The convolutional layers are core building blocks of neural network architectures. In general, a convolutional filter applies to the entire frequency spectrum of the input data. We explore artificially constraining the frequency spectra of these filters and data, called band-limiting, during training. The frequency domain constraints apply to both the feed-forward and back-propagation steps. Experimentally, we observe that Convolutional Neural Networks (CNNs) are resilient to this compression scheme and results suggest that CNNs learn to leverage lower-frequency components. In particular, we found: (1) band-limited training can effectively control the resource usage (GPU and memory); (2) models trained with band-limited layers retain high prediction accuracy; and (3) requires no modification to existing training algorithms or neural network architectures to use unlike other compression schemes.

--Global physical event detection has traditionally relied on dense coverage of physical sensors around the world; while this is an expensive undertaking, there have not been alternatives until recently. The ubiquity of social networks and human sensors in the field provides a tremendous amount of real-time, live data about true physical events from around the world. However, while such human sensor data have been exploited for retrospective large-scale event detection, such as hurricanes or earthquakes, they has been limited to no success in exploiting this rich resource for general physical event detection. Prior implementation approaches have suffered from the concept drift phenomenon, where real-world data exhibits constant, unknown, unbounded changes in its data distribution, making static machine learning models ineffective in the long term. We propose and implement an end-to-end collaborative drift adaptive system that integrates corroborative and probabilistic sources to deliver real-time predictions. Furthermore, out system is adaptive to concept drift and performs automated continuous learning to maintain high performance. We demonstrate our approach in a real-time demo available online for landslide disaster detection, with extensibility to other real-world physical events such as flooding, wildfires, hurricanes, and earthquakes. Physical event detection, such as extreme weather events or traffic accidents have long been the domain of static event processors operating on numeric sensor data or human actors manually identifying event types. However, the emergence of big data and associated data processing and analytics tools and systems have led to several applications in large-scale event and trend detection in the streaming domain [1]-[7]. However, it is important to note that many of these works are a form of retrospective analysis, as opposed to true real-time event detection, since they perform analyses on cleaned and processed data within a short-time frame in the past, with the assumption that their approaches are sustainable and will continue to function over time.

Recently, techniques have been developed to provably guarantee the robustness of a classifier to adversarial perturbations of bounded L 1 and L 2 magnitudes by using randomized smoothing: the robust classification is a consensus of base classifications on randomly noised samples where the noise is additive. In this paper, we extend this technique to the L 0 threat model. We propose an efficient and certifiably robust defense against sparse adversarial attacks by randomly ablating input features, rather than using additive noise. Experimentally, on MNIST, we can certify the classifications of over 50% of images to be robust to any distortion of at most 8 pixels. This is comparable to the observed empirical robustness of unprotected classifiers on MNIST to modern L 0 attacks, demonstrating the tightness of the proposed robustness certificate. We also evaluate our certificate on ImageNet and CIFAR-10. Our certificates represent an improvement on those provided in a concurrent work (Lee et al. 2019) which uses random noise rather than ablation (median certificates of 8 pixels versus 4 pixels on MNIST; 16 pixels versus 1 pixel on ImageNet.) Additionally, we empirically demonstrate that our classifier is highly robust to modern sparse adversarial attacks on MNIST. Our classifications are robust, in median, to adversarial perturbations of up to 31 pixels, compared to 22 pixels reported as the state-of-the-art defense, at the cost of a slight decrease (around 2. 3%) in the classification accuracy. Introduction Adversarial attacks, and defenses against these attacks, have been active topics of research in machine learning in recent years (Szegedy et al. 2013; Carlini and Wagner 2017; Madry et al. 2017).

Resource is an important constraint when deploying Deep Neural Networks (DNNs) on mobile and edge devices. Existing works commonly adopt the cell-based search approach, which limits the flexibility of network patterns in learned cell structures. Moreover, due to the topology-agnostic nature of existing works, including both cell-based and node-based approaches, the search process is time consuming and the performance of found architecture may be sub-optimal. To address these problems, we propose AutoShrink, a topology-aware Neural Architecture Search(NAS) for searching efficient building blocks of neural architectures. Our method is node-based and thus can learn flexible network patterns in cell structures within a topological search space. Directed Acyclic Graphs (DAGs) are used to abstract DNN architectures and progressively optimize the cell structure through edge shrinking. As the search space intrinsically reduces as the edges are progressively shrunk, AutoShrink explores more flexible search space with even less search time. We evaluate AutoShrink on image classification and language tasks by crafting ShrinkCNN and ShrinkRNN models. ShrinkCNN is able to achieve up to 48% parameter reduction and save 34% Multiply-Accumulates (MACs) on ImageNet-1K with comparable accuracy of state-of-the-art (SOTA) models. Specifically, both ShrinkCNN and ShrinkRNN are crafted within 1.5 GPU hours, which is 7.2x and 6.7x faster than the crafting time of SOTA CNN and RNN models, respectively.

