While the need for interpretable machine learning has been established, many common approaches are slow, lack fidelity, or hard to evaluate. Amortized explanation methods reduce the cost of providing interpretations by learning a global selector model that returns feature importances for a single instance of data. The selector model is trained to optimize the fidelity of the interpretations, as evaluated by a predictor model for the target. Popular methods learn the selector and predictor model in concert, which we show allows predictions to be encoded within interpretations. We introduce EVAL-X as a method to quantitatively evaluate interpretations and REAL-X as an amortized explanation method, which learn a predictor model that approximates the true data generating distribution given any subset of the input. We show EVAL-X can detect when predictions are encoded in interpretations and show the advantages of REAL-X through quantitative and radiologist evaluation.

Bayesian inference provides a uniquely rigorous approach to obtain principled justification for uncertainty in predictions, yet it is difficult to articulate suitably general prior belief in the machine learning context, where computational architectures are pure abstractions subject to frequent modifications by practitioners attempting to improve results. Parsimonious inference is an information-theoretic formulation of inference over arbitrary architectures that formalizes Occam's Razor; we prefer simple and sufficient explanations. Our universal hyperprior assigns plausibility to prior descriptions, encoded as sequences of symbols, by expanding on the core relationships between program length, Kolmogorov complexity, and Solomonoff's algorithmic probability. We then cast learning as information minimization over our composite change in belief when an architecture is specified, training data are observed, and model parameters are inferred. By distinguishing model complexity from prediction information, our framework also quantifies the phenomenon of memorization. Although our theory is general, it is most critical when datasets are limited, e.g. small or skewed. We develop novel algorithms for polynomial regression and random forests that are suitable for such data, as demonstrated by our experiments. Our approaches combine efficient encodings with prudent sampling strategies to construct predictive ensembles without cross-validation, thus addressing a fundamental challenge in how to efficiently obtain predictions from data.

In real-world applications of large-scale time series, one often encounters the situation where the temporal patterns of time series, while drifting over time, differ from one another in the same dataset. In this paper, we provably show under such heterogeneity, training a forecasting model with commonly used stochastic optimizers (e.g. SGD) potentially suffers large gradient variance, and thus requires long time training. To alleviate this issue, we propose a sampling strategy named Subgroup Sampling, which mitigates the large variance via sampling over pre-grouped time series. We further introduce SCott, a variance reduced SGD-style optimizer that co-designs subgroup sampling with the control variate method. In theory, we provide the convergence guarantee of SCott on smooth non-convex objectives. Empirically, we evaluate SCott and other baseline optimizers on both synthetic and real-world time series forecasting problems, and show SCott converges faster with respect to both iterations and wall clock time. Additionally, we show two SCott variants that can speed up Adam and Adagrad without compromising generalization of forecasting models.

Random Forest's performance can be matched by a single slow-growing tree (SGT), which uses a learning rate to tame CART's greedy algorithm. SGT exploits the view that CART is an extreme case of an iterative weighted least square procedure. Moreover, a unifying view of Boosted Trees (BT) and Random Forests (RF) is presented. Greedy ML algorithms' outcomes can be improved using either "slow learning" or diversification. SGT applies the former to estimate a single deep tree, and Booging (bagging stochastic BT with a high learning rate) uses the latter with additive shallow trees. The performance of this tree ensemble quaternity (Booging, BT, SGT, RF) is assessed on simulated and real regression tasks.

A widely recognized difficulty in federated learning arises from the statistical heterogeneity among clients: local datasets often come from different but not entirely unrelated distributions, and personalization is, therefore, necessary to achieve optimal results from each individual's perspective. In this paper, we show how the excess risks of personalized federated learning with a smooth, strongly convex loss depend on data heterogeneity from a minimax point of view. Our analysis reveals a surprising theorem of the alternative for personalized federated learning: there exists a threshold such that (a) if a certain measure of data heterogeneity is below this threshold, the FedAvg algorithm [McMahan et al., 2017] is minimax optimal; (b) when the measure of heterogeneity is above this threshold, then doing pure local training (i.e., clients solve empirical risk minimization problems on their local datasets without any communication) is minimax optimal. As an implication, our results show that the presumably difficult (infinite-dimensional) problem of adapting to client-wise heterogeneity can be reduced to a simple binary decision problem of choosing between the two baseline algorithms. Our analysis relies on a new notion of algorithmic stability that takes into account the nature of federated learning.

Embedding high dimensional data into low dimensional spaces, such as with tSNE [van der Maaten and Hinton, 2008] or UMAP [McInnes et al., 2018], allow us to visually inspect and discover meaningful structure from the data that would otherwise be difficult or impossible to see. These methods are as popular as they are useful, but, at the same time limited in that they are one-shot only: they embed the data as is, and that is that. If the resulting embedding reveals novel knowledge, all is well, but, what if the structure that dominates it is something we already know, something we are no longer interested in, or, if we want to discover whether the data has meaningful structure other than what the first result revealed? In word embeddings, for example, we may already know that certain words are synonyms, while in single cell sequencing we may want to discover structure other than known cell types, or factor out family relationships. The question at hand is therefore, how can we obtain low-dimensional embeddings that reveal structure beyond what we already know, i.e. how to factor out prior knowledge from low-dimensional embeddings? For conditional embeddings, research so far mostly focused on emphasizing rather than factoring out prior knowledge [Barshan et al., 2011, De Ridder et al., 2003, Hanhijärvi et al., 2009], with conditional tSNE as notable exception, which, however, can only factor out label information [Kang et al., 2019]. Here, we propose two techniques for factoring out a more general form of prior knowledge from low-dimensional embeddings of arbitrary data types. In particular, we consider background knowledge in the form of pairwise distances between samples. This formulation allows us to cover a plethora of practical instances including labels, clustering structure, family trees, user-defined distances, but also, and especially important for unstructured data, kernel matrices.

