In many domains, there is significant interest in capturing novel relationships between time series that represent activities recorded at different nodes of a highly complex system. In this paper, we introduce multipoles, a novel class of linear relationships between more than two time series. A multipole is a set of time series that have strong linear dependence among themselves, with the requirement that each time series makes a significant contribution to the linear dependence. We demonstrate that most interesting multipoles can be identified as cliques of negative correlations in a correlation network. Such cliques are typically rare in a real-world correlation network, which allows us to find almost all multipoles efficiently using a clique-enumeration approach. Using our proposed framework, we demonstrate the utility of multipoles in discovering new physical phenomena in two scientific domains: climate science and neuroscience. In particular, we discovered several multipole relationships that are reproducible in multiple other independent datasets and lead to novel domain insights.

Communication structure plays a key role in the learning capability of decentralized systems. Structural self-adaptation, by means of self-organization, changes the order as well as the input information of the agents' collective decision-making. This paper studies the role of agents' repositioning on the same communication structure, i.e. a tree, as the means to expand the learning capacity in complex combinatorial optimization problems, for instance, load-balancing power demand to prevent blackouts or efficient utilization of bike sharing stations. The optimality of structural self-adaptations is rigorously studied by constructing a novel large-scale benchmark that consists of 4000 agents with synthetic and real-world data performing 4 million structural self-adaptations during which almost 320 billion learning messages are exchanged. Based on this benchmark dataset, 124 deterministic structural criteria, applied as learning meta-features, are systematically evaluated as well as two online structural self-adaptation strategies designed to expand learning capacity. Experimental evaluation identifies metrics that capture agents with influential information and their optimal positioning. Significant gain in learning performance is observed for the two strategies especially under low-performing initialization. Strikingly, the strategy that triggers structural self-adaptation in a more exploratory fashion is the most cost-effective.

This work tackles the problem of robust zero-shot planning in non-stationary stochastic environments. We study Markov Decision Processes (MDPs) evolving over time and consider Model-Based Reinforcement Learning algorithms in this setting. We make two hypotheses: 1) the environment evolves continuously and its evolution rate is bounded, 2) a current model is known at each decision epoch but not its evolution. Our contribution can be presented in four points. First, we define this specific class of MDPs that we call Non-Stationary MDPs (NSMDPs). We introduce the notion of regular evolution by making an hypothesis of Lipschitz-Continuity on the transition and reward functions w.r.t. time. Secondly, we consider a planning agent using the current model of the environment, but unaware of its future evolution. This leads us to consider a worst-case method where the environment is seen as an adversarial agent. Third, following this approach, we propose the Risk-Averse Tree-Search (RATS) algorithm. This is a zero-shot Model-Based method similar to Minimax search. Finally, we illustrate the benefits brought by RATS empirically and compare its performance with reference Model-Based algorithms.

We study online convex optimization under stochastic sub-gradient observation faults, where we introduce adaptive algorithms with minimax optimal regret guarantees. We specifically study scenarios where our sub-gradient observations can be noisy or even completely missing in a stochastic manner. To this end, we propose algorithms based on sub-gradient descent method, which achieve tight minimax optimal regret bounds. When necessary, these algorithms utilize properties of the underlying stochastic settings to optimize their learning rates (step sizes). These optimizations are the main factor in providing the minimax optimal performance guarantees, especially when observations are stochastically missing. However, in real world scenarios, these properties of the underlying stochastic settings may not be revealed to the optimizer. For such a scenario, we propose a blind algorithm that estimates these properties empirically in a generally applicable manner. Through extensive experiments, we show that this empirical approach is a natural combination of regular stochastic gradient descent and the minimax optimal algorithms (which work best for randomized and adversarial function sequences, respectively).

