Considering music as a sequence of events with multiple complex dependencies, the Long Short-Term Memory (LSTM) architecture has proven very efficient in learning and reproducing musical styles. However, the generation of rhythms requires additional information regarding musical structure and accompanying instruments. In this paper we present DeepDrum, an adaptive Neural Network capable of generating drum rhythms under constraints imposed by Feed-Forward (Conditional) Layers which contain musical parameters along with given instrumentation information (e.g. bass and guitar notes). Results on generated drum sequences are presented indicating that DeepDrum is effective in producing rhythms that resemble the learned style, while at the same time conforming to given constraints that were unknown during the training process.

We study the problem of visibility-based exploration, reconstruction and surveillance in the context of supervised learning. Using a level set representation of data and information, we train a convolutional neural network to determine vantage points that maximize visibility. We show that this method drastically reduces the on-line computational cost and determines a small set of vantage points that solve the problem. This enables us to efficiently produce highly-resolved and topologically accurate maps of complex 3D environments. We present realistic simulations on 2D and 3D urban environments.

Tasks involving the analysis of geometric (graph- and manifold-structured) data have recently gained prominence in the machine learning community, giving birth to a rapidly developing field of geometric deep learning. In this work, we leverage graph neural networks to improve signal detection in the IceCube neutrino observatory. The IceCube detector array is modeled as a graph, where vertices are sensors and edges are a learned function of the sensors' spatial coordinates. As only a subset of IceCube's sensors is active during a given observation, we note the adaptive nature of our GNN, wherein computation is restricted to the input signal support. We demonstrate the effectiveness of our GNN architecture on a task classifying IceCube events, where it outperforms both a traditional physics-based method as well as classical 3D convolution neural networks.

Sequence-to-sequence (encoder-decoder) models with attention constitute a cornerstone of deep learning research, as they have enabled unprecedented sequential data modeling capabilities. This effectiveness largely stems from the capacity of these models to infer salient temporal dynamics over long horizons; these are encoded into the obtained neural attention (NA) distributions. However, existing NA formulations essentially constitute point-wise selection mechanisms over the observed source sequences; that is, attention weights computation relies on the assumption that each source sequence element is independent of the rest. Unfortunately, although convenient, this assumption fails to account for higher-order dependencies which might be prevalent in real-world data. This paper addresses these limitations by leveraging Quantum-Statistical modeling arguments. Specifically, our work broadens the notion of NA, by attempting to account for the case that the NA model becomes inherently incapable of discerning between individual source elements; this is assumed to be the case due to higher-order temporal dynamics. On the contrary, we postulate that in some cases selection may be feasible only at the level of pairs of source sequence elements. To this end, we cast NA into inference of an attention density matrix (ADM) approximation. We derive effective training and inference algorithms, and evaluate our approach in the context of a machine translation (MT) application. We perform experiments with challenging benchmark datasets. As we show, our approach yields favorable outcomes in terms of several evaluation metrics.

This paper addresses the question of identifying the time-window in short-term past from which the information regarding the future occupant's window opening actions and resulting window states in buildings can be predicted. The addressed sequence duration was in the range between 30 and 240 time-steps of indoor climate data, where the applied temporal discretization was one minute. For that purpose, a deep neural network is trained to predict the window states, where the input sequence duration is handled as an additional hyperparameter. Eventually, the relationship between the prediction accuracy and the time-lag of the predicted window state in future is analyzed. The results pointed out, that the optimal predictive performance was achieved for the case where 60 time-steps of the indoor climate data were used as input. Additionally, the results showed that very long sequences (120-240 time-steps) could be addressed efficiently, given the right hyperprameters. Hence, the use of the memory over previous hours of high-resolution indoor climate data did not improve the predictive performance, when compared to the case where 30/60 minutes indoor sequences were used. The analysis of the prediction accuracy in the form of F1 score for the different time-lag of future window states dropped from 0.51 to 0.27, when shifting the prediction target from 10 to 60 minutes in future.

Black box discrete optimization (BBDO) appears in wide range of engineering tasks. Evolutionary or other BBDO approaches have been applied, aiming at automating necessary tuning of system parameters, such as hyper parameter tuning of machine learning based systems when being installed for a specific task. However, automation is often jeopardized by the need of strategy parameter tuning for BBDO algorithms. An expert with the domain knowledge must undergo time-consuming strategy parameter tuning. This paper proposes a parameterless BBDO algorithm based on information geometric optimization, a recent framework for black box optimization using stochastic natural gradient. Inspired by some theoretical implications, we develop an adaptation mechanism for strategy parameters of the stochastic natural gradient method for discrete search domains. The proposed algorithm is evaluated on commonly used test problems. It is further extended to two examples of simultaneous optimization of the hyper parameters and the connection weights of deep learning models, leading to a faster optimization than the existing approaches without any effort of parameter tuning.

