Machine learning has become omnipresent with applications in various safety-critical domains such as medical, law, and transportation. In these domains, high-stake decisions provided by machine learning necessitate researchers to design interpretable models, where the prediction is understandable to a human. In interpretable machine learning, rule-based classifiers are particularly effective in representing the decision boundary through a set of rules comprising input features. Examples of such classifiers include decision trees, decision lists, and decision sets. The interpretability of rule-based classifiers is in general related to the size of the rules, where smaller rules are considered more interpretable. To learn such a classifier, the brute-force direct approach is to consider an optimization problem that tries to learn the smallest classification rule that has close to maximum accuracy. This optimization problem is computationally intractable due to its combinatorial nature and thus, the problem is not scalable in large datasets. To this end, in this paper we study the triangular relationship among the accuracy, interpretability, and scalability of learning rule-based classifiers. The contribution of this paper is an interpretable learning framework IMLI, that is based on maximum satisfiability (MaxSAT) for synthesizing classification rules expressible in proposition logic. IMLI considers a joint objective function to optimize the accuracy and the interpretability of classification rules and learns an optimal rule by solving an appropriately designed MaxSAT query. Despite the progress of MaxSAT solving in the last decade, the straightforward MaxSAT-based solution cannot scale to practical classification datasets containing thousands to millions of samples. Therefore, we incorporate an efficient incremental learning technique inside the MaxSAT formulation by integrating mini-batch learning and iterative rule-learning. The resulting framework learns a classifier by iteratively covering the training data, wherein in each iteration, it solves a sequence of smaller MaxSAT queries corresponding to each mini-batch. In our experiments, IMLI achieves the best balance among prediction accuracy, interpretability, and scalability. For instance, IMLI attains a competitive prediction accuracy and interpretability w.r.t. existing interpretable classifiers and demonstrates impressive scalability on large datasets where both interpretable and non-interpretable classifiers fail. As an application, we deploy IMLI in learning popular interpretable classifiers such as decision lists and decision sets. The source code is available at https://github.com/meelgroup/mlic.

In recent years exemplar clustering has become a popular tool for applications in document and video summarization, active learning, and clustering with general similarity, where cluster centroids are required to be a subset of the data samples rather than their linear combinations. The problem is also well-known as facility location in the operations research literature. While the problem has well-developed convex relaxation with approximation and recovery guarantees, it has number of variables grows quadratically with the number of samples. Therefore, state-of-the-art methods can hardly handle more than 10^4 samples (i.e. 10^8 variables). In this work, we propose an Augmented-Lagrangian with Block Coordinate Descent (AL-BCD) algorithm that utilizes problem structure to obtain closed-form solution for each block sub-problem, and exploits low-rank representation of the dissimilarity matrix to search active columns without computing the entire matrix. Experiments show our approach to be orders of magnitude faster than existing approaches and can handle problems of up to 10^6 samples. We also demonstrate successful applications of the algorithm on world-scale facility location, document summarization and active learning.

Logarithms of determinants of large positive definite matrices appear ubiquitously in machine learning applications including Gaussian graphical and Gaussian process models, partition functions of discrete graphical models, minimum-volume ellipsoids and metric and kernel learning. Log-determinant computation involves the Cholesky decomposition at the cost cubic in the number of variables (i.e., the matrix dimension), which makes it prohibitive for large-scale applications. We propose a linear-time randomized algorithm to approximate log-determinants for very large-scale positive definite and general non-singular matrices using a stochastic trace approximation, called the Hutchinson method, coupled with Chebyshev polynomial expansions that both rely on efficient matrix-vector multiplications. We establish rigorous additive and multiplicative approximation error bounds depending on the condition number of the input matrix. In our experiments, the proposed algorithm can provide very high accuracy solutions at orders of magnitude faster time than the Cholesky decomposition and Shur completion, and enables us to compute log-determinants of matrices involving tens of millions of variables.

Sparse reconstruction approaches using the re-weighted ℓ1-penalty have been shown, both empirically and theoretically, to provide a significant improvement in recovering sparse signals in comparison to the ℓ1-relaxation. However, numerical optimization of such penalties involves solving problems with ℓ1-norms in the objective many times. Using the direct link of reweighted ℓ1-penalties to the concave log-regularizer for sparsity, we derive a simple proximal-like algorithm for the log-regularized formulation. The proximal splitting step of the algorithm has a closed form solution, and we call the algorithm log-thresholding in analogy to soft thresholding for the ℓ1-penalty. We establish convergence results, and demonstrate that log-thresholding provides more accurate sparse reconstructions compared to both soft and hard thresholding. Furthermore, the approach can be directly extended to optimization over matrices with penalty for rank (i.e. the nuclear norm penalty and its re-weighted version), where we suggest a singular-value log-thresholding approach.

We study the total least squares (TLS) problem that generalizes least squares regression by allowing measurement errors in both dependent and independent variables. TLS is widely used in applied fields including computer vision, system identification and econometrics. The special case when all dependent and independent variables have the same level of uncorrelated Gaussian noise, known as ordinary TLS, can be solved by singular value decomposition (SVD). However, SVD cannot solve many important practical TLS problems with realistic noise structure, such as having varying measurement noise, known structure on the errors, or large outliers requiring robust error-norms. To solve such problems, we develop convex relaxation approaches for a general class of structured TLS (STLS). We show both theoretically and experimentally, that while the plain nuclear norm relaxation incurs large approximation errors for STLS, the re-weighted nuclear norm approach is very effective, and achieves better accuracy on challenging STLS problems than popular non-convex solvers. We describe a fast solution based on augmented Lagrangian formulation, and apply our approach to an important class of biological problems that use population average measurements to infer cell-type and physiological-state specific expression levels that are very hard to measure directly.

We describe the architecture for our distributed video coding (DVC) system. Some key differences between our work and previous systems include a new method of enabling decoder motion compensation, and the use of serially concatenated accumulate syndrome codes for distributed source coding. To evaluate performance, we compare our system to the H.263+ and H.264/AVC video codecs. Experiments show that our system is comparable to DVC systems from Stanford and Berkeley in the sense that our system performs better than H.263+Intra, but worse than H.263+Inter and H.264/AVC

We describe a practical method for distributed compression of q-ary sources using multi-level serially concatenated-accumulate codes. Our approach works well at high compression rates, and allows for graceful and incremental rate-adaptivity. Simulations show that the compression efficiency is near the information-theoretic limits for correlations between sources that obey a Gaussian or Laplacian distribution.