Multi-agent reinforcement learning (MARL) extends (single-agent) reinforcement learning (RL) by introducing additional agents and (potentially) partial observability of the environment. Consequently, algorithms for solving MARL problems incorporate various extensions beyond traditional RL methods, such as a learned communication protocol between cooperative agents that enables exchange of private information or adaptive modeling of opponents in competitive settings. One popular algorithmic construct is a memory mechanism such that an agent's decisions can depend not only upon the current state but also upon the history of observed states and actions. In this paper, we study how a memory mechanism can be useful in environments with different properties, such as observability, internality and presence of a communication channel. Using both prior work and new experiments, we show that a memory mechanism is helpful when learning agents need to model other agents and/or when communication is constrained in some way; however we must to be cautious of agents achieving effective memoryfulness through other means.

Recent research studies revealed that neural networks are vulnerable to adversarial attacks. State-of-the-art defensive techniques add various adversarial examples in training to improve models' adversarial robustness. However, these methods are not universal and can't defend unknown or non-adversarial evasion attacks. In this paper, we analyze the model robustness in the decision space. A feedback learning method is then proposed, to understand how well a model learns and to facilitate the retraining process of remedying the defects. The evaluations according to a set of distance-based criteria show that our method can significantly improve models' accuracy and robustness against different types of evasion attacks. Moreover, we observe the existence of inter-class inequality and propose to compensate it by changing the proportions of examples generated in different classes.

Session-based Recurrent Neural Networks (RNNs) are gaining increasing popularity for recommendation task, due to the high autocorrelation of user's behavior on the latest session and the effectiveness of RNN to capture the sequence order information. However, most existing session-based RNN recommender systems still solely focus on the short-term interactions within a single session and completely discard all the other long-term data across different sessions. While traditional Collaborative Filtering (CF) methods have many advanced research works on exploring long-term dependency, which show great value to be explored and exploited in deep learning models. Therefore, in this paper, we propose ASARS, a novel framework that effectively imports the temporal dynamics methodology in CF into session-based RNN system in DL, such that the temporal info can act as scalable weights by a parallel attentional network. Specifically, we first conduct an extensive data analysis to show the distribution and importance of such temporal interactions data both within sessions and across sessions. And then, our ASARS framework promotes two novel models: (1) an inter-session temporal dynamic model that captures the long-term user interaction for RNN recommender system. We integrate the time changes in session RNN and add user preferences as model drifting; and (2) a novel triangle parallel attention network that enhances the original RNN model by incorporating time information. Such triangle parallel network is also specially designed for realizing data argumentation in sequence-to-scalar RNN architecture, and thus it can be trained very efficiently. Our extensive experiments on four real datasets from different domains demonstrate the effectiveness and large improvement of ASARS for personalized recommendation.

Deep neural networks (DNNs) provide high image classification accuracy, but experience significant performance degradation when perturbation from various sources are present in the input. The lack of resilience to input perturbations makes DNN less reliable for systems interacting with physical world such as autonomous vehicles, robotics, to name a few, where imperfect input is the normal condition. We present a hybrid deep network architecture with spike-assisted contextual information extraction (ScieNet). ScieNet integrates unsupervised learning using spiking neural network (SNN) for unsupervised contextual informationextraction with a back-end DNN trained for classification. The integrated network demonstrates high resilience to input perturbations without relying on prior training on perturbed inputs. We demonstrate ScieNet with different back-end DNNs for image classification using CIFAR dataset considering stochastic (noise) and structured (rain) input perturbations. Experimental results demonstrate significant improvement in accuracy on noisy and rainy images without prior training, while maintaining state-of-the-art accuracy on clean images.

Graph Convolutional Networks (GCNs) have recently become the primary choice for learning from graph-structured data, superseding hash fingerprints in representing chemical compounds. However, GCNs lack the ability to take into account the ordering of node neighbors, even when there is a geometric interpretation of the graph vertices that provides an order based on their spatial positions. To remedy this issue, we propose Geometric Graph Convolutional Network (geo-GCN) which uses spatial features to efficiently learn from graphs that can be naturally located in space. Our contribution is threefold: we propose a GCN-inspired architecture which (i) leverages node positions, (ii) is a proper generalisation of both GCNs and Convolutional Neural Networks (CNNs), (iii) benefits from augmentation which further improves the performance and assures invariance with respect to the desired properties. Empirically, geo-GCN outperforms state-of-the-art graph-based methods on image classification and chemical tasks. Introduction Convolutional Neural Networks (CNNs) outperform humans on visual learning tasks, such as image classification (Krizhevsky, Sutskever, and Hinton 2012), object detection (Seferbekov et al. 2018) or image captioning (Y ang et al. 2017). They have also been successfully applied to text processing (Kim 2014) and time series analysis (Y ang et al. 2015). Nevertheless, CNNs cannot be easily adapted to irregular entities, such as graphs, where data representation is not organised in a grid-like structure. Graph Convolutional Networks (GCNs) attempt to mimic CNNs by operating on spatially close neighbors. Motivated by spectral graph theory, Kipf and Welling (Kipf and Welling 2016) use fixed weights determined by the adjacency matrix of a graph to aggregate labels of the neighbors.

