Towards a Unified Analysis of Neural Networks in Nonparametric Instrumental Variable Regression: Optimization and Generalization

A new framework for real-time crash detection has been developed, addressing the challenges of trajectory acquisition and vehicle tracking. This two-stage, trajectory-free approach utilizes road segment maps to identify crashes by generating future maps through a diffusion-based model called Mapfusion. The model incorporates temporal dynamics and background context to enhance the accuracy of crash detection, aiming to improve traffic safety and efficiency.

Causal discovery is a method that integrates data with expert knowledge to identify the directed acyclic graph (DAG) that illustrates causal relationships among variables. In scenarios with limited data, bagging techniques are employed to assess confidence in an average DAG derived from bootstrapped DAGs. However, the aggregation process has been underexplored, as it typically focuses solely on individual edge confidence, neglecting complex higher-order structures. This paper presents a new theoretical framework and a DAG aggregation algorithm, demonstrating its efficiency and effectiveness, p…

Importance sampling is a technique utilized to enhance the efficiency of deep neural network (DNN) training by minimizing the variance of gradient estimators. This paper introduces a method for estimating variance reduction during DNN training using only minibatches sampled through importance sampling. Additionally, it suggests an optimal minibatch size for automatic learning rate adjustment and presents a metric to quantify the efficiency of importance sampling, supported by theoretical analysis and experiments demonstrating improved training efficiency and model accuracy.

The article discusses a novel method for detecting low-dimensional structures in high-dimensional probability measures, crucial for efficient sampling. This approach approximates a target measure as a perturbation of a reference measure along significant directions in Euclidean space. The reference measure can be Gaussian or a nonlinear transformation of it, commonly used in generative modeling. The study establishes a link between the dimensional logarithmic Sobolev inequality and Kullback-Leibler divergence minimization, enhancing approximation techniques.

Open Set Recognition (OSR) is a critical area in machine learning that involves not only classifying known categories but also rejecting unknown samples. A significant challenge arises when unknown samples resemble known classes, leading to overconfidence in model predictions and misclassifications. This paper introduces a framework designed to mitigate overconfidence through a two-component system: a perturbation-based uncertainty estimation module and an unknown detection module that employs distinct classifiers.

The article examines the running time of differentially private empirical and population risk minimization for Lipschitz convex losses, focusing on scenarios with non-smooth losses. It establishes that an expected running time of Ω(min{√d/α², d/log(1/α)}) is required to achieve α excess risk for problems with dimension d when d ≥ 1/α². The findings indicate that these results are tight for dimensions exceeding Ω(1/α⁴) and also provide strengthened lower bounds for algorithms using smaller minibatch sizes.

The study investigates the impact of two information feeds, a closed-form Markov estimator and an online trained actor-critic, on reneging and jockeying behaviors in a dual M/M/1 system. It reveals that with unequal service rates and total-time patience, total wait increases linearly, leading to inevitable abandonment. The probability of successful jockeying diminishes as backlog increases. Both information models converge to the same asymptotic limits under certain conditions, highlighting the importance of value-of-information in finite regimes.

The article presents a distributionally robust approach for estimating both the covariance and precision matrices of a random vector. This model minimizes the worst-case weighted sum of the Frobenius loss for covariance estimation and Stein's loss for precision estimation, using an ambiguity set centered around a nominal distribution. The method is formulated as a convex optimization problem, leading to quasi-analytical estimators that correct spectral bias through nonlinear shrinkage, enhancing the reliability of statistical analyses.