Resilience against stragglers is a critical element of prediction serving systems, tasked with executing inferences on input data for a pre-trained machine-learning model. In this paper, we propose NeRCC, as a general straggler-resistant framework for approximate coded computing. NeRCC includes three layers: (1) encoding regression and sampling, which generates coded data points, as a combination of original data points, (2) computing, in which a cluster of workers run inference on the coded data points, (3) decoding regression and sampling, which approximately recovers the predictions of the original data points from the available predictions on the coded data points. We argue that the overall objective of the framework reveals an underlying interconnection between two regression models in the encoding and decoding layers. We propose a solution to the nested regressions problem by summarizing their dependence on two regularization terms that are jointly optimized. Our extensive experiments on different datasets and various machine learning models, including LeNet5, RepVGG, and Vision Transformer (ViT), demonstrate that NeRCC accurately approximates the original predictions in a wide range of stragglers, outperforming the state-of-the-art by up to 23%.

Background and objective Risk prediction models aim at identifying people at higher risk of developing a target disease. Feature selection is particularly important to improve the prediction model performance avoiding overfitting and to identify the leading cancer risk (and protective) factors. Assessing the stability of feature selection/ranking algorithms becomes an important issue when the aim is to analyze the features with more prediction power. Methods This work is focused on colorectal cancer, assessing several feature ranking algorithms in terms of performance for a set of risk prediction models (Neural Networks, Support Vector Machines (SVM), Logistic Regression, k-Nearest Neighbors and Boosted Trees). Additionally, their robustness is evaluated following a conventional approach with scalar stability metrics and a visual approach proposed in this work to study both similarity among feature ranking techniques as well as their individual stability. A comparative analysis is carried out between the most relevant features found out in this study and features provided by the experts according to the state-of-the-art knowledge. Results The two best performance results in terms of Area Under the ROC Curve (AUC) are achieved with a SVM classifier using the top-41 features selected by the SVM wrapper approach (AUC=0.693) and Logistic Regression with the top-40 features selected by the Pearson (AUC=0.689). Experiments showed that performing feature selection contributes to classification performance with a 3.9% and 1.9% improvement in AUC for the SVM and Logistic Regression classifier, respectively, with respect to the results using the full feature set. The visual approach proposed in this work allows to see that the Neural Network-based wrapper ranking is the most unstable while the Random Forest is the most stable.

This work studies algorithms for learning from aggregate responses. We focus on the construction of aggregation sets (called bags in the literature) for event-level loss functions. We prove for linear regression and generalized linear models (GLMs) that the optimal bagging problem reduces to one-dimensional size-constrained $k$-means clustering. Further, we theoretically quantify the advantage of using curated bags over random bags. We then propose the PriorBoost algorithm, which adaptively forms bags of samples that are increasingly homogeneous with respect to (unobserved) individual responses to improve model quality. We study label differential privacy for aggregate learning, and we also provide extensive experiments showing that PriorBoost regularly achieves optimal model quality for event-level predictions, in stark contrast to non-adaptive algorithms.

Large language models (LLMs) exhibit remarkable performance improvement through in-context learning (ICL) by leveraging task-specific examples in the input. However, the mechanisms behind this improvement remain elusive. In this work, we investigate how LLM embeddings and attention representations change following in-context-learning, and how these changes mediate improvement in behavior. We employ neuroscience-inspired techniques such as representational similarity analysis (RSA) and propose novel methods for parameterized probing and measuring ratio of attention to relevant vs. irrelevant information in Llama-2 70B and Vicuna 13B. We designed two tasks with a priori relationships among their conditions: linear regression and reading comprehension. We formed hypotheses about expected similarities in task representations and measured hypothesis alignment of LLM representations before and after ICL as well as changes in attention. Our analyses revealed a meaningful correlation between improvements in behavior after ICL and changes in both embeddings and attention weights across LLM layers. This empirical framework empowers a nuanced understanding of how latent representations shape LLM behavior, offering valuable tools and insights for future research and practical applications.

We study online changepoint detection in the context of a linear regression model. We propose a class of heavily weighted statistics based on the CUSUM process of the regression residuals, which are specifically designed to ensure timely detection of breaks occurring early on during the monitoring horizon. We subsequently propose a class of composite statistics, constructed using different weighing schemes; the decision rule to mark a changepoint is based on the largest statistic across the various weights, thus effectively working like a veto-based voting mechanism, which ensures fast detection irrespective of the location of the changepoint. Our theory is derived under a very general form of weak dependence, thus being able to apply our tests to virtually all time series encountered in economics, medicine, and other applied sciences. Monte Carlo simulations show that our methodologies are able to control the procedure-wise Type I Error, and have short detection delays in the presence of breaks.

