We develop a theory of online learning by defining several complexity measures. Among them are analogues of Rademacher complexity, covering numbers and fat-shattering dimension from statistical learning theory. Relationship among these complexity measures, their connection to online learning, and tools for bounding them are provided. We apply these results to various learning problems. We provide a complete characterization of online learnability in the supervised setting.

In discriminative machine learning one is interested in training a system to optimize a certain desired measure of performance, or loss. In binary classification one typically tries to minimizes the error rate. But in structured prediction each task often has its own measure of performance such as the BLEU score in machine translation or the intersection-over-union score in PASCAL segmentation. The most common approaches to structured prediction, structural SVMs and CRFs, do not minimize the task loss: the former minimizes a surrogate loss with no guarantees for task loss and the latter minimizes log loss independent of task loss. The main contribution of this paper is a theorem stating that a certain perceptron-like learning rule, involving features vectors derived from loss-adjusted inference, directly corresponds to the gradient of task loss. We give empirical results on phonetic alignment of a standard test set from the TIMIT corpus, which surpasses all previously reported results on this problem.

We present a model that describes the structure in the responses of different brain areas to a set of stimuli in terms of stimulus categories" (clusters of stimuli) and "functional units" (clusters of voxels). We assume that voxels within a unit respond similarly to all stimuli from the same category, and design a nonparametric hierarchical model to capture inter-subject variability among the units. The model explicitly captures the relationship between brain activations and fMRI time courses. A variational inference algorithm derived based on the model can learn categories, units, and a set of unit-category activation probabilities from data. When applied to data from an fMRI study of object recognition, the method finds meaningful and consistent clusterings of stimuli into categories and voxels into units."

We introduce a stochastic process with Wishart marginals: the generalised Wishart process (GWP). It is a collection of positive semi-definite random matrices indexed by any arbitrary dependent variable. We use it to model dynamic (e.g. time varying) covariance matrices. Unlike existing models, it can capture a diverse class of covariance structures, it can easily handle missing data, the dependent variable can readily include covariates other than time, and it scales well with dimension; there is no need for free parameters, and optional parameters are easy to interpret. We describe how to construct the GWP, introduce general procedures for inference and predictions, and show that it outperforms its main competitor, multivariate GARCH, even on financial data that especially suits GARCH. We also show how to predict the mean of a multivariate process while accounting for dynamic correlations.

Singular Value Decomposition (and Principal Component Analysis) is one of the most widely used techniques for dimensionality reduction: successful and efficiently computable, it is nevertheless plagued by a well-known, well-documented sensitivity to outliers. Recent work has considered the setting where each point has a few arbitrarily corrupted components. Yet, in applications of SVD or PCA such as robust collaborative filtering or bioinformatics, malicious agents, defective genes, or simply corrupted or contaminated experiments may effectively yield entire points that are completely corrupted. We present an efficient convex optimization-based algorithm we call Outlier Pursuit, that under some mild assumptions on the uncorrupted points (satisfied, e.g., by the standard generative assumption in PCA problems) recovers the exact optimal low-dimensional subspace, and identifies the corrupted points. Such identification of corrupted points that do not conform to the low-dimensional approximation, is of paramount interest in bioinformatics and financial applications, and beyond. Our techniques involve matrix decomposition using nuclear norm minimization, however, our results, setup, and approach, necessarily differ considerably from the existing line of work in matrix completion and matrix decomposition, since we develop an approach to recover the correct column space of the uncorrupted matrix, rather than the exact matrix itself. In any problem where one seeks to recover a structure rather than the exact initial matrices, techniques developed thus far relying on certificates of optimality, will fail. We present an important extension of these methods, that allows the treatment of such problems.

Software cost estimation is one of the prerequisite managerial activities carried out at the software development initiation stages and also repeated throughout the whole software life-cycle so that amendments to the total cost are made. In software cost estimation typically, a selection of project attributes is employed to produce effort estimations of the expected human resources to deliver a software product. However, choosing the appropriate project cost drivers in each case requires a lot of experience and knowledge on behalf of the project manager which can only be obtained through years of software engineering practice. A number of studies indicate that popular methods applied in the literature for software cost estimation, such as linear regression, are not robust enough and do not yield accurate predictions. Recently the dual variables Ridge Regression (RR) technique has been used for effort estimation yielding promising results. In this work we show that results may be further improved if an AI method is used to automatically select appropriate project cost drivers (inputs) for the technique. We propose a hybrid approach combining RR with a Genetic Algorithm, the latter evolving the subset of attributes for approximating effort more accurately. The proposed hybrid cost model has been applied on a widely known high-dimensional dataset of software project samples and the results obtained show that accuracy may be increased if redundant attributes are eliminated.

Intrusion detection systems (IDSs) fall into two high-level categories: network-based systems (NIDS) that monitor network behaviors, and host-based systems (HIDS) that monitor system calls. In this work, we present a general technique for both systems. We use anomaly detection, which identifies patterns not conforming to a historic norm. In both types of systems, the rates of change vary dramatically over time (due to burstiness) and over components (due to service difference). To efficiently model such systems, we use continuous time Bayesian networks (CTBNs) and avoid specifying a fixed update interval common to discrete-time models. We build generative models from the normal training data, and abnormal behaviors are flagged based on their likelihood under this norm. For NIDS, we construct a hierarchical CTBN model for the network packet traces and use Rao-Blackwellized particle filtering to learn the parameters. We illustrate the power of our method through experiments on detecting real worms and identifying hosts on two publicly available network traces, the MAWI dataset and the LBNL dataset. For HIDS, we develop a novel learning method to deal with the finite resolution of system log file time stamps, without losing the benefits of our continuous time model. We demonstrate the method by detecting intrusions in the DARPA 1998 BSM dataset.

We consider adaptive system identification problems with convex constraints and propose a family of regularized Least-Mean-Square (LMS) algorithms. We show that with a properly selected regularization parameter the regularized LMS provably dominates its conventional counterpart in terms of mean square deviations. We establish simple and closed-form expressions for choosing this regularization parameter. For identifying an unknown sparse system we propose sparse and group-sparse LMS algorithms, which are special examples of the regularized LMS family. Simulation results demonstrate the advantages of the proposed filters in both convergence rate and steady-state error under sparsity assumptions on the true coefficient vector.

We consider the problem of estimating the inverse covariance matrix by maximizing the likelihood function with a penalty added to encourage the sparsity of the resulting matrix. We propose a new approach based on the split Bregman method to solve the regularized maximum likelihood estimation problem. We show that our method is significantly faster than the widely used graphical lasso method, which is based on blockwise coordinate descent, on both artificial and real-world data. More importantly, different from the graphical lasso, the split Bregman based method is much more general, and can be applied to a class of regularization terms other than the $\ell_1$ norm

We study the problem of estimating a temporally varying coefficient and varying structure (VCVS) graphical model underlying nonstationary time series data, such as social states of interacting individuals or microarray expression profiles of gene networks, as opposed to i.i.d. data from an invariant model widely considered in current literature of structural estimation. In particular, we consider the scenario in which the model evolves in a piece-wise constant fashion. We propose a procedure that minimizes the so-called TESLA loss (i.e., temporally smoothed L1 regularized regression), which allows jointly estimating the partition boundaries of the VCVS model and the coefficient of the sparse precision matrix on each block of the partition. A highly scalable proximal gradient method is proposed to solve the resultant convex optimization problem; and the conditions for sparsistent estimation and the convergence rate of both the partition boundaries and the network structure are established for the first time for such estimators.