Recently, there has been growing interest in estimating optimal treatment regimes which are individualized decision rules that can achieve maximal average outcomes. This paper considers the problem of inference for optimal treatment regimes in the model-free setting, where the specification of an outcome regression model is not needed. Existing model-free estimators are usually not suitable for the purpose of inference because they either have nonstandard asymptotic distributions, or are designed to achieve fisher-consistent classification performance. This paper first studies a smoothed robust estimator that directly targets estimating the parameters corresponding to the Bayes decision rule for estimating the optimal treatment regime. This estimator is shown to have an asymptotic normal distribution. Furthermore, it is proved that a resampling procedure provides asymptotically accurate inference for both the parameters indexing the optimal treatment regime and the optimal value function. A new algorithm is developed to calculate the proposed estimator with substantially improved speed and stability. Numerical results demonstrate the satisfactory performance of the new methods.

In the last two years, more than 200 papers have been written on how machine learning (ML) systems can fail because of adversarial attacks on the algorithms and data; this number balloons if we were to incorporate papers covering non-adversarial failure modes. The spate of papers has made it difficult for ML practitioners, let alone engineers, lawyers, and policymakers, to keep up with the attacks against and defenses of ML systems. However, as these systems become more pervasive, the need to understand how they fail, whether by the hand of an adversary or due to the inherent design of a system, will only become more pressing. In order to equip software developers, security incident responders, lawyers, and policy makers with a common vernacular to talk about this problem, we developed a framework to classify failures into "Intentional failures" where the failure is caused by an active adversary attempting to subvert the system to attain her goals; and "Unintentional failures" where the failure is because an ML system produces an inherently unsafe outcome. After developing the initial version of the taxonomy last year, we worked with security and ML teams across Microsoft, 23 external partners, standards organization, and governments to understand how stakeholders would use our framework. Throughout the paper, we attempt to highlight how machine learning failure modes are meaningfully different from traditional software failures from a technology and policy perspective.

Learning performance can show non-monotonic behavior. That is, more data does not necessarily lead to better models, even on average. We propose three algorithms that take a supervised learning model and make it perform more monotone. We prove consistency and monotonicity with high probability, and evaluate the algorithms on scenarios where non-monotone behaviour occurs. Our proposed algorithm $\text{MT}_{\text{HT}}$ makes less than $1\%$ non-monotone decisions on MNIST while staying competitive in terms of error rate compared to several baselines.

Currently, the diagnosis of Autism Spectrum Disorder (ASD) is dependent upon a subjective, time-consuming evaluation of behavioral tests by an expert clinician. Non-invasive functional MRI (fMRI) characterizes brain connectivity and may be used to inform diagnoses and democratize medicine. However, successful construction of deep learning models from fMRI requires addressing key choices about the model's architecture, including the number of layers and number of neurons per layer. Meanwhile, deriving functional connectivity (FC) features from fMRI requires choosing an atlas with an appropriate level of granularity. Once a model has been built, it is vital to determine which features are predictive of ASD and if similar features are learned across atlas granularity levels. To identify aptly suited architectural configurations, probability distributions of the configurations of high versus low performing models are compared. To determine the effect of atlas granularity, connectivity features are derived from atlases with 3 levels of granularity and important features are ranked with permutation feature importance. Results show the highest performing models use between 2-4 hidden layers and 16-64 neurons per layer, granularity dependent. Connectivity features identified as important across all 3 atlas granularity levels include FC to the supplementary motor gyrus and language association cortex, regions associated with deficits in social and sensory processing in ASD. Importantly, the cerebellum, often not included in functional analyses, is also identified as a region whose abnormal connectivity is highly predictive of ASD. Results of this study identify important regions to include in future studies of ASD, help assist in the selection of network architectures, and help identify appropriate levels of granularity to facilitate the development of accurate diagnostic models of ASD.