Constant (naive) imputation is still widely used in practice as this is a first easy-to-use technique to deal with missing data. Yet, this simple method could be expected to induce a large bias for prediction purposes, as the imputed input may strongly differ from the true underlying data. However, recent works suggest that this bias is low in the context of high-dimensional linear predictors when data is supposed to be missing completely at random (MCAR). This paper completes the picture for linear predictors by confirming the intuition that the bias is negligible and that surprisingly naive imputation also remains relevant in very low dimension.To this aim, we consider a unique underlying random features model, which offers a rigorous framework for studying predictive performances, whilst the dimension of the observed features varies.Building on these theoretical results, we establish finite-sample bounds on stochastic gradient (SGD) predictors applied to zero-imputed data, a strategy particularly well suited for large-scale learning.If the MCAR assumption appears to be strong, we show that similar favorable behaviors occur for more complex missing data scenarios.

We propose a novel ensemble method called Riemann-Lebesgue Forest (RLF) for regression. The core idea of RLF is to mimic the way how a measurable function can be approximated by partitioning its range into a few intervals. With this idea in mind, we develop a new tree learner named Riemann-Lebesgue Tree which has a chance to split the node from response $Y$ or a direction in feature space $\mathbf{X}$ at each non-terminal node. We generalize the asymptotic performance of RLF under different parameter settings mainly through Hoeffding decomposition \cite{Vaart} and Stein's method \cite{Chen2010NormalAB}. When the underlying function $Y=f(\mathbf{X})$ follows an additive regression model, RLF is consistent with the argument from \cite{Scornet2014ConsistencyOR}. The competitive performance of RLF against original random forest \cite{Breiman2001RandomF} is demonstrated by experiments in simulation data and real world datasets.

Diffusion models have emerged as a powerful tool rivaling GANs in generating high-quality samples with improved fidelity, flexibility, and robustness. A key component of these models is to learn the score function through score matching. Despite empirical success on various tasks, it remains unclear whether gradient-based algorithms can learn the score function with a provable accuracy. As a first step toward answering this question, this paper establishes a mathematical framework for analyzing score estimation using neural networks trained by gradient descent. Our analysis covers both the optimization and the generalization aspects of the learning procedure. In particular, we propose a parametric form to formulate the denoising score-matching problem as a regression with noisy labels. Compared to the standard supervised learning setup, the score-matching problem introduces distinct challenges, including unbounded input, vector-valued output, and an additional time variable, preventing existing techniques from being applied directly. In this paper, we show that with a properly designed neural network architecture, the score function can be accurately approximated by a reproducing kernel Hilbert space induced by neural tangent kernels. Furthermore, by applying an early-stopping rule for gradient descent and leveraging certain coupling arguments between neural network training and kernel regression, we establish the first generalization error (sample complexity) bounds for learning the score function despite the presence of noise in the observations. Our analysis is grounded in a novel parametric form of the neural network and an innovative connection between score matching and regression analysis, facilitating the application of advanced statistical and optimization techniques.

An accurate prediction of house prices is a fundamental requirement for various sectors including real estate and mortgage lending. It is widely recognized that a property value is not solely determined by its physical attributes but is significantly influenced by its surrounding neighbourhood. Meeting the diverse housing needs of individuals while balancing budget constraints is a primary concern for real estate developers. To this end, we addressed the house price prediction problem as a regression task and thus employed various machine learning techniques capable of expressing the significance of independent variables. We made use of the housing dataset of Ames City in Iowa, USA to compare support vector regressor, random forest regressor, XGBoost, multilayer perceptron and multiple linear regression algorithms for house price prediction. Afterwards, we identified the key factors that influence housing costs. Our results show that XGBoost is the best performing model for house price prediction.

To estimate accurately the parameters of a regression model, the sample size must be large enough relative to the number of possible predictors for the model. In practice, sufficient data is often lacking, which can lead to overfitting of the model and, as a consequence, unreliable predictions of the outcome of new patients. Pooling data from different data sets collected in different (medical) centers would alleviate this problem, but is often not feasible due to privacy regulation or logistic problems. An alternative route would be to analyze the local data in the centers separately and combine the statistical inference results with the Bayesian Federated Inference (BFI) methodology. The aim of this approach is to compute from the inference results in separate centers what would have been found if the statistical analysis was performed on the combined data. We explain the methodology under homogeneity and heterogeneity across the populations in the separate centers, and give real life examples for better understanding. Excellent performance of the proposed methodology is shown. An R-package to do all the calculations has been developed and is illustrated in this paper. The mathematical details are given in the Appendix.