Modern black-box predictive models are often accompanied by weak performance guarantees that only hold asymptotically in the size of the dataset or require strong parametric assumptions. In response to this, split conformal prediction represents a promising avenue to obtain finite-sample guarantees under minimal distribution-free assumptions. Although prediction set validity most often concerns marginal coverage, we explore the related but different guarantee of tolerance regions, reformulating known results in the language of nested prediction sets and extending on the duality between marginal coverage and tolerance regions. Furthermore, we highlight the connection between split conformal prediction and classical tolerance predictors developed in the 1940s, as well as recent developments in distribution-free risk control. One result that transfers from classical tolerance predictors is that the coverage of a prediction set based on order statistics, conditional on the calibration set, is a random variable stochastically dominating the Beta distribution. We demonstrate the empirical effectiveness of our findings on synthetic and real datasets using a popular split conformal prediction procedure called conformalized quantile regression (CQR).
相關內容
Training adversarially robust discriminative (i.e., softmax) classifier has been the dominant approach to robust classification. Building on recent work on adversarial training (AT)-based generative models, we investigate using AT to learn unnormalized class-conditional density models and then performing generative robust classification. Our result shows that, under the condition of similar model capacities, the generative robust classifier achieves comparable performance to a baseline softmax robust classifier when the test data is clean or when the test perturbation is of limited size, and much better performance when the test perturbation size exceeds the training perturbation size. The generative classifier is also able to generate samples or counterfactuals that more closely resemble the training data, suggesting that the generative classifier can better capture the class-conditional distributions. In contrast to standard discriminative adversarial training where advanced data augmentation techniques are only effective when combined with weight averaging, we find it straightforward to apply advanced data augmentation to achieve better robustness in our approach. Our result suggests that the generative classifier is a competitive alternative to robust classification, especially for problems with limited number of classes.
Max-stable processes provide natural models for the modelling of spatial extreme values observed at a set of spatial sites. Full likelihood inference for max-stable data is, however, complicated by the form of the likelihood function as it contains a sum over all partitions of sites. As such, the number of terms to sum over grows rapidly with the number of sites and quickly becomes prohibitively burdensome to compute. We propose a variational inference approach to full likelihood inference that circumvents the problematic sum. To achieve this, we first posit a parametric family of partition distributions from which partitions can be sampled. Second, we optimise the parameters of the family in conjunction with the max-stable model to find the partition distribution best supported by the data, and to estimate the max-stable model parameters. In a simulation study we show that our method enables full likelihood inference in higher dimensions than previous methods, and is readily applicable to data sets with a large number of observations. Furthermore, our method can easily be extended to a Bayesian setting. Code is available at //github.com/LPAndersson/MaxStableVI.jl.
The multi-armed bandit(MAB) problem is a simple yet powerful framework that has been extensively studied in the context of decision-making under uncertainty. In many real-world applications, such as robotic applications, selecting an arm corresponds to a physical action that constrains the choices of the next available arms (actions). Motivated by this, we study an extension of MAB called the graph bandit, where an agent travels over a graph to maximize the reward collected from different nodes. The graph defines the agent's freedom in selecting the next available nodes at each step. We assume the graph structure is fully available, but the reward distributions are unknown. Built upon an offline graph-based planning algorithm and the principle of optimism, we design a learning algorithm, G-UCB, that balances long-term exploration-exploitation using the principle of optimism. We show that our proposed algorithm achieves $O(\sqrt{|S|T\log(T)}+D|S|\log T)$ learning regret, where $|S|$ is the number of nodes and $D$ is the diameter of the graph, which matches the theoretical lower bound $\Omega(\sqrt{|S|T})$ up to logarithmic factors. To our knowledge, this result is among the first tight regret bounds in non-episodic, un-discounted learning problems with known deterministic transitions. Numerical experiments confirm that our algorithm outperforms several benchmarks.
Bayesian inference for high-dimensional inverse problems is computationally costly and requires selecting a suitable prior distribution. Amortized variational inference addresses these challenges via a neural network that acts as a surrogate conditional distribution, matching the posterior distribution not only for one instance of data, but a distribution of data pertaining to a specific inverse problem. During inference, the neural network -- in our case a conditional normalizing flow -- provides posterior samples with virtually no cost. However, the accuracy of Amortized variational inference relies on the availability of high-fidelity training data, which seldom exists in geophysical inverse problems due to the Earth's heterogeneity. In addition, the network is prone to errors if evaluated over out-of-distribution data. As such, we propose to increases the resilience of amortized variational inference in presence of moderate data distribution shifts. We achieve this via a correction to the latent distribution that improves the posterior distribution approximation for the data at hand. The correction involves relaxing the standard Gaussian assumption on the latent distribution and parameterizing it via a Gaussian distribution with an unknown mean and (diagonal) covariance. These unknowns are then estimated by minimizing the Kullback-Leibler divergence between the corrected and (physics-based) true posterior distributions. While generic and applicable to other inverse problems, by means of a linearized seismic imaging example, we show that our correction step improves the robustness of amortized variational inference with respect to changes in number of seismic sources, noise variance, and shifts in the prior distribution. This approach provides a seismic image with limited artifacts and an assessment of its uncertainty with approximately the same cost as five reverse-time migrations.
This paper studies a class of multi-agent reinforcement learning (MARL) problems where the reward that an agent receives depends on the states of other agents, but the next state only depends on the agent's own current state and action. We name it REC-MARL standing for REward-Coupled Multi-Agent Reinforcement Learning. REC-MARL has a range of important applications such as real-time access control and distributed power control in wireless networks. This paper presents a distributed and optimal policy gradient algorithm for REC-MARL. The proposed algorithm is distributed in two aspects: (i) the learned policy is a distributed policy that maps a local state of an agent to its local action and (ii) the learning/training is distributed, during which each agent updates its policy based on its own and neighbors' information. The learned policy is provably optimal among all local policies and its regret bounds depend on the dimension of local states and actions. This distinguishes our result from most existing results on MARL, which often obtain stationary-point policies. The experimental results of our algorithm for the real-time access control and power control in wireless networks show that our policy significantly outperforms the state-of-the-art algorithms and well-known benchmarks.
