There is growing interest in multi-label image classification due to its critical role in web-based image analytics-based applications, such as large-scale image retrieval and browsing. Matrix completion has recently been introduced as a method for transductive (semi-supervised) multi-label classification, and has several distinct advantages, including robustness to missing data and background noise in both feature and label space. However, it is limited by only considering data represented by a single-view feature, which cannot precisely characterize images containing several semantic concepts. To utilize multiple features taken from different views, we have to concatenate the different features as a long vector. But this concatenation is prone to over-fitting and often leads to very high time complexity in MC based image classification. Therefore, we propose to weightedly combine the MC outputs of different views, and present the multi-view matrix completion (MVMC) framework for transductive multi-label image classification. To learn the view combination weights effectively, we apply a cross validation strategy on the labeled set. In the learning process, we adopt the average precision (AP) loss, which is particular suitable for multi-label image classification. A least squares loss formulation is also presented for the sake of efficiency, and the robustness of the algorithm based on the AP loss compared with the other losses is investigated. Experimental evaluation on two real world datasets (PASCAL VOC' 07 and MIR Flickr) demonstrate the effectiveness of MVMC for transductive (semi-supervised) multi-label image classification, and show that MVMC can exploit complementary properties of different features and output-consistent labels for improved multi-label image classification.

To obtain interpretable machine learning models, either interpretable models are constructed from the outset - e.g. shallow decision trees, rule lists, or sparse generalized linear models - or post-hoc interpretation methods - e.g. partial dependence or ALE plots - are employed. Both approaches have disadvantages. While the former can restrict the hypothesis space too conservatively, leading to potentially suboptimal solutions, the latter can produce too verbose or misleading results if the resulting model is too complex, especially w.r.t. feature interactions. We propose to make the compromise between predictive power and interpretability explicit by quantifying the complexity / interpretability of machine learning models. Based on functional decomposition, we propose measures of number of features used, interaction strength and main effect complexity. We show that post-hoc interpretation of models that minimize the three measures becomes more reliable and compact. Furthermore, we demonstrate the application of such measures in a multi-objective optimization approach which considers predictive power and interpretability at the same time.

Representation and learning of long-range dependencies is a central challenge confronted in modern applications of machine learning to sequence data. Yet despite the prominence of this issue, the basic problem of measuring long-range dependence, either in a given data source or as represented in a trained deep model, remains largely limited to heuristic tools. We contribute a statistical framework for investigating long-range dependence in current applications of sequence modeling, drawing on the statistical theory of long memory stochastic processes. By analogy with their linear predecessors in the time series literature, we identify recurrent neural networks (RNNs) as nonlinear processes that simultaneously attempt to learn both a feature representation for and the long-range dependency structure of an input sequence. We derive testable implications concerning the relationship between long memory in real-world data and its learned representation in a deep network architecture, which are explored through a semiparametric framework adapted to the high-dimensional setting. We establish the validity of statistical inference for a simple estimator, which yields a decision rule for long memory in RNNs. Experiments illustrating this statistical framework confirm the presence of long memory in a diverse collection of natural language and music data, but show that a variety of RNN architectures fail to capture this property even after training to benchmark accuracy in a language model.

Monte Carlo Tree Search (MCTS) algorithms perform simulation-based search to improve policies online. During search, the simulation policy is adapted to explore the most promising lines of play. MCTS has been used by state-of-the-art programs for many problems, however a disadvantage to MCTS is that it estimates the values of states with Monte Carlo averages, stored in a search tree; this does not scale to games with very high branching factors. We propose an alternative simulation-based search method, Policy Gradient Search (PGS), which adapts a neural network simulation policy online via policy gradient updates, avoiding the need for a search tree. In Hex, PGS achieves comparable performance to MCTS, and an agent trained using Expert Iteration with PGS was able defeat MoHex 2.0, the strongest open-source Hex agent, in 9x9 Hex.

Deep neural networks (DNN) are ubiquitous in a growing number of domains ranging from computer vision to healthcare. State-of-the-art DNN models are typically overparameterized and contain more parameters than the size of the training dataset. It is well understood that in this overparameterized regime, DNNs are highly expressive and have the capacity to (over)fit arbitrary training datasets including pure noise [56]. Mysteriously however neural network models trained via simple algorithms such as stochastic gradient descent continue to predict well on yet unseen test data. In such over-parametrized scenarios there maybe infinitely many globally optimal network parameters consistent with the training data, the key challenge is to understand which network parameters (stochastic) gradient descent converges to and what are its properties. Indeed, a recent series of papers [16, 52, 56], suggest that solutions found by first order methods tend to have favorable generalization properties. As DNNs begin to be deployed in safety critical applications, the need for foundational understanding of their noise robustness and their unique prediction capabilities intensifies. This paper focuses on an intriguing phenomena: overparameterized neural networks are surprisingly robust to label noise when first order methods with early stopping is used to train them [25]. To observe this phenomena consider Figure 1 where we perform experiments on the MNIST data set.

