While humans can solve a visual puzzle that requires logical reasoning by observing only few samples, it would require training over large amount of data for state-of-the-art deep reasoning models to obtain similar performance on the same task. In this work, we propose to solve such a few-shot (or low-shot) visual reasoning problem, by resorting to analogical reasoning, which is a unique human ability to identify structural or relational similarity between two sets. Specifically, given training and test sets that contain the same type of visual reasoning problems, we extract the structural relationships between elements in both domains, and enforce them to be as similar as possible with analogical learning. We repeatedly apply this process with slightly modified queries of the same problem under the assumption that it does not affect the relationship between a training and a test sample. This allows to learn the relational similarity between the two samples in an effective manner even with a single pair of samples. We validate our method on RAVEN dataset, on which it outperforms state-of-the-art method, with larger gains when the training data is scarce. We further meta-learn our analogical contrastive learning model over the same tasks with diverse attributes, and show that it generalizes to the same visual reasoning problem with unseen attributes.

Designing task-oriented dialogue systems is a challenging research topic, since it needs not only to generate utterances fulfilling user requests but also to guarantee the comprehensibility. Many previous works trained end-to-end (E2E) models with supervised learning (SL), however, the bias in annotated system utterances remains as a bottleneck. Reinforcement learning (RL) deals with the problem through using non-differentiable evaluation metrics (e.g., the success rate) as rewards. Nonetheless, existing works with RL showed that the comprehensibility of generated system utterances could be corrupted when improving the performance on fulfilling user requests. In our work, we (1) propose modelling the hierarchical structure between dialogue policy and natural language generator (NLG) with the option framework, called HDNO; (2) train HDNO with hierarchical reinforcement learning (HRL), as well as suggest alternating updates between dialogue policy and NLG during HRL inspired by fictitious play, to preserve the comprehensibility of generated system utterances while improving fulfilling user requests; and (3) propose using a discriminator modelled with language models as an additional reward to further improve the comprehensibility. We test HDNO on MultiWoz 2.0 and MultiWoz 2.1, the datasets on multi-domain dialogues, in comparison with word-level E2E model trained with RL, LaRL and HDSA, showing a significant improvement on the total performance evaluated with automatic metrics.

Why do agents often obtain better reinforcement learning policies when imitating a worse expert? We show that privileged information used by the expert is marginalized in the learned agent policy, resulting in an "imitation gap." Prior work bridges this gap via a progression from imitation learning to reinforcement learning. While often successful, gradual progression fails for tasks that require frequent switches between exploration and memorization skills. To better address these tasks and alleviate the imitation gap we propose 'Adaptive Insubordination' (ADVISOR), which dynamically reweights imitation and reward-based reinforcement learning losses during training, enabling switching between imitation and exploration. On a suite of challenging tasks, we show that ADVISOR outperforms pure imitation, pure reinforcement learning, as well as sequential combinations of these approaches.

Accurate forecasting is important for decision-makers. Recently, the Amazon rainforest is reaching record levels of the number of fires, a situation that concerns both climate and public health problems. Obtaining the desired forecasting accuracy becomes difficult and challenging. In this paper were developed a novel heterogeneous decomposition-ensemble model by using Seasonal and Trend decomposition based on Loess in combination with algorithms for short-term load forecasting multi-month-ahead, to explore temporal patterns of Amazon rainforest fires in Brazil. The results demonstrate the proposed decomposition-ensemble models can provide more accurate forecasting evaluated by performance measures. Diebold-Mariano statistical test showed the proposed models are better than other compared models, but it is statistically equal to one of them.

A primary difficulty with unsupervised discovery of structure in large data sets is a lack of quantitative evaluation criteria. In this work, we propose and investigate several metrics for evaluating and comparing generative models of behavior learned using imitation learning. Compared to the commonly-used model log-likelihood, these criteria look at longer temporal relationships in behavior, are relevant if behavior has some properties that are inherently unpredictable, and highlight biases in the overall distribution of behaviors produced by the model. Pointwise metrics compare real to model-predicted trajectories given true past information. Distribution metrics compare statistics of the model simulating behavior in open loop, and are inspired by how experimental biologists evaluate the effects of manipulations on animal behavior. We show that the proposed metrics correspond with biologists' intuitions about behavior, and allow us to evaluate models, understand their biases, and enable us to propose new research directions.

