We consider the problem of learning decision rules for prediction with feature budget constraint. In particular, we are interested in pruning an ensemble of decision trees to reduce expected feature cost while maintaining high prediction accuracy for any test example. We propose a novel 0-1 integer program formulation for ensemble pruning. Our pruning formulation is general - it takes any ensemble of decision trees as input. By explicitly accounting for feature-sharing across trees together with accuracy/cost trade-off, our method is able to significantly reduce feature cost by pruning subtrees that introduce more loss in terms of feature cost than benefit in terms of prediction accuracy gain. Theoretically, we prove that a linear programming relaxation produces the exact solution of the original integer program. This allows us to use efficient convex optimization tools to obtain an optimally pruned ensemble for any given budget. Empirically, we see that our pruning algorithm significantly improves the performance of the state of the art ensemble method BudgetRF.

Gaussian Processes (GPs) are widely used tools in statistics, machine learning, robotics, computer vision, and scientific computation. However, despite their popularity, they can be difficult to apply; all but the simplest classification or regression applications require specification and inference over complex covariance functions that do not admit simple analytical posteriors. This paper shows how to embed Gaussian processes in any higher-order probabilistic programming language, using an idiom based on memoization, and demonstrates its utility by implementing and extending classic and state-of-the-art GP applications. The interface to Gaussian processes, called gpmem, takes an arbitrary real-valued computational process as input and returns a statistical emulator that automatically improve as the original process is invoked and its input-output behavior is recorded. The flexibility of gpmem is illustrated via three applications: (i) robust GP regression with hierarchical hyper-parameter learning, (ii) discovering symbolic expressions from time-series data by fully Bayesian structure learning over kernels generated by a stochastic grammar, and (iii) a bandit formulation of Bayesian optimization with automatic inference and action selection. All applications share a single 50-line Python library and require fewer than 20 lines of probabilistic code each.

We propose multi-way, multilingual neural machine translation. The proposed approach enables a single neural translation model to translate between multiple languages, with a number of parameters that grows only linearly with the number of languages. This is made possible by having a single attention mechanism that is shared across all language pairs. We train the proposed multi-way, multilingual model on ten language pairs from WMT'15 simultaneously and observe clear performance improvements over models trained on only one language pair. In particular, we observe that the proposed model significantly improves the translation quality of low-resource language pairs.

The community detection problem refers to finding the underlying communities within a network using only the knowledge of the network topology [16, 31]. This paper considers the following probabilistic model for generating a network with underlying community structures: Suppose that out of a total of n vertices, rK of them are partitioned into r clusters of size K, and the remaining n rK vertices do not belong to any clusters (called outlier vertices); a random graph G is generated based on the cluster structure, where each pair of vertices is connected independently with probability p if they are in the same cluster or q otherwise. In particular, an outlier vertex is connected to any other vertex with probability q. This random graph ensemble is known as the planted cluster model [10] with parameters n, r, K N and p, q [0, 1] such that n rK. In particular, we call p and q the in-cluster and cross-cluster edge density, respectively. The planted cluster model encompasses several classical planted random graph models including planted clique [5], planted coloring [4], planted dense subgraph [6], planted partition [11], and the stochastic block model [23], which have been widely used for studying the community detection and graph partitioning problem (see, e.g., [26, 12, 30, 9] and the references therein). This paper was accepted to IEEE Transactions on Information Theory on January 3, 2016. The material in this paper was presented in part at the IEEE International Symposium on Information Theory, Hong Kong, June, 2015 [20].

This paper considers probabilistic estimation of a low-rank matrix from non-linear element-wise measurements of its elements. We derive the corresponding approximate message passing (AMP) algorithm and its state evolution. Relying on non-rigorous but standard assumptions motivated by statistical physics, we characterize the minimum mean squared error (MMSE) achievable information theoretically and with the AMP algorithm. Unlike in related problems of linear estimation, in the present setting the MMSE depends on the output channel only trough a single parameter - its Fisher information. We illustrate this striking finding by analysis of submatrix localization, and of detection of communities hidden in a dense stochastic block model. For this example we locate the computational and statistical boundaries that are not equal for rank larger than four.