The information bottleneck (IB) problem tackles the issue of obtaining relevant compressed representations T of some random variable X for the task of predicting Y. It is defined as a constrained optimization problem which maximizes the information the representation has about the task, I(T;Y), while ensuring that a minimum level of compression r is achieved; i.e., I(X;T) <= r. For practical reasons the problem is usually solved by maximizing the IB Lagrangian for many values of the Lagrange multiplier, therefore drawing the IB curve (i.e., the curve of maximal I(T;Y) for a given I(X;T)) and selecting the representation of desired predictability and compression. It is known when Y is a deterministic function of X, the IB curve cannot be explored, and other Lagrangians have been proposed to tackle this problem; e.g., the squared IB Lagrangian. In this paper, we (i) present a general family of Lagrangians which allow for the exploration of the IB curve in all scenarios; and (ii) prove that if these Lagrangians are used, there is a (and we know the) one-to-one mapping between the Lagrange multiplier and the desired compression rate r for known IB curve shapes, hence, freeing us from the burden of solving the optimization problem for many values of the Lagrange multiplier. That is, we can solve the original constrained problem with a single optimization.

In this paper we present a model for unsupervised topic discovery in texts corpora. The proposed model uses documents, words, and topics lookup table embedding as neural network model parameters to build probabilities of words given topics, and probabilities of topics given documents. These probabilities are used to recover by marginalization probabilities of words given documents. For very large corpora where the number of documents can be in the order of billions, using a neural auto-encoder based document embedding is more scalable then using a lookup table embedding as classically done. We thus extended the lookup based document embedding model to continuous auto-encoder based model. Our models are trained using probabilistic latent semantic analysis (PLSA) assumptions. We evaluated our models on six datasets with a rich variety of contents. Conducted experiments demonstrate that the proposed neural topic models are very effective in capturing relevant topics. Furthermore, considering perplexity metric, conducted evaluation benchmarks show that our topic models outperform latent Dirichlet allocation (LDA) model which is classically used to address topic discovery tasks.

Iterative learning to infer approaches have become popular solvers for inverse problems. However, their memory requirements during training grow linearly with model depth, limiting in practice model expressiveness. In this work, we propose an iterative inverse model with constant memory that relies on invertible networks to avoid storing intermediate activations. As a result, the proposed approach allows us to train models with 400 layers on 3D volumes in an MRI image reconstruction task. In experiments on a public data set, we demonstrate that these deeper, and thus more expressive, networks perform state-of-the-art image reconstruction.

Existing methods produce very inefficient architectures and do not scale. In this paper, we introduce a new method for generating data from a random forest and learning a neural network that imitates it. Without any additional training data, this transformation creates very efficient neural networks that learn the decision boundaries of a random forest. The generated model is fully differentiable and can be combined with the feature extraction in a single pipeline enabling further end-to-end processing. Experiments on several real-world benchmark datasets demonstrate outstanding performance in terms of scalability, accuracy, and learning with very few training examples. Compared to state-of-the-art mappings, we significantly reduce the network size while achieving the same or even improved accuracy due to better generalization.

Problems of segmentation, denoising, registration and 3D reconstruction are often addressed with the graph cut algorithm. However, solving an unconstrained graph cut problem is NP-hard. For tractable optimization, pairwise potentials have to fulfill the submodularity inequality. In our learning paradigm, pairwise potentials are created as the dot product of a learned vector w with positive feature vectors. In order to constrain such a model to remain tractable, previous approaches have enforced the weight vector to be positive for pairwise potentials in which the labels differ, and set pairwise potentials to zero in the case that the label remains the same. Such constraints are sufficient to guarantee that the resulting pairwise potentials satisfy the submodularity inequality. However, we show that such an approach unnecessarily restricts the capacity of the learned models. Guaranteeing submodularity for all possible inputs, no matter how improbable, reduces inference error to effectively zero, but increases model error. In contrast, we relax the requirement of guaranteed submodularity to solutions that are probably approximately submodular. We show that the conceptually simple strategy of enforcing submodularity on the training examples guarantees with low sample complexity that test images will also yield submodular pairwise potentials. Results are presented in the binary and muticlass settings, showing substantial improvement from the resulting increased model capacity.

Random projections offer an appealing and flexible approach to a wide range of large-scale statistical problems. They are particularly useful in high-dimensional settings, where we have many covariates recorded for each observation. In classification problems there are two general techniques using random projections. The first involves many projections in an ensemble -- the idea here is to aggregate the results after applying different random projections, with the aim of achieving superior statistical accuracy. The second class of methods include hashing and sketching techniques, which are straightforward ways to reduce the complexity of a problem, perhaps therefore with a huge computational saving, while approximately preserving the statistical efficiency.