There are many problems in machine learning and data mining which are equivalent to selecting a non-redundant, high "quality" set of objects. Recommender systems, feature selection, and data summarization are among many applications of this. In this paper, we consider this problem as an optimization problem that seeks to maximize the sum of a sum-sum diversity function and a non-negative monotone submodular function. The diversity function addresses the redundancy, and the submodular function controls the predictive quality. We consider the problem in big data settings (in other words, distributed and streaming settings) where the data cannot be stored on a single machine or the process time is too high for a single machine. We show that a greedy algorithm achieves a constant factor approximation of the optimal solution in these settings. Moreover, we formulate the multi-label feature selection problem as such an optimization problem. This formulation combined with our algorithm leads to the first distributed multi-label feature selection method. We compare the performance of this method with centralized multi-label feature selection methods in the literature, and we show that its performance is comparable or in some cases is even better than current centralized multi-label feature selection methods.

The procedural generation of levels and content in video games is a challenging AI problem. Often such generation relies on an intelligent way of evaluating the content being generated so that constraints are satisfied and/or objectives maximized. In this work, we address the problem of creating levels that are not only playable but also revolve around specific mechanics in the game. We use constrained evolutionary algorithms and quality-diversity algorithms to generate small sections of Super Mario Bros levels called scenes, using three different simulation approaches: Limited Agents, Punishing Model, and Mechanics Dimensions. All three approaches are able to create scenes that give opportunity for a player to encounter or use targeted mechanics with different properties. We conclude by discussing the advantages and disadvantages of each approach and compare them to each other.

In many cases an intelligent agent may want to learn how to mimic a single observed demonstrated trajectory. In this work we consider how to perform such procedural learning from observation, which could help to enable agents to better use the enormous set of video data on observation sequences. Our approach exploits the properties of this setting to incrementally build an open loop action plan that can yield the desired subsequence, and can be used in both Markov and partially observable Markov domains. In addition, procedures commonly involve repeated extended temporal action subsequences. Our method optimistically explores actions to leverage potential repeated structure in the procedure. In comparing to some state-of-the-art approaches we find that our explicit procedural learning from observation method is about 100 times faster than policy-gradient based approaches that learn a stochastic policy and is faster than model based approaches as well. We also find that performing optimistic action selection yields substantial speed ups when latent dynamical structure is present.

Good parameter settings are crucial to achieve high performance in many areas of artificial intelligence (AI), such as propositional satisfiability solving, AI planning, scheduling, and machine learning (in particular deep learning). Automated algorithm configuration methods have recently received much attention in the AI community since they replace tedious, irreproducible and error-prone manual parameter tuning and can lead to new state-of-the-art performance. However, practical applications of algorithm configuration are prone to several (often subtle) pitfalls in the experimental design that can render the procedure ineffective. We identify several common issues and propose best practices for avoiding them. As one possibility for automatically handling as many of these as possible, we also propose a tool called GenericWrapper4AC.

We propose a method for constructing artificial intelligence (AI) of mahjong, which is a multiplayer imperfect information game. Since the size of the game tree is huge, constructing an expert-level AI player of mahjong is challenging. We define multiple Markov decision processes (MDPs) as abstractions of mahjong to construct effective search trees. We also introduce two methods of inferring state values of the original mahjong using these MDPs. We evaluated the effectiveness of our method using gameplays vis-\`{a}-vis the current strongest AI player.

Globally normalized neural sequence models are considered superior to their locally normalized equivalents because they may ameliorate the effects of label bias. However, when considering high-capacity neural parametrizations that condition on the whole input sequence, both model classes are theoretically equivalent in terms of the distributions they are capable of representing. Thus, the practical advantage of global normalization in the context of modern neural methods remains unclear. In this paper, we attempt to shed light on this problem through an empirical study. We extend an approach for search-aware training via a continuous relaxation of beam search (Goyal et al., 2017b) in order to enable training of globally normalized recurrent sequence models through simple backpropagation. We then use this technique to conduct an empirical study of the interaction between global normalization, high-capacity encoders, and search-aware optimization. We observe that in the context of inexact search, globally normalized neural models are still more effective than their locally normalized counterparts. Further, since our training approach is sensitive to warm-starting with pre-trained models, we also propose a novel initialization strategy based on self-normalization for pre-training globally normalized models. We perform analysis of our approach on two tasks: CCG supertagging and Machine Translation, and demonstrate the importance of global normalization under different conditions while using search-aware training.