Many large-scale machine learning (ML) applications need to train ML models over decentralized datasets that are generated at different devices and locations. These decentralized datasets pose a fundamental challenge to ML because they are typically generated in very different contexts, which leads to significant differences in data distribution across devices/locations (i.e., they are not independent and identically distributed (IID)). In this work, we take a step toward better understanding this challenge, by presenting the first detailed experimental study of the impact of such non-IID data on the decentralized training of deep neural networks (DNNs). Our study shows that: (i) the problem of non-IID data partitions is fundamental and pervasive, as it exists in all ML applications, DNN models, training datasets, and decentralized learning algorithms in our study; (ii) this problem is particularly difficult for DNN models with batch normalization layers; and (iii) the degree of deviation from IID (the skewness) is a key determinant of the difficulty level of the problem. With these findings in mind, we present SkewScout, a system-level approach that adapts the communication frequency of decentralized learning algorithms to the (skew-induced) accuracy loss between data partitions. We also show that group normalization can recover much of the skew-induced accuracy loss of batch normalization.

Imaging-based early diagnosis of Alzheimer Disease (AD) has become an effective approach, especially by using nuclear medicine imaging techniques such as Positron Emission Topography (PET). In various literature it has been found that PET images can be better modeled as signals (e.g. uptake of florbetapir) defined on a network (non-Euclidean) structure which is governed by its underlying graph patterns of pathological progression and metabolic connectivity. In order to effectively apply deep learning framework for PET image analysis to overcome its limitation on Euclidean grid, we develop a solution for 3D PET image representation and analysis under a generalized, graph-based CNN architecture (PETNet), which analyzes PET signals defined on a group-wise inferred graph structure. Computations in PETNet are defined in non-Euclidean, graph (network) domain, as it performs feature extraction by convolution operations on spectral-filtered signals on the graph and pooling operations based on hierarchical graph clustering. Effectiveness of the PETNet is evaluated on the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset, which shows improved performance over both deep learning and other machine learning-based methods.

A BSTRACT It has been shown that instead of learning actual object features, deep networks tend to exploit non-robust (spurious) discriminative features that are shared between training and test sets. Therefore, while they achieve state of the art performance on such test sets, they achieve poor generalization on out of distribution (OOD) samples where the IID (independent, identical distribution) assumption breaks and the distribution of non-robust features shifts. Through theoretical and empirical analysis, we show that this happens because maximum likelihood training (without appropriate regularization) leads the model to depend on all the correlations (including spurious ones) present between inputs and targets in the dataset. We then show evidence that the information bottleneck (IB) principle can address this problem. To do so, we propose a regularization approach based on IB, called Entropy Penalty, that reduces the model's dependence on spurious features-features corresponding to such spurious correlations. This allows deep networks trained with Entropy Penalty to generalize well even under distribution shift of spurious features. As a controlled test-bed for evaluating our claim, we train deep networks with Entropy Penalty on a colored MNIST (C-MNIST) dataset and show that it is able to generalize well on vanilla MNIST, MNIST -M and SVHN datasets in addition to an OOD version of C-MNIST itself. The baseline regularization methods we compare against fail to generalize on this test-bed. An example of non-robust feature is the presence of desert in camel images, which may correlate well with this object class. More realistically, models can learn to exploit the abundance of input-target correlations present in datasets, not all of which may be invariant under different environments. Interestingly, such classifiers can achieve good performance on test sets which share the same non-robust features. However, due to this exploitation, these classifiers perform poorly under distribution shift (Geirhos et al., 2018a; Hendrycks & Dietterich, 2019) because it violates the IID assumption which is the foundation of existing generalization theory (Bartlett & Mendelson, 2002; McAllester, 1999b;a).

This paper introduces Meta-Q-Learning (MQL), a new off-policy algorithm for meta-Reinforcement Learning (meta-RL). MQL builds upon three simple ideas. First, we show that Q-learning is competitive with state of the art meta-RL algorithms if given access to a context variable that is a representation of the past trajectory. Second, using a multi-task objective to maximize the average reward across the training tasks is an effective method to meta-train RL policies. Third, past data from the meta-training replay buffer can be recycled to adapt the policy on a new task using off-policy updates. MQL draws upon ideas in propensity estimation to do so and thereby amplifies the amount of available data for adaptation. Experiments on standard continuous-control benchmarks suggest that MQL compares favorably with state of the art meta-RL algorithms.

Deep Learning has established itself to be a common occurrence in the business lexicon. The unprecedented success of deep learning in recent years can be attributed to: abundance of data, availability of gargantuan compute capabilities offered by GPUs, and adoption of open-source philosophy by the researchers and industry. Deep neural networks can be decomposed into a series of different operators. MIOpen, AMD's open-source deep learning primitives library for GPUs, provides highly optimized implementations of such operators, shielding researchers from internal implementation details and hence, accelerating the time to discovery. This paper introduces MIOpen and provides details about the internal workings of the library and supported features. MIOpen innovates on several fronts, such as implementing fusion to optimize for memory bandwidth and GPU launch overheads, providing an auto-tuning infrastructure to overcome the large design space of problem configurations, and implementing different algorithms to optimize convolutions for different filter and input sizes. MIOpen is one of the first libraries to publicly support the bfloat16 data-type for convolutions, allowing efficient training at lower precision without the loss of accuracy.

