Multi-instance learning (MIL) deals with tasks where each example is represented by a bag of instances. Unlike traditional supervised learning, only the bag labels are observed whereas the label for each instance in the bags is not available. Previous MIL studies typically assume that training and the test data follow the same distribution, which is often violated in real-world applications. Existing methods address distribution changes by reweighting the training bags with the density ratio between the test and the training data. However, models are frequently trained without prior knowledge of the testing distribution which renders existing methods ineffective. In this paper, we propose a novel multi-instance learning algorithm which links MIL with causal inference to achieve stable prediction without knowing the distribution of the test dataset. Experimental results show that the performance of our approach is stable to the distribution changes.

We give the first efficient algorithm for learning the structure of an Ising model that tolerates independent failures; that is, each entry of the observed sample is missing with some unknown probability p. Our algorithm matches the essentially optimal runtime and sample complexity bounds of recent work for learning Ising models due to Klivans and Meka (2017). We devise a novel unbiased estimator for the gradient of the Interaction Screening Objective (ISO) due to Vuffray et al. (2016) and apply a stochastic multiplicative gradient descent algorithm to minimize this objective. Solutions to this minimization recover the neighborhood information of the underlying Ising model on a node by node basis.

This paper introduces a new method for model selection and more generally hyperparameter selection in machine learning. The paper first proves a relationship between generalization error and a difference of description lengths of the training data; we call this difference differential description length (DDL). This allows prediction of generalization error from the training data \emph{alone} by performing encoding of the training data. This can now be used for model selection by choosing the model that has the smallest predicted generalization error. We show how this encoding can be done for linear regression and neural networks. We provide experiments showing that this leads to smaller generalization error than cross-validation and traditional MDL and Bayes methods.

Data privacy is an important concern in machine learning, and is fundamentally at odds with the task of training useful learning models, which typically require the acquisition of large amounts of private user data. One possible way of fulfilling the machine learning task while preserving user privacy is to train the model on a transformed, noisy version of the data, which does not reveal the data itself directly to the training procedure. In this work, we analyze the privacy-utility trade-off of two such schemes for the problem of linear regression: additive noise, and random projections. In contrast to previous work, we consider a recently proposed notion of differential privacy that is based on conditional mutual information (MI-DP), which is stronger than the conventional $(\epsilon, \delta)$-differential privacy, and use relative objective error as the utility metric. We find that projecting the data to a lower-dimensional subspace before adding noise attains a better trade-off in general. We also make a connection between privacy problem and (non-coherent) SIMO, which has been extensively studied in wireless communication, and use tools from there for the analysis. We present numerical results demonstrating the performance of the schemes.

Many modern neural network architectures are trained in an overparameterized regime where the parameters of the model exceed the size of the training dataset. Sufficiently overparameterized neural network architectures in principle have the capacity to fit any set of labels including random noise. However, given the highly nonconvex nature of the training landscape it is not clear what level and kind of overparameterization is required for first order methods to converge to a global optima that perfectly interpolate any labels. A number of recent theoretical works have shown that for very wide neural networks where the number of hidden units is polynomially large in the size of the training data gradient descent starting from a random initialization does indeed converge to a global optima. However, in practice much more moderate levels of overparameterization seems to be sufficient and in many cases overparameterized models seem to perfectly interpolate the training data as soon as the number of parameters exceed the size of the training data by a constant factor. Thus there is a huge gap between the existing theoretical literature and practical experiments. In this paper we take a step towards closing this gap. Focusing on shallow neural nets and smooth activations, we show that (stochastic) gradient descent when initialized at random converges at a geometric rate to a nearby global optima as soon as the square-root of the number of network parameters exceeds the size of the training data. Our results also benefit from a fast convergence rate and continue to hold for non-differentiable activations such as Rectified Linear Units (ReLUs).

We prove that α-loss has an equivalent margin-based form and is classification-calibrated, two desirable properties for a good surrogate loss function for the ideal yet intractable 0-1 loss. For logistic regression-based classification, we provide an upper bound on the difference between the empirical and expected risk for α-loss by exploiting its Lipschitzianity along with recent results on the landscape features of empirical risk functions. Finally, we show that α-loss with α 2 performs better than log-loss on MNIST for logistic regression. I. INTRODUCTION In learning theory, the performance of a classification algorithm interms of accuracy, tractability, and convergence guarantees is contingent on the choice of a loss function. Consider a feature vector X X, an unknown finite label Y Y, and a hypothesis test h: X Y. The canonical 0-1 loss, given by 1[h(X) Y ], is considered an ideal loss function that captures the probability of incorrectly guessing the true label Y using h(X).

State-of-the-art models are now trained with billions of parameters, reaching hardware limits in terms of memory consumption. This has created a recent demand for memory-efficient optimizers. To this end, we investigate the limits and performance tradeoffs of memory-efficient adaptively preconditioned gradient methods. We propose extreme tensoring for high-dimensional stochastic optimization, showing that an optimizer needs very little memory to benefit from adaptive preconditioning. Our technique applies to arbitrary models (not necessarily with tensor-shaped parameters), and is accompanied by regret and convergence guarantees, which shed light on the tradeoffs between preconditioner quality and expressivity. On a large-scale NLP model, we reduce the optimizer memory overhead by three orders of magnitude, without degrading performance.

Privacy-preserving data analysis is a rising challenge in contemporary statistics, as the privacy guarantees of statistical methods are often achieved at the expense of accuracy. In this paper, we investigate the tradeoff between statistical accuracy and privacy in mean estimation and linear regression, under both the classical low-dimensional and modern high-dimensional settings. A primary focus is to establish minimax optimality for statistical estimation with the $(\varepsilon,\delta)$-differential privacy constraint. To this end, we find that classical lower bound arguments fail to yield sharp results, and new technical tools are called for. We first develop a general lower bound argument for estimation problems with differential privacy constraints, and then apply the lower bound argument to mean estimation and linear regression. For these statistical problems, we also design computationally efficient algorithms that match the minimax lower bound up to a logarithmic factor. In particular, for the high-dimensional linear regression, a novel private iterative hard thresholding pursuit algorithm is proposed, based on a privately truncated version of stochastic gradient descent. The numerical performance of these algorithms is demonstrated by simulation studies and applications to real data containing sensitive information, for which privacy-preserving statistical methods are necessary.

We consider the two-group classification problem and propose a kernel classifier based on the optimal scoring framework. Unlike previous approaches, we provide theoretical guarantees on the expected risk consistency of the method. We also allow for feature selection by imposing structured sparsity using weighted kernels. We propose fully-automated methods for selection of all tuning parameters, and in particular adapt kernel shrinkage ideas for ridge parameter selection. Numerical studies demonstrate the superior classification performance of the proposed approach compared to existing nonparametric classifiers.

Understanding learning and generalization of deep architectures has been a major research objective in the recent years with notable theoretical progress. A main focal point of generalization studies stems from the success of excessively large networks which defy the classical wisdom of uniform convergence and learnability. We study empirically the layer-wise functional structure of overparameterized deep models. We provide evidence for the heterogeneous characteristic of layers. To do so, we introduce the notion of (post training) re-initialization and re-randomization robustness. We show that layers can be categorized into either `robust' or `critical'. In contrast to critical layers, resetting the robust layers to their initial value has no negative consequence, and in many cases they barely change throughout training. Our study provides further evidence that mere parameter counting or norm accounting is too coarse in studying generalization of deep models, and flatness or robustness analysis of the model parameters needs to respect the network architectures.