This material may be downloaded for personal use only. Any other use requires prior permission of the American Society of Civil Engineers. This material may be found at https://doi.org/10.1061/AJRUA6.0001106 ABSTRACT In data-driven SHM, the signals recorded from systems in operation can be noisy and incomplete. Data corresponding to each of the operational, environmental, and damage states are rarely available a priori; furthermore, labelling to describe the measurements is often unavailable. In consequence, the algorithms used to implement SHM should be robust and adaptive, while accommodating for missing information in the training-data - such that new information can be included if it becomes available. By reviewing novel techniques for statistical learning (introduced in previous work), it is argued that probabilistic algorithms offer a natural solution to the modelling of SHM data in practice. In three case-studies, probabilistic methods are adapted for applications to SHM signals -- including semi-supervised learning, active learning, and multi-task learning. Various machine learning tools have been applied in the literature, for example (Vanik et al. 2000; Sohn et al. 2003; Chatzi and Smyth 2009), and used to infer the health or performance state of the monitored system, either directly or indirectly. Generally, algorithms for regression, classification, density estimation, or clustering learn patterns in the measured signals (available for training), and the associated patterns can be used to infer the state of the system in operation, given future measurements (Worden and Manson 2006). Unsurprisingly, there are numerous ways to apply machine learning to SHM. Notably (and categorised generally), advances have focussed on various probabilistic (e.g. Each approach has its advantages; however, considering certain challenges associated with SHM data (outlined in the next section) the current work focusses on probabilistic (i.e. Additionally, probabilistic methods can lead to predictions under uncertainty (Papoulis 1965) - a significant advantage in risk-based applications.

Label-imbalanced and group-sensitive classification seeks to appropriately modify standard training algorithms to optimize relevant metrics such as balanced error and/or equal opportunity. For label imbalances, recent works have proposed a logit-adjusted loss modification to standard empirical risk minimization. We show that this might be ineffective in general and, in particular so, in the overparameterized regime where training continues in the zero training-error regime. Specifically for binary linear classification of a separable dataset, we show that the modified loss converges to the max-margin SVM classifier despite the logit adjustment. Instead, we propose a more general vector-scaling loss that directly relates to the cost-sensitive SVM (CS-SVM), thus favoring larger margin to the minority class. Through an insightful sharp asymptotic analysis for a Gaussian-mixtures data model, we demonstrate the efficacy of CS-SVM in balancing the errors of the minority/majority classes. Our analysis also leads to a simple strategy for optimally tuning the involved margin-ratio parameter. Then, we show how our results extend naturally to binary classification with sensitive groups, thus treating the two common types of imbalances (label/group) in a unifying way. We corroborate our theoretical findings with numerical experiments on both synthetic and real-world datasets.

Stochastic convex optimization over an $\ell_1$-bounded domain is ubiquitous in machine learning applications such as LASSO but remains poorly understood when learning with differential privacy. We show that, up to logarithmic factors the optimal excess population loss of any $(\varepsilon,\delta)$-differentially private optimizer is $\sqrt{\log(d)/n} + \sqrt{d}/\varepsilon n.$ The upper bound is based on a new algorithm that combines the iterative localization approach of~\citet{FeldmanKoTa20} with a new analysis of private regularized mirror descent. It applies to $\ell_p$ bounded domains for $p\in [1,2]$ and queries at most $n^{3/2}$ gradients improving over the best previously known algorithm for the $\ell_2$ case which needs $n^2$ gradients. Further, we show that when the loss functions satisfy additional smoothness assumptions, the excess loss is upper bounded (up to logarithmic factors) by $\sqrt{\log(d)/n} + (\log(d)/\varepsilon n)^{2/3}.$ This bound is achieved by a new variance-reduced version of the Frank-Wolfe algorithm that requires just a single pass over the data. We also show that the lower bound in this case is the minimum of the two rates mentioned above.

Batch Normalization (BN) is a commonly used technique to accelerate and stabilize training of deep neural networks. Despite its empirical success, a full theoretical understanding of BN is yet to be developed. In this work, we analyze BN through the lens of convex optimization. We introduce an analytic framework based on convex duality to obtain exact convex representations of weight-decay regularized ReLU networks with BN, which can be trained in polynomial-time. Our analyses also show that optimal layer weights can be obtained as simple closed-form formulas in the high-dimensional and/or overparameterized regimes. Furthermore, we find that Gradient Descent provides an algorithmic bias effect on the standard non-convex BN network, and we design an approach to explicitly encode this implicit regularization into the convex objective. Experiments with CIFAR image classification highlight the effectiveness of this explicit regularization for mimicking and substantially improving the performance of standard BN networks.