The scalability of submodular optimization methods is critical for their usability in practice. In this paper, we study the reducibility of submodular functions, a property that enables us to reduce the solution space of submodular optimization problems without performance loss. We introduce the concept of reducibility using marginal gains. Then we show that by adding perturbation, we can endow irreducible functions with reducibility, based on which we propose the perturbation-reduction optimization framework. Our theoretical analysis proves that given the perturbation scales, the reducibility gain could be computed, and the performance loss has additive upper bounds. We further conduct empirical studies and the results demonstrate that our proposed framework significantly accelerates existing optimization methods for irreducible submodular functions with a cost of only small performance losses.

The aim of this paper is to provide a new method for learning the relationships between data that have been obtained independently. Unlike existing methods like matching, the proposed technique does not require any contextual information, provided that the dependency between the variables of interest is monotone. It can therefore be easily combined with matching in order to exploit the advantages of both methods. This technique can be described as a mix between quantile matching, and deconvolution. We provide for it a theoretical and an empirical validation.

The spectral $k$-support norm enjoys good estimation properties in low rank matrix learning problems, empirically outperforming the trace norm. Its unit ball is the convex hull of rank $k$ matrices with unit Frobenius norm. In this paper we generalize the norm to the spectral $(k,p)$-support norm, whose additional parameter $p$ can be used to tailor the norm to the decay of the spectrum of the underlying model. We characterize the unit ball and we explicitly compute the norm. We further provide a conditional gradient method to solve regularization problems with the norm, and we derive an efficient algorithm to compute the Euclidean projection on the unit ball in the case $p=\infty$. In numerical experiments, we show that allowing $p$ to vary significantly improves performance over the spectral $k$-support norm on various matrix completion benchmarks, and better captures the spectral decay of the underlying model.

Recent works demonstrated the usefulness of temporal coherence to regularize supervised training or to learn invariant features with deep architectures. In particular, enforcing smooth output changes while presenting temporally-closed frames from video sequences, proved to be an effective strategy. In this paper we prove the efficacy of temporal coherence for semi-supervised incremental tuning. We show that a deep architecture, just mildly trained in a supervised manner, can progressively improve its classification accuracy, if exposed to video sequences of unlabeled data. The extent to which, in some cases, a semi-supervised tuning allows to improve classification accuracy (approaching the supervised one) is somewhat surprising. A number of control experiments pointed out the fundamental role of temporal coherence.

Motivation: Public and private repositories of experimental data are growing to sizes that require dedicated methods for finding relevant data. To improve on the state of the art of keyword searches from annotations, methods for content-based retrieval have been proposed. In the context of gene expression experiments, most methods retrieve gene expression profiles, requiring each experiment to be expressed as a single profile, typically of case vs. control. A more general, recently suggested alternative is to retrieve experiments whose models are good for modelling the query dataset. However, for very noisy and high-dimensional query data, this retrieval criterion turns out to be very noisy as well. Results: We propose doing retrieval using a denoised model of the query dataset, instead of the original noisy dataset itself. To this end, we introduce a general probabilistic framework, where each experiment is modelled separately and the retrieval is done by finding related models. For retrieval of gene expression experiments, we use a probabilistic model called product partition model, which induces a clustering of genes that show similar expression patterns across a number of samples. The suggested metric for retrieval using clusterings is the normalized information distance. Empirical results finally suggest that inference for the full probabilistic model can be approximated with good performance using computationally faster heuristic clustering approaches (e.g. $k$-means). The method is highly scalable and straightforward to apply to construct a general-purpose gene expression experiment retrieval method. Availability: The method can be implemented using standard clustering algorithms and normalized information distance, available in many statistical software packages.