We propose a multi-resolution convolutional autoencoder (MrCAE) architecture that integrates and leverages three highly successful mathematical architectures: (i) multigrid methods, (ii) convolutional autoencoders and (iii) transfer learning. The method provides an adaptive, hierarchical architecture that capitalizes on a progressive training approach for multiscale spatio-temporal data. This framework allows for inputs across multiple scales: starting from a compact (small number of weights) network architecture and low-resolution data, our network progressively deepens and widens itself in a principled manner to encode new information in the higher resolution data based on its current performance of reconstruction. Basic transfer learning techniques are applied to ensure information learned from previous training steps can be rapidly transferred to the larger network. As a result, the network can dynamically capture different scaled features at different depths of the network. The performance gains of this adaptive multiscale architecture are illustrated through a sequence of numerical experiments on synthetic examples and real-world spatial-temporal data.

Most algorithms for representation learning and link prediction in relational data have been designed for static data. However, the data they are applied to usually evolves with time, such as friend graphs in social networks or user interactions with items in recommender systems. This is also the case for knowledge bases, which contain facts such as (US, has president, B. Obama, [2009-2017]) that are valid only at certain points in time. For the problem of link prediction under temporal constraints, i.e., answering queries such as (US, has president, ?, 2012), we propose a solution inspired by the canonical decomposition of tensors of order 4. We introduce new regularization schemes and present an extension of ComplEx (Trouillon et al., 2016) that achieves state-of-the-art performance. Additionally, we propose a new dataset for knowledge base completion constructed from Wikidata, larger than previous benchmarks by an order of magnitude, as a new reference for evaluating temporal and non-temporal link prediction methods.

Nowadays, the utilization of the ever expanding amount of data has made a huge impact on web technologies while also causing various types of security concerns. On one hand, potential gains are highly anticipated if different organizations could somehow collaboratively share their data for technological improvements. On the other hand, data security concerns may arise for both data holders and data providers due to commercial or sociological concerns. To make a balance between technical improvements and security limitations, we implement secure and scalable protocols for multiple data holders to train linear regression and logistic regression models. We build our protocols based on the secret sharing scheme, which is scalable and efficient in applications. Moreover, our proposed paradigm can be generalized to any secure multiparty training scenarios where only matrix summation and matrix multiplications are used. We demonstrate our approach by experiments which shows the scalability and efficiency of our proposed protocols, and finally present its real-world applications.

Over past years, the philosophy for designing the artificial intelligence algorithms has significantly shifted towards automatically extracting the composable systems from massive data volumes. This paradigm shift has been expedited by the big data booming which enables us to easily access and analyze the highly large data sets. The most well-known class of big data analysis techniques is called deep learning. These models require significant computation power and extremely high memory accesses which necessitate the design of novel approaches to reduce the memory access and improve power efficiency while taking into account the development of domain-specific hardware accelerators to support the current and future data sizes and model structures. The current trends for designing application-specific integrated circuits barely consider the essential requirement for maintaining the complex neural network computation to be resilient in the presence of soft errors. The soft errors might strike either memory storage or combinational logic in the hardware accelerator that can affect the architectural behavior such that the precision of the results fall behind the minimum allowable correctness. In this study, we demonstrate that the impact of soft errors on a customized deep learning algorithm called Binarized Neural Network might cause drastic image misclassification. Our experimental results show that the accuracy of image classifier can drastically drop by 76.70% and 19.25% in lfcW1A1 and cnvW1A1 networks, respectively across CIFAR-10 and MNIST datasets during the fault injection for the worst-case scenarios.

Previous studies have suggested that being imitated by an adult is an effective intervention with children with autism and developmental delay. The purpose of this study is to investigate if an imitation game with a robot can arise interest from children and constitute an effective tool to be used in clinical activities. In this paper, we describe the design of our nursery rhyme imitation game, its implementation based on RGB image pose recognition and the preliminary tests we performed.