In this paper, we proposed a unified and principled method for both querying and training process in deep batch active learning. We provided the theoretical insights from the intuition of modeling the interactive procedure in active learning as distribution matching, by adopting Wasserstein distance. As a consequence, we derived a new training loss from the theoretical analysis, which is decomposed into optimizing deep neural network parameter and batch query selection through alternative optimization. In addition, the loss for training deep neural network is naturally formulated as a min-max optimization problem through leveraging the unlabeled data information. Moreover, the proposed principles also indicate an explicit uncertainty-diversity trade-off in the query batch selection. Finally we evaluated our proposed method for different benchmarks, showing consistently better empirical performances and more time efficient query strategy, comparing to several baselines.

Often the development of novel materials is not amenable to high-throughput or purely computational screening methods. Instead, materials must be synthesized one at a time in a process that does not generate significant amounts of data. One way this method can be improved is by ensuring that each experiment provides the best improvement in both material properties and predictive modeling accuracy. In this work, we study the effectiveness of active learning, which optimizes the order of experiments, and meta learning, which transfers knowledge from one context to another, to reduce the number of experiments necessary to build a predictive model. We present a novel multi-task benchmark database of peptides designed to advance active, few-shot, and meta-learning methods for experimental design. Each task is binary classification of peptides represented as a sequence string. We show results of standard active learning and meta-learning methods across these datasets to assess their ability to improve predictive models with the fewest number of experiments. We find the ensemble query by committee active learning method to be effective. The meta-learning method Reptile was found to improve accuracy. The robustness of these conclusions were tested across multiple model choices.

We propose a novel approach for generating unrestricted adversarial examples by manipulating fine-grained aspects of image generation. Unlike existing unrestricted attacks that typically hand-craft geometric transformations, we learn stylistic and stochastic modifications leveraging state-of-the-art generative models. This allows us to manipulate an image in a controlled, fine-grained manner without being bounded by a norm threshold. Our model can be used for both targeted and non-targeted unrestricted attacks. We demonstrate that our attacks can bypass certified defenses, yet our adversarial images look indistinguishable from natural images as verified by human evaluation. Adversarial training can be used as an effective defense without degrading performance of the model on clean images. We perform experiments on LSUN and CelebA-HQ as high resolution datasets to validate efficacy of our proposed approach.

Note that most widely used benchmark datasets for point cloud classification only contain foreground objects. Therefore, we generate a new dataset, where each point cloud contains both the foreground object and the background. In this new dataset, the background is composed of points that carry no relevant information of the foreground. We will introduce details in Section 5. Metric 3, rotation robustness: The rotation robustness is proposed to measure whether a DNN uses similar subsets of two point clouds to compute the intermediate-layer feature, if the two point clouds have the same shape but different orientations. Let X θ 1 and X θ 2 denote the point clouds that have the same global shape but different orientations θ 1 and θ 2. To quantify the similarity of the attention on the two point clouds, we compute the Jensen-Shannon divergence between the distributions of the perturbed inputs ˆ X θ 1 X θ 1 δ 1 and ˆ X θ 2 X θ 2 δ 2. ˆ X θ 1 and ˆ X θ 2 denote the perturbed inputs, which are computed to measure information discarding in Equation (1).

Crop yield prediction is extremely challenging due to its dependence on multiple factors such as crop genotype, environmental factors, management practices, and their interactions. This paper presents a deep learning framework using convolutional neural networks (CNN) and recurrent neural networks (RNN) for crop yield prediction based on environmental data and management practices. The proposed CNN-RNN model, along with other popular methods such as random forest (RF), deep fully-connected neural networks (DFNN), and LASSO, was used to forecast corn and soybean yield across the entire Corn Belt (including 13 states) in the United States for years 2016, 2017, and 2018 using historical data. The new model achieved a root-mean-square-error (RMSE) 9% and 8% of their respective average yields, substantially outperforming all other methods that were tested. The CNN-RNN have three salient features that make it a potentially useful method for other crop yield prediction studies. (1) The CNN-RNN model was designed to capture the time dependencies of environmental factors and the genetic improvement of seeds over time without having their genotype information. (2) The model demonstrated the capability to generalize the yield prediction to untested environments without significant drop in the prediction accuracy. (3) Coupled with the backpropagation method, the model could reveal the extent to which weather conditions, accuracy of weather predictions, soil conditions, and management practices were able to explain the variation in the crop yields.

This paper proposes a set of rules to revise various neural networks for 3D point cloud processing to rotation-equivariant quaternion neural networks (REQNNs). W e find that when a neural network uses quaternion features under certain conditions, the network feature naturally has the rotation-equivariance property. Rotation equivariance means that applying a specific rotation transformation to the input point cloud is equivalent to applying the same rotation transformation to all intermediate-layer quaternion features. Besides, the REQNN also ensures that the intermediate-layer features are invariant to the permutation of input points. Compared with the original neural network, the REQNN exhibits higher rotation robustness.