Recent developments in in-situ monitoring and process control in Additive Manufacturing (AM), also known as 3D-printing, allows the collection of large amounts of emission data during the build process of the parts being manufactured. This data can be used as input into 3D and 2D representations of the 3D-printed parts. However the analysis and use, as well as the characterization of this data still remains a manual process. The aim of this paper is to propose an adaptive human-in-the-loop approach using Machine Learning techniques that automatically inspect and annotate the emissions data generated during the AM process. More specifically, this paper will look at two scenarios: firstly, using convolutional neural networks (CNNs) to automatically inspect and classify emission data collected by in-situ monitoring and secondly, applying Active Learning techniques to the developed classification model to construct a human-in-the-loop mechanism in order to accelerate the labeling process of the emission data. The CNN-based approach relies on transfer learning and fine-tuning, which makes the approach applicable to other industrial image patterns. The adaptive nature of the approach is enabled by uncertainty sampling strategy to automatic selection of samples to be presented to human experts for annotation.
A graph $G$ is called self-ordered (a.k.a asymmetric) if the identity permutation is its only automorphism. Equivalently, there is a unique isomorphism from $G$ to any graph that is isomorphic to $G$. We say that $G=(V,E)$ is robustly self-ordered if the size of the symmetric difference between $E$ and the edge-set of the graph obtained by permuting $V$ using any permutation $\pi:V\to V$ is proportional to the number of non-fixed-points of $\pi$. In this work, we initiate the study of the structure, construction and utility of robustly self-ordered graphs. We show that robustly self-ordered bounded-degree graphs exist (in abundance), and that they can be constructed efficiently, in a strong sense. Specifically, given the index of a vertex in such a graph, it is possible to find all its neighbors in polynomial-time (i.e., in time that is poly-logarithmic in the size of the graph). We also consider graphs of unbounded degree, seeking correspondingly unbounded robustness parameters. We again demonstrate that such graphs (of linear degree) exist (in abundance), and that they can be constructed efficiently, in a strong sense. This turns out to require very different tools. Specifically, we show that the construction of such graphs reduces to the construction of non-malleable two-source extractors (with very weak parameters but with some additional natural features). We demonstrate that robustly self-ordered bounded-degree graphs are useful towards obtaining lower bounds on the query complexity of testing graph properties both in the bounded-degree and the dense graph models. One of the results that we obtain, via such a reduction, is a subexponential separation between the query complexities of testing and tolerant testing of graph properties in the bounded-degree graph model.
Wasserstein distributionally robust optimization (DRO) has found success in operations research and machine learning applications as a powerful means to obtain solutions with favourable out-of-sample performances. Two compelling explanations for the success are the generalization bounds derived from Wasserstein DRO and the equivalency between Wasserstein DRO and the regularization scheme commonly applied in machine learning. Existing results on generalization bounds and the equivalency to regularization are largely limited to the setting where the Wasserstein ball is of a certain type and the decision criterion takes certain forms of an expected function. In this paper, we show that by focusing on Wasserstein DRO problems with affine decision rules, it is possible to obtain generalization bounds and the equivalency to regularization in a significantly broader setting where the Wasserstein ball can be of a general type and the decision criterion can be a general measure of risk, i.e., nonlinear in distributions. This allows for accommodating many important classification, regression, and risk minimization applications that have not been addressed to date using Wasserstein DRO. Our results are strong in that the generalization bounds do not suffer from the curse of dimensionality and the equivalency to regularization is exact. As a byproduct, our regularization results broaden considerably the class of Wasserstein DRO models that can be solved efficiently via regularization formulations.
Non-negative matrix factorization is a popular unsupervised machine learning algorithm for extracting meaningful features from data which are inherently non-negative. However, such data sets may often contain privacy-sensitive user data, and therefore, we may need to take necessary steps to ensure the privacy of the users while analyzing the data. In this work, we focus on developing a Non-negative matrix factorization algorithm in the privacy-preserving framework. More specifically, we propose a novel privacy-preserving algorithm for non-negative matrix factorisation capable of operating on private data, while achieving results comparable to those of the non-private algorithm. We design the framework such that one has the control to select the degree of privacy grantee based on the utility gap. We show our proposed framework's performance in six real data sets. The experimental results show that our proposed method can achieve very close performance with the non-private algorithm under some parameter regime, while ensuring strict privacy.
Sampling methods (e.g., node-wise, layer-wise, or subgraph) has become an indispensable strategy to speed up training large-scale Graph Neural Networks (GNNs). However, existing sampling methods are mostly based on the graph structural information and ignore the dynamicity of optimization, which leads to high variance in estimating the stochastic gradients. The high variance issue can be very pronounced in extremely large graphs, where it results in slow convergence and poor generalization. In this paper, we theoretically analyze the variance of sampling methods and show that, due to the composite structure of empirical risk, the variance of any sampling method can be decomposed into \textit{embedding approximation variance} in the forward stage and \textit{stochastic gradient variance} in the backward stage that necessities mitigating both types of variance to obtain faster convergence rate. We propose a decoupled variance reduction strategy that employs (approximate) gradient information to adaptively sample nodes with minimal variance, and explicitly reduces the variance introduced by embedding approximation. We show theoretically and empirically that the proposed method, even with smaller mini-batch sizes, enjoys a faster convergence rate and entails a better generalization compared to the existing methods.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191