Model mismatches prevail in real-world applications. Hence it is important to design robust safe control algorithms for systems with uncertain dynamic models. The major challenge is that uncertainty results in difficulty in finding a feasible safe control in real-time. Existing methods usually simplify the problem such as restricting uncertainty type, ignoring control limits, or forgoing feasibility guarantees. In this work, we overcome these issues by proposing a robust safe control framework for bounded state-dependent uncertainties. We first guarantee the feasibility of safe control for uncertain dynamics by learning a control-limits-aware, uncertainty-robust safety index. Then we show that robust safe control can be formulated as convex problems (Convex Semi-Infinite Programming or Second-Order Cone Programming) and propose corresponding optimal solvers that can run in real-time. In addition, we analyze when and how safety can be preserved under unmodeled uncertainties. Experiment results show that our method successfully finds robust safe control in real-time for different uncertainties and is much less conservative than a strong baseline algorithm.
相關內容
Model Predictive Control (MPC) is a well-established approach to solve infinite horizon optimal control problems. Since optimization over an infinite time horizon is generally infeasible, MPC determines a suboptimal feedback control by repeatedly solving finite time optimal control problems. Although MPC has been successfully used in many applications, applying MPC to large-scale systems -- arising, e.g., through discretization of partial differential equations -- requires the solution of high-dimensional optimal control problems and thus poses immense computational effort. We consider systems governed by parametrized parabolic partial differential equations and employ the reduced basis (RB) method as a low-dimensional surrogate model for the finite time optimal control problem. The reduced order optimal control serves as feedback control for the original large-scale system. We analyze the proposed RB-MPC approach by first developing a posteriori error bounds for the errors in the optimal control and associated cost functional. These bounds can be evaluated efficiently in an offline-online computational procedure and allow us to guarantee asymptotic stability of the closed-loop system using the RB-MPC approach in several practical scenarios. We also propose an adaptive strategy to choose the prediction horizon of the finite time optimal control problem. Numerical results are presented to illustrate the theoretical properties of our approach.
Uncertainty approximation in text classification is an important area with applications in domain adaptation and interpretability. The most widely used uncertainty approximation method is Monte Carlo Dropout, which is computationally expensive as it requires multiple forward passes through the model. A cheaper alternative is to simply use a softmax to estimate model uncertainty. However, prior work has indicated that the softmax can generate overconfident uncertainty estimates and can thus be tricked into producing incorrect predictions. In this paper, we perform a thorough empirical analysis of both methods on five datasets with two base neural architectures in order to reveal insight into the trade-offs between the two. We compare the methods' uncertainty approximations and downstream text classification performance, while weighing their performance against their computational complexity as a cost-benefit analysis, by measuring runtime (cost) and the downstream performance (benefit). We find that, while Monte Carlo produces the best uncertainty approximations, using a simple softmax leads to competitive uncertainty estimation for text classification at a much lower computational cost, suggesting that softmax can in fact be a sufficient uncertainty estimate when computational resources are a concern.
We consider the problem of forecasting debt recovery from large portfolios of non-performing unsecured consumer loans under management. The state of the art in industry is to use stochastic processes to approximately model payment behaviour of individual customers based on several covariates, including credit scores and payment history. Monte Carlo simulation of these stochastic processes can enable forecasting of the possible returns from portfolios of defaulted debt, and the quantification of uncertainty. Despite the fact that the individual-level models are relatively simple, it is challenging to carry out simulations at the portfolio level because of the very large number of accounts. The accounts are also heterogeneous, with a broad range of values for the collection variances. We aim to solve two main problems: efficient allocation of computational resources in the simulations to estimate the likely collections as precisely as possible, and quantification of the uncertainty in the forecasts. We show that under certain conditions, robust estimators of population-level variance can be constructed by summing over coarse unbiased estimators of the variance of individual accounts. The proposed methods are demonstrated through application to a model which shares key features with those that are used in practice.
Empirical detection of long range dependence (LRD) of a time series often consists of deciding whether an estimate of the memory parameter $d$ corresponds to LRD. Surprisingly, the literature offers numerous spectral domain estimators for $d$ but there are only a few estimators in the time domain. Moreover, the latter estimators are criticized for relying on visual inspection to determine an observation window $[n_1, n_2]$ for a linear regression to run on. Theoretically motivated choices of $n_1$ and $n_2$ are often missing for many time series models. In this paper, we take the well-known variance plot estimator and provide rigorous asymptotic conditions on $[n_1, n_2]$ to ensure the estimator's consistency under LRD. We establish these conditions for a large class of square-integrable time series models. This large class enables one to use the variance plot estimator to detect LRD for infinite-variance time series (after suitable transformation). Thus, detection of LRD for infinite-variance time series is another novelty of our paper. A simulation study indicates that the variance plot estimator can detect LRD better than the popular spectral domain GPH estimator.