We consider inference about a scalar parameter under a non-parametric model based on a one-step estimator computed as a plug in estimator plus the empirical mean of an estimator of the parameter's influence function. We focus on a class of parameters that have influence function which depends on two infinite dimensional nuisance functions and such that the bias of the one-step estimator of the parameter of interest is the expectation of the product of the estimation errors of the two nuisance functions. Our class includes many important treatment effect contrasts of interest in causal inference and econometrics, such as ATE, ATT, an integrated causal contrast with a continuous treatment, and the mean of an outcome missing not at random. We propose estimators of the target parameter that entertain approximately sparse regression models for the nuisance functions allowing for the number of potential confounders to be even larger than the sample size. By employing sample splitting, cross-fitting and $\ell_1$-regularized regression estimators of the nuisance functions based on objective functions whose directional derivatives agree with those of the parameter's influence function, we obtain estimators of the target parameter with two desirable robustness properties: (1) they are rate doubly-robust in that they are root-n consistent and asymptotically normal when both nuisance functions follow approximately sparse models, even if one function has a very non-sparse regression coefficient, so long as the other has a sufficiently sparse regression coefficient, and (2) they are model doubly-robust in that they are root-n consistent and asymptotically normal even if one of the nuisance functions does not follow an approximately sparse model so long as the other nuisance function follows an approximately sparse model with a sufficiently sparse regression coefficient.

Matrix completion has a long-time history of usage as the core technique of recommender systems. In particular, 1-bit matrix completion, which considers the prediction as a ``Recommended'' or ``Not Recommended'' question, has proved its significance and validity in the field. However, while customers and products aggregate into interacted clusters, state-of-the-art model-based 1-bit recommender systems do not take the consideration of grouping bias. To tackle the gap, this paper introduced Group-Specific 1-bit Matrix Completion (GS1MC) by first-time consolidating group-specific effects into 1-bit recommender systems under the low-rank latent variable framework. Additionally, to empower GS1MC even when grouping information is unobtainable, Cluster Developing Matrix Completion (CDMC) was proposed by integrating the sparse subspace clustering technique into GS1MC. Namely, CDMC allows clustering users/items and to leverage their group effects into matrix completion at the same time. Experiments on synthetic and real-world data show that GS1MC outperforms the current 1-bit matrix completion methods. Meanwhile, it is compelling that CDMC can successfully capture items' genre features only based on sparse binary user-item interactive data. Notably, GS1MC provides a new insight to incorporate and evaluate the efficacy of clustering methods while CDMC can be served as a new tool to explore unrevealed social behavior or market phenomenon.

Relevance vector machine (RVM) can be seen as a probabilistic version of support vector machines which is able to produce sparse solutions by linearly weighting a small number of basis functions instead using all of them. Regardless of a few merits of RVM such as giving probabilistic predictions and relax of parameter tuning, it has poor prediction for test instances that are far away from the relevance vectors. As a solution, we propose a new combination of RVM and k-nearest neighbor (k-NN) rule which resolves this issue with regionally dealing with every test instance. In our settings, we obtain the relevance vectors for each test instance in the local area given by k-NN rule. In this way, relevance vectors are closer and more relevant to the test instance which results in a more accurate model. This can be seen as a piece-wise learner which locally classifies test instances. The model is hence called localized relevance vector machine (LRVM). The LRVM is examined on several datasets of the University of California, Irvine (UCI) repository. Results supported by statistical tests indicate that the performance of LRVM is competitive as compared with a few state-of-the-art classifiers.

Training neural networks is traditionally done by providing a sequence of random mini-batches sampled uniformly from the entire training data. In this work, we analyze the effects of curriculum learning, which involves the dynamic non-uniform sampling of mini-batches, on the training of deep networks, and specifically CNNs trained on image recognition. To employ curriculum learning, the training algorithm must resolve 2 problems: (i) sort the training examples by difficulty; (ii) compute a series of mini-batches that exhibit an increasing level of difficulty. We address challenge (i) using two methods: transfer learning from some competitive "teacher" network, and bootstrapping. We show that both methods show similar benefits in terms of increased learning speed and improved final performance on test data. We address challenge (ii) by investigating different pacing functions to guide the sampling. The empirical investigation includes a variety of network architectures, using images from CIFAR-10, CIFAR-100 and subsets of ImageNet. We conclude with a novel theoretical analysis of curriculum learning, where we show how it effectively modifies the optimization landscape. We then define the concept of an ideal curriculum, and show that under mild conditions it does not change the corresponding global minimum of the optimization function.

We introduce Bayesian least-squares policy iteration (BLSPI), an off-policy, model-free, policy iteration algorithm that uses the Bayesian least-squares temporal-difference (BLSTD) learning algorithm to evaluate policies. An online variant of BLSPI has been also proposed, called randomised BLSPI (RBLSPI), that improves its policy based on an incomplete policy evaluation step. In online setting, the exploration-exploitation dilemma should be addressed as we try to discover the optimal policy by using samples collected by ourselves. RBLSPI exploits the advantage of BLSTD to quantify our uncertainty about the value function. Inspired by Thompson sampling, RBLSPI first samples a value function from a posterior distribution over value functions, and then selects actions based on the sampled value function. The effectiveness and the exploration abilities of RBLSPI are demonstrated experimentally in several environments.