In this work we introduce a semi-supervised approach to the voice conversion problem, in which speech from a source speaker is converted into speech of a target speaker. The proposed method makes use of both parallel and non-parallel utterances from the source and target simultaneously during training. This approach can be used to extend existing parallel data voice conversion systems such that they can be trained with semi-supervision. We show that incorporating semi-supervision improves the voice conversion performance compared to fully supervised training when the number of parallel utterances is limited as in many practical applications. Additionally, we find that increasing the number non-parallel utterances used in training continues to improve performance when the amount of parallel training data is held constant.

In many review classification applications, a fine-grained analysis of the reviews is desirable, because different segments (e.g., sentences) of a review may focus on different aspects of the entity in question. However, training supervised models for segment-level classification requires segment labels, which may be more difficult or expensive to obtain than review labels. In this paper, we employ Multiple Instance Learning (MIL) and use only weak supervision in the form of a single label per review. First, we show that when inappropriate MIL aggregation functions are used, then MIL-based networks are outperformed by simpler baselines. Second, we propose a new aggregation function based on the sigmoid attention mechanism and show that our proposed model outperforms the state-of-the-art models for segment-level sentiment classification (by up to 9.8% in F1). Finally, we highlight the importance of fine-grained predictions in an important public-health application: finding actionable reports of foodborne illness. We show that our model achieves 48.6% higher recall compared to previous models, thus increasing the chance of identifying previously unknown foodborne outbreaks.

October 1, 2019 Abstract We focus on the problem of black-box adversarial attacks, where the aim is to generate adversarial examples using information limited to loss function evaluations of input-output pairs. We use Bayesian optimization (BO) to specifically cater to scenarios involving low query budgets to develop query efficient adversarial attacks. We alleviate the issues surrounding BO in regards to optimizing high dimensional deep learning models by effective dimension upsampling techniques. Our proposed approach achieves performance comparable to the state of the art black-box adversarial attacks albeit with a much lower average query count. In particular, in low query budget regimes, our proposed method reduces the query count up to 80% with respect to the state of the art methods. 1 Introduction Neural networks are now well-known to be vulnerable to adversarial examples: additive perturbations that, when applied to the input, change the network's output classification [9]. Work investigating this lack of robustness to adversarial examples often takes the form of a back-and-forth between newly proposed adversarial attacks, methods for quickly and efficiently crafting adversarial examples, and corresponding defenses that modify the classifier at either training or test time to improve robustness. The most successful adversarial attacks use gradient-based optimization methods [9, 17], which require complete knowledge of the architecture and parameters of the target network; this assumption is referred to as the white-box attack setting.

Applying deep neural networks (DNNs) in mobile and safety-critical systems, such as autonomous vehicles, demands a reliable and efficient execution on hardware. Optimized dedicated hardware accelerators are being developed to achieve this. However, the design of efficient and reliable hardware has become increasingly difficult, due to the increased complexity of modern integrated circuit technology and its sensitivity against hardware faults, such as random bit-flips. It is thus desirable to exploit optimization potential for error resilience and efficiency also at the algorithmic side, e.g. by optimizing the architecture of the DNN. Since there are numerous design choices for the architecture of DNNs, with partially opposing effects on the preferred characteristics (such as small error rates at low latency), multi-objective optimization strategies are necessary. In this paper, we develop an evolutionary optimization technique for the automated design of hardware-optimized DNN architectures. For this purpose, we derive a set of easily computable objective functions, which enable the fast evaluation of DNN architectures with respect to their hardware efficiency and error resilience solely based on the network topology. We observe a strong correlation between predicted error resilience and actual measurements obtained from fault injection simulations. Keywords Neural Network Hardware · Error Resilience · Hardware Faults · Neural Architecture Search · Multi-Objective Optimization · AutoML 1 Introduction The application of deep neural networks (DNNs) in safety-critical perception systems, for example autonomous vehicles (A Vs), poses some challenges on the design of the underlying hardware platforms. On the one hand, efficient and fast accelerators are needed, since DNNs for computer vision exhibit massive computational requirements [55]. On the other hand, resilience against random hardware faults has to be ensured. In many driving scenarios, entering a fail-safe state is not sufficient, but fail-operational behavior and fault tolerance are required [48]. However, fault tolerance techniques at the hardware level often entail large redundancy overheads in silicon area, latency, and power consumption. These overheads stand in contrast to the low-power and low-latency requirements of embedded real-time DNN accelerators. Reliability concerns in nanoscale integrated circuits, for instance soft errors in memory and logic, represent an additional challenge for the realization of fault tolerance mechanisms at the hardware level [2, 33, 36, 68, 83]. Moreover, techniques such as near-threshold computing [26] and approximate computing [65] are desirable to meet power constraints, but can further increase error rates.

The Hamiltonian formalism plays a central role in classical and quantum physics. Hamiltonians are the main tool for modelling the continuous time evolution of systems with conserved quantities, and they come equipped with many useful properties, like time reversibility and smooth interpolation in time. These properties are important for many machine learning problems - from sequence prediction to reinforcement learning and density modelling - but are not typically provided out of the box by standard tools such as recurrent neural networks. In this paper, we introduce the Hamiltonian Generative Network (HGN), the first approach capable of consistently learning Hamiltonian dynamics from high-dimensional observations (such as images) without restrictive domain assumptions. Once trained, we can use HGN to sample new trajectories, perform rollouts both forward and backward in time and even speed up or slow down the learned dynamics. We demonstrate how a simple modification of the network architecture turns HGN into a powerful normalising flow model, called Neural Hamiltonian Flow (NHF), that uses Hamiltonian dynamics to model expressive densities. We hope that our work serves as a first practical demonstration of the value that the Hamiltonian formalism can bring to deep learning.