaging properties of such methods by training an ensemble of weight matrices with stochastic regularization while explicitly averaging outputs. This provides stronger regularization with no additional computational cost at test time. We show consistent improvement on various image classification tasks using standard network topologies. Index T erms-- neural networks, regularization, dropout, model averaging, ensemble methods 1. INTRODUCTION In order to generalize well to new inputs, modern deep neural networks require heavy regularization.

Approaches to continual learning aim to successfully learn a set of related tasks that arrive in an online manner. Recently, several frameworks have been developed which enable deep learning to be deployed in this learning scenario. A key modelling decision is to what extent the architecture should be shared across tasks. On the one hand, separately modelling each task avoids catastrophic forgetting but it does not support transfer learning and leads to large models. On the other hand, rigidly specifying a shared component and a task-specific part enables task transfer and limits the model size, but it is vulnerable to catastrophic forgetting and restricts the form of task-transfer that can occur. Ideally, the network should adaptively identify which parts of the network to share in a data driven way. Here we introduce such an approach called Continual Learning with Adaptive Weights (CLAW), which is based on probabilistic modelling and variational inference. Experiments show that CLAW achieves state-of-the-art performance on six benchmarks in terms of overall continual learning performance, as measured by classification accuracy, and in terms of addressing catastrophic forgetting.

Machine and deep learning-based algorithms are the emerging approaches in addressing prediction problems in time series. These techniques have been shown to produce more accurate results than conventional regression-based modeling. It has been reported that artificial Recurrent Neural Networks (RNN) with memory, such as Long Short-Term Memory (LSTM), are superior compared to Autoregressive Integrated Moving Average (ARIMA) with a large margin. The LSTM-based models incorporate additional "gates" for the purpose of memorizing longer sequences of input data. The major question is that whether the gates incorporated in the LSTM architecture already offers a good prediction and whether additional training of data would be necessary to further improve the prediction. Bidirectional LSTMs (BiLSTMs) enable additional training by traversing the input data twice (i.e., 1) left-to-right, and 2) right-to-left). The research question of interest is then whether BiLSTM, with additional training capability, outperforms regular unidirectional LSTM. This paper reports a behavioral analysis and comparison of BiLSTM and LSTM models. The objective is to explore to what extend additional layers of training of data would be beneficial to tune the involved parameters. The results show that additional training of data and thus BiLSTM-based modeling offers better predictions than regular LSTM-based models. More specifically, it was observed that BiLSTM models provide better predictions compared to ARIMA and LSTM models. It was also observed that BiLSTM models reach the equilibrium much slower than LSTM-based models.