Adversarial machine learning in the context of image processing and related applications has received a large amount of attention. However, adversarial machine learning, especially adversarial deep learning, in the context of malware detection has received much less attention despite its apparent importance. In this paper, we present a framework for enhancing the robustness of Deep Neural Networks (DNNs) against adversarial malware samples, dubbed Hashing Transformation Deep Neural Networks} (HashTran-DNN). The core idea is to use hash functions with a certain locality-preserving property to transform samples to enhance the robustness of DNNs in malware classification. The framework further uses a Denoising Auto-Encoder (DAE) regularizer to reconstruct the hash representations of samples, making the resulting DNN classifiers capable of attaining the locality information in the latent space. We experiment with two concrete instantiations of the HashTran-DNN framework to classify Android malware. Experimental results show that four known attacks can render standard DNNs useless in classifying Android malware, that known defenses can at most defend three of the four attacks, and that HashTran-DNN can effectively defend against all of the four attacks.

In this paper we first present a class of algorithms for training multi-level neural networks with a quadratic cost function one layer at a time starting from the input layer. The algorithm is based on the fact that for any layer to be trained, the effect of a direct connection to an optimized linear output layer can be computed without the connection being made. Thus, starting from the input layer, we can train each layer in succession in isolation from the other layers. Once trained, the weights are kept fixed and the outputs of the trained layer then serve as the inputs to the next layer to be trained. The result is a very fast algorithm. The simplicity of this training arrangement allows the activation function and step size in weight adjustment to be adaptive and self-adjusting. Furthermore, the stability of the training process allows relatively large steps to be taken and thereby achieving in even greater speeds. Finally, in our context configuring the network means determining the number of outputs for each layer. By decomposing the overall cost function into separate components related to approximation and estimation, we obtain an optimization formula for determining the number of outputs for each layer. With the ability to self-configure and set parameters, we now have more than a fast training algorithm, but the ability to build automatically a fully trained deep neural network starting with nothing more than data.

Scattering networks are a class of designed Convolutional Neural Networks (CNNs) with fixed weights. We argue they can serve as generic representations for modelling images. In particular, by working in scattering space, we achieve competitive results both for supervised and unsupervised learning tasks, while making progress towards constructing more interpretable CNNs. For supervised learning, we demonstrate that the early layers of CNNs do not necessarily need to be learned, and can be replaced with a scattering network instead. Indeed, using hybrid architectures, we achieve the best results with predefined representations to-date, while being competitive with end-to-end learned CNNs. Specifically, even applying a shallow cascade of small-windowed scattering coefficients followed by 1$\times$1-convolutions results in AlexNet accuracy on the ILSVRC2012 classification task. Moreover, by combining scattering networks with deep residual networks, we achieve a single-crop top-5 error of 11.4% on ILSVRC2012. Also, we show they can yield excellent performance in the small sample regime on CIFAR-10 and STL-10 datasets, exceeding their end-to-end counterparts, through their ability to incorporate geometrical priors. For unsupervised learning, scattering coefficients can be a competitive representation that permits image recovery. We use this fact to train hybrid GANs to generate images. Finally, we empirically analyze several properties related to stability and reconstruction of images from scattering coefficients.

Abstract--The recent advances of hardware technology have made the intelligent analysis equipped at the front-end with deep learning more prevailing and practical. To better enable the intelligent sensing at the front-end, instead of compressing and transmitting visual signals or the ultimately utilized toplayer deep learning features, we propose to compactly represent and convey the intermediate-layer deep learning features of high generalization capability, to facilitate the collaborating approach between front and cloud ends. This strategy enables a good balance among the computational load, transmission load and the generalization ability for cloud servers when deploying the deep neural networks for large scale cloud based visual analysis. Moreover, the presented strategy also makes the standardization of deep feature coding more feasible and promising, as a series of tasks can simultaneously benefit from the transmitted intermediate layers. We also present the results for evaluation of lossless deep feature compression with four benchmark data compression methods, which provides meaningful investigations and baselines for future research and standardization activities. ECENTLY, deep neural networks (DNNs) have demonstrated the state-of-the-art performance in various computer vision tasks, e.g., image classification [1], [2], [3], [4], image object detection [5], [6], visual tracking [7], visual retrieval [8]. In contrast to the handcrafted features such as Scale-Invariant Feature Transform (SIFT) [9], deep learning based approaches are able to learn representative features directly from the vast amounts of data. For image classification, which is the fundamental task of computer vision, the AlexNet model [1] has achieved 9% better classification accuracy than the previous handcrafted methods in the 2012 ImageNet competition [10], which provides a large scale training dataset with 1.2 million images and one thousand categories. Inspired by the fantastic progress of AlexNet, DNN models continue to be the undisputed leaders in the competition of ImageNet. In particular, both VGGNet [2] and GoogLeNet [11] announced promising performance in the ILSVRC 2014 classification challenge, which demonstrated that deeper and wider architectures can bring great benefits in learning better representations via large scale datasets. In 2016, He et al. also proposed residual blocks to enable very deep learning structure [3]. With the advances of network infrastructure, cloud-based applications are springing up in recent years. In particular, the front-end devices acquire information from users or the physical world, which are subsequently transmitted to the cloud end (i.e., data center) for further process and analyses. In particular, for visual analysis, the front-end devices deployed in the real world such as surveillance cameras and wearable devices acquire massive visual data which are transmitted to the cloud side for analyses, as shown in Figure 1.