Model-based control requires an accurate model of the system dynamics for precisely and safely controlling the robot in complex and dynamic environments. Moreover, in presence of variations in the operating conditions, the model should be continuously refined to compensate for dynamics changes. In this paper, we propose a self-supervised learning approach to actively model robot discrete-time dynamics. We combine offline learning from past experience and online learning from present robot interaction with the unknown environment. These two ingredients enable highly sample-efficient and adaptive learning for accurate inference of the model dynamics in real-time even in operating regimes significantly different from the training distribution. Moreover, we design an uncertainty-aware model predictive controller that is conditioned to the aleatoric (data) uncertainty of the learned dynamics. The controller actively selects the optimal control actions that (i) optimize the control performance and (ii) boost the online learning sample efficiency. We apply the proposed method to a quadrotor system in multiple challenging real-world experiments. Our approach exhibits high flexibility and generalization capabilities by consistently adapting to unseen flight conditions, while it significantly outperforms classical and adaptive control baselines.
Long-term fairness is an important factor of consideration in designing and deploying learning-based decision systems in high-stake decision-making contexts. Recent work has proposed the use of Markov Decision Processes (MDPs) to formulate decision-making with long-term fairness requirements in dynamically changing environments, and demonstrated major challenges in directly deploying heuristic and rule-based policies that worked well in static environments. We show that policy optimization methods from deep reinforcement learning can be used to find strictly better decision policies that can often achieve both higher overall utility and less violation of the fairness requirements, compared to previously-known strategies. In particular, we propose new methods for imposing fairness requirements in policy optimization by regularizing the advantage evaluation of different actions. Our proposed methods make it easy to impose fairness constraints without reward engineering or sacrificing training efficiency. We perform detailed analyses in three established case studies, including attention allocation in incident monitoring, bank loan approval, and vaccine distribution in population networks.
We consider the problem of performing Bayesian inference in probabilistic models where observations are accompanied by uncertainty, referred to as `uncertain evidence'. In many real-world scenarios, such uncertainty stems from measurement errors associated with observable quantities in probabilistic models. We explore how to interpret uncertain evidence, and by extension the importance of proper interpretation as it pertains to inference about latent variables. We consider a recently-proposed method `stochastic evidence' as well as revisit two older methods: Jeffrey's rule and virtual evidence. We devise concrete guidelines on how to account for uncertain evidence and we provide new insights, particularly regarding consistency. To showcase the impact of different interpretations of the same uncertain evidence, we carry out experiments in which we compare inference results associated with each interpretation.
In domains where sample sizes are limited, efficient learning algorithms are critical. Learning using privileged information (LuPI) offers increased sample efficiency by allowing prediction models access to auxiliary information at training time which is unavailable when the models are used. In recent work, it was shown that for prediction in linear-Gaussian dynamical systems, a LuPI learner with access to intermediate time series data is never worse and often better in expectation than any unbiased classical learner. We provide new insights into this analysis and generalize it to nonlinear prediction tasks in latent dynamical systems, extending theoretical guarantees to the case where the map connecting latent variables and observations is known up to a linear transform. In addition, we propose algorithms based on random features and representation learning for the case when this map is unknown. A suite of empirical results confirm theoretical findings and show the potential of using privileged time-series information in nonlinear prediction.
It has been shown that deep neural networks are prone to overfitting on biased training data. Towards addressing this issue, meta-learning employs a meta model for correcting the training bias. Despite the promising performances, super slow training is currently the bottleneck in the meta learning approaches. In this paper, we introduce a novel Faster Meta Update Strategy (FaMUS) to replace the most expensive step in the meta gradient computation with a faster layer-wise approximation. We empirically find that FaMUS yields not only a reasonably accurate but also a low-variance approximation of the meta gradient. We conduct extensive experiments to verify the proposed method on two tasks. We show our method is able to save two-thirds of the training time while still maintaining the comparable or achieving even better generalization performance. In particular, our method achieves the state-of-the-art performance on both synthetic and realistic noisy labels, and obtains promising performance on long-tailed recognition on standard benchmarks.
Ensembles over neural network weights trained from different random initialization, known as deep ensembles, achieve state-of-the-art accuracy and calibration. The recently introduced batch ensembles provide a drop-in replacement that is more parameter efficient. In this paper, we design ensembles not only over weights, but over hyperparameters to improve the state of the art in both settings. For best performance independent of budget, we propose hyper-deep ensembles, a simple procedure that involves a random search over different hyperparameters, themselves stratified across multiple random initializations. Its strong performance highlights the benefit of combining models with both weight and hyperparameter diversity. We further propose a parameter efficient version, hyper-batch ensembles, which builds on the layer structure of batch ensembles and self-tuning networks. The computational and memory costs of our method are notably lower than typical ensembles. On image classification tasks, with MLP, LeNet, and Wide ResNet 28-10 architectures, our methodology improves upon both deep and batch ensembles.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191