Everybody loves Occam's Razor, the heuristic often phrased as "When in doubt, choose the simplest option," and elegantly expressed by Albert Einstein via his maxim that theories should be "As simple as possible, but no simpler." This sort of advice sounds intuitively sensible, but without some precise understanding of what "simplicity" means, it's not particularly crisp guidance. My own interest in Occam's Razor arises largely from my work in artificial intelligence. A host of theorists have argued for Occam's central role in AI - going back to Ray Solomonoff in the late 1960s, whose theory of "Solomonoff induction" involves, essentially, AIs that understand the world via choosing the hypothesis represented by the shortest computer program [Sol64]. Marcus Hutter [Hut05] has built a rigorous theory of general intelligence under infinite or near-infinite computing resources, founded on this idea; and Eric Baum has argued the merits of similar ideas from a broad conceptual perspective [Bau04]. Occam's Razor has also been considered foundational in the philosophy of science, by many different thinkers [Gau03]. There is a lot of power in the idea that complex hypotheses, like the Ptolemaic epicycles, have been systematically cast aside in favor of simpler, more compact hypotheses like the Copernican model. However, all these applications of Occam's Razor either rely on very specialized formalizations of the "simplicity" concept (e.g.

What kind of programming language would be most appropriate to serve the needs of integrative, multi-paradigm, multi-software-system approaches to AGI? This question is broached via exploring the more particular question of how to create a more scalable and usable version of the "Atomese" programming language that forms a key component of the OpenCog AGI design (an "Atomese 2.0") . It is tentatively proposed that the core of Atomese 2.0 should be a very flexible framework of rewriting rules for rewriting a metagraph (where the rules themselves are represented within the same metagraph, and some of the intermediate data created and used during the rule-interpretation process may be represented in the same metagraph). This framework should support concurrent rewriting of the metagraph according to rules that are labeled with various sorts of uncertainty-quantifications, and that are labeled with various sorts of types associated with various type systems. A gradual typing approach should be used to enable mixture of rules and other metagraph nodes/links associated with various type systems, and untyped metagraph nodes/links not associated with any type system. This must be done in a way that allows reasonable efficiency and scalability, including in concurrent and distributed processing contexts, in the case where a large percentage of of processing time is occupied with evaluating static pattern-matching queries on specific subgraphs of a large metagraph (including a rich variety of queries such as matches against nodes representing variables, and matches against whole subgraphs, etc.).

A variety of natural language tasks require processing of textual data which contains a mix of natural language and formal languages such as mathematical expressions. In this paper, we take unit conversions as an example and propose a data augmentation technique which leads to models learning both translation and conversion tasks as well as how to adequately switch between them for end-to-end localization.

We focus on scenarios in which multiple rational, non-communicating agents navigate in close proximity, such as unsignalized street intersections. In these situations, decentralized coordination to achieve safe and efficient motion demands nuanced implicit communication between the agents. Often, the spatial structure of such environments constrains multi-agent trajectories to belong to a finite set of modes, each corresponding to a distinct spatiotemporal topology. Our key insight is that empowering agents with a model of this structure can enable effective coordination through implicit communication, realized via intent signals encoded in agents' actions. In this paper, we do so by representing modes of joint behavior as topological braids. We design a decentralized planning framework that incorporates a mechanism for inferring the emerging braid in the decision-making process. By executing actions that minimize the uncertainty over the upcoming braid, agents converge rapidly to a consensus over a joint collision avoidance strategy despite lacking knowledge of the destinations of others and accurate models of their behaviors. We validate our approach with a case study in a four-way unsignalized intersection involving a series of challenging multi-agent scenarios. Our findings show that our model reduces frequency of collisions by at least 65% over a set of explicit trajectory prediction baselines, while maintaining comparable efficiency.

This work presents a modular and hierarchical approach to learn policies for exploring 3D environments, called `Active Neural SLAM'. Our approach leverages the strengths of both classical and learning-based methods, by using analytical path planners with learned SLAM module, and global and local policies. The use of learning provides flexibility with respect to input modalities (in the SLAM module), leverages structural regularities of the world (in global policies), and provides robustness to errors in state estimation (in local policies). Such use of learning within each module retains its benefits, while at the same time, hierarchical decomposition and modular training allow us to sidestep the high sample complexities associated with training end-to-end policies. Our experiments in visually and physically realistic simulated 3D environments demonstrate the effectiveness of our approach over past learning and geometry-based approaches. The proposed model can also be easily transferred to the PointGoal task and was the winning entry of the CVPR 2019 Habitat PointGoal Navigation Challenge.