We propose a novel deep learning architecture suitable for the prediction of investor interest for a given asset in a given timeframe. This architecture performs both investor clustering and modelling at the same time. We first verify its superior performance on a simulated scenario inspired by real data and then apply it to a large proprietary database from BNP Paribas Corporate and Institutional Banking.

It has long been suggested that living systems, in particular the brain, may operate near some critical point. Through dynamical stability analysis on various computer vision models, we find direct evidence that optimal deep neural network performance occur near the transition point separating stable and chaotic attractors. In fact modern neural network architectures push the model closer to this edge of chaos during the training process. Our dissection into their fully connected layers reveals that they achieve the stability transition through self-adjusting an oscillation-diffusion process embedded in the weights. Further analogy to the logistic map leads us to believe that the optimality near the edge of chaos is a consequence of maximal diversity of stable states, which maximize the effective expressivity. There has been abundant suggestive evidence that many natural systems operate around the critical point between order and disorder ( 1). In particular the brain activities exhibit various spatiotemporal patterns of scale-invariance, which resemble that of critical phase transitions in statistical mechanics ( 2, 3). On the theoretical front, self-organized criticality ( 4) was proposed to explain the prevalence of scale-invariance in nature.

Regularization methods are often employed in deep learning neural networks (DNNs) to prevent overfitting. For penalty based methods for DNN regularization, typically only convex penalties are considered because of their optimization guarantees. Recent theoretical work have shown that non-convex penalties that satisfy certain regularity conditions are also guaranteed to perform well with standard optimization algorithms. In this paper, we examine new and currently existing non-convex penalties for DNN regularization. We provide theoretical justifications for the new penalties and also assess the performance of all penalties on DNN analysis of real datasets. Introduction The success of DNNs in learning complex relationships between the inputs and outputs may be mainly attributed to multiple nonlinear hidden layers [1,2]. Corresponding author, address: 350 Community Drive, Manhasset, NY 11030. Such large number of parameters gives the method incredible amount of flexibility. However on the downside, this may lead to overfitting the data, especially if the training sample is not large enough.

This success can be just ified based on their universal approximation properties [6], whi ch allow them to approximate any function that associates each train ing set input to its corresponding class. But this is also a double-edg ed sword, as the resulting function may not handle well domain shifts ( i.e., it does not generalize well to previously unseen inputs). Adve rsarial attacks (i.e., imperceptible changes to the input built spe cifically to fool the network function) [7, 8] illustrate the risks of bad generalization. Isotropic noise [9] or corrupted inputs [10] are al so likely to produce similar misclassifications. In applications tha t are very sensitive to errors, such as autonomous vehicles or robotic assisted surgery, robustness to such deviations is a key challenge.

Model compression techniques on Deep Neural Network (DNN) have been widely acknowledged as an effective way to achieve acceleration on a variety of platforms, and DNN weight pruning is a straightforward and effective method. There are currently two mainstreams of pruning methods representing two extremes of pruning regularity: non-structured, fine-grained pruning can achieve high sparsity and accuracy, but is not hardware friendly; structured, coarse-grained pruning exploits hardware-efficient structures in pruning, but suffers from accuracy drop when the pruning rate is high. In this paper, we introduce PCONV, comprising a new sparsity dimension, -- fine-grained pruning patterns inside the coarse-grained structures. PCONV comprises two types of sparsities, Sparse Convolution Patterns (SCP) which is generated from intra-convolution kernel pruning and connectivity sparsity generated from inter-convolution kernel pruning. Essentially, SCP enhances accuracy due to its special vision properties, and connectivity sparsity increases pruning rate while maintaining balanced workload on filter computation. To deploy PCONV, we develop a novel compiler-assisted DNN inference framework and execute PCONV models in real-time without accuracy compromise, which cannot be achieved in prior work. Our experimental results show that, PCONV outperforms three state-of-art end-to-end DNN frameworks, TensorFlow-Lite, TVM, and Alibaba Mobile Neural Network with speedup up to 39.2x, 11.4x, and 6.3x, respectively, with no accuracy loss. Mobile devices can achieve real-time inference on large-scale DNNs.