Mutual information (MI) minimization has gained considerable interests in various machine learning tasks. However, estimating and minimizing MI in high-dimensional spaces remains a challenging problem, especially when only samples, rather than distribution forms, are accessible. Previous works mainly focus on MI lower bound approximation, which is not applicable to MI minimization problems. In this paper, we propose a novel Contrastive Log-ratio Upper Bound (CLUB) of mutual information. We provide a theoretical analysis of the properties of CLUB and its variational approximation. Based on this upper bound, we introduce a MI minimization training scheme and further accelerate it with a negative sampling strategy. Simulation studies on Gaussian distributions show the reliable estimation ability of CLUB. Real-world MI minimization experiments, including domain adaptation and information bottleneck, demonstrate the effectiveness of the proposed method. The code is at https://github.com/Linear95/CLUB.

In online learning from non-stationary data streams, it is both necessary to learn robustly to outliers and to adapt to changes of underlying data generating mechanism quickly. In this paper, we refer to the former nature of online learning algorithms as robustness and the latter as adaptivity. There is an obvious tradeoff between them. It is a fundamental issue to quantify and evaluate the tradeoff because it provides important information on the data generating mechanism. However, no previous work has considered the tradeoff quantitatively. We propose a novel algorithm called the Stochastic approximation-based Robustness-Adaptivity algorithm (SRA) to evaluate the tradeoff. The key idea of SRA is to update parameters of distribution or sufficient statistics with the biased stochastic approximation scheme, while dropping data points with large values of the stochastic update. We address the relation between two parameters, one of which is the step size of the stochastic approximation, and the other is the threshold parameter of the norm of the stochastic update. The former controls the adaptivity and the latter does the robustness. We give a theoretical analysis for the non-asymptotic convergence of SRA in the presence of outliers, which depends on both the step size and the threshold parameter. Since SRA is formulated on the majorization-minimization principle, it is a general algorithm including many algorithms, such as the online EM algorithm and stochastic gradient descent. Empirical experiments for both synthetic and real datasets demonstrated that SRA was superior to previous methods.

How can we find the right graph for semi-supervised learning? In real world applications, the choice of which edges to use for computation is the first step in any graph learning process. Interestingly, there are often many types of similarity available to choose as the edges between nodes, and the choice of edges can drastically affect the performance of downstream semi-supervised learning systems. However, despite the importance of graph design, most of the literature assumes that the graph is static. In this work, we present Grale, a scalable method we have developed to address the problem of graph design for graphs with billions of nodes. Grale operates by fusing together different measures of(potentially weak) similarity to create a graph which exhibits high task-specific homophily between its nodes. Grale is designed for running on large datasets. We have deployed Grale in more than 20 different industrial settings at Google, including datasets which have tens of billions of nodes, and hundreds of trillions of potential edges to score. By employing locality sensitive hashing techniques,we greatly reduce the number of pairs that need to be scored, allowing us to learn a task specific model and build the associated nearest neighbor graph for such datasets in hours, rather than the days or even weeks that might be required otherwise. We illustrate this through a case study where we examine the application of Grale to an abuse classification problem on YouTube with hundreds of million of items. In this application, we find that Grale detects a large number of malicious actors on top of hard-coded rules and content classifiers, increasing the total recall by 89% over those approaches alone.

We consider the problem of online boosting for regression tasks, when only limited information is available to the learner. We give an efficient regret minimization method that has two implications: an online boosting algorithm with noisy multi-point bandit feedback, and a new projection-free online convex optimization algorithm with stochastic gradient, that improves state-of-the-art guarantees in terms of efficiency.

In this work, two new initialization methods for K-means clustering are proposed. Both proposals are based on applying a divide-and-conquer approach for the K-means|| type of an initialization strategy. The second proposal also utilizes multiple lower-dimensional subspaces produced by the random projection method for the initialization. The proposed methods are scalable and can be run in parallel, which make them suitable for initializing large-scale problems. In the experiments, comparison of the proposed methods to the K-means++ and K-means|| methods is conducted using an extensive set of reference and synthetic large-scale datasets. Concerning the latter, a novel high-dimensional clustering data generation algorithm is given. The experiments show that the proposed methods compare favorably to the state-of-the-art. We also observe that the currently most popular K-means++ initialization behaves like the random one in the very high-dimensional cases.