We present a new family of information-theoretic generalization bounds, in which the training loss and the population loss are compared through a jointly convex function. This function is upper-bounded in terms of the disintegrated, samplewise, evaluated conditional mutual information (CMI), an information measure that depends on the losses incurred by the selected hypothesis, rather than on the hypothesis itself, as is common in probably approximately correct (PAC)-Bayesian results. We demonstrate the generality of this framework by recovering and extending previously known information-theoretic bounds. Furthermore, using the evaluated CMI, we derive a samplewise, average version of Seeger's PAC-Bayesian bound, where the convex function is the binary KL divergence. In some scenarios, this novel bound results in a tighter characterization of the population loss of deep neural networks than previous bounds. Finally, we derive high-probability versions of some of these average bounds. We demonstrate the unifying nature of the evaluated CMI bounds by using them to recover average and high-probability generalization bounds for multiclass classification with finite Natarajan dimension.
相關內容
We propose a multicountry quantile factor augmeneted vector autoregression (QFAVAR) to model heterogeneities both across countries and across characteristics of the distributions of macroeconomic time series. The presence of quantile factors allows for summarizing these two heterogeneities in a parsimonious way. We develop two algorithms for posterior inference that feature varying level of trade-off between estimation precision and computational speed. Using monthly data for the euro area, we establish the good empirical properties of the QFAVAR as a tool for assessing the effects of global shocks on country-level macroeconomic risks. In particular, QFAVAR short-run tail forecasts are more accurate compared to a FAVAR with symmetric Gaussian errors, as well as univariate quantile autoregressions that ignore comovements among quantiles of macroeconomic variables. We also illustrate how quantile impulse response functions and quantile connectedness measures, resulting from the new model, can be used to implement joint risk scenario analysis.
The Adaptive Large Neighborhood Search (ALNS) algorithm has shown considerable success in solving complex combinatorial optimization problems (COPs). ALNS selects various heuristics adaptively during the search process, leveraging their strengths to find good solutions for optimization problems. However, the effectiveness of ALNS depends on the proper configuration of its selection and acceptance parameters. To address this limitation, we propose a Deep Reinforcement Learning (DRL) approach that selects heuristics, adjusts parameters, and controls the acceptance criteria during the search process. The proposed method aims to learn, based on the state of the search, how to configure the next iteration of the ALNS to obtain good solutions to the underlying optimization problem. We evaluate the proposed method on a time-dependent orienteering problem with stochastic weights and time windows, used in an IJCAI competition. The results show that our approach outperforms vanilla ALNS and ALNS tuned with Bayesian Optimization. In addition, it obtained better solutions than two state-of-the-art DRL approaches, which are the winning methods of the competition, with much fewer observations required for training. The implementation of our approach will be made publicly available.
Under some regularity assumptions, we report an a priori error analysis of a dG scheme for the Poisson and Stokes flow problem in their dual mixed formulation. Both formulations satisfy a Babu\v{s}ka-Brezzi type condition within the space H(div) x L2. It is well known that the lowest order Crouzeix-Raviart element paired with piecewise constants satisfies such a condition on (broken) H1 x L2 spaces. In the present article, we use this pair. The continuity of the normal component is weakly imposed by penalizing jumps of the broken H(div) component. For the resulting methods, we prove well-posedness and convergence with constants independent of data and mesh size. We report error estimates in the methods natural norms and optimal local error estimates for the divergence error. In fact, our finite element solution shares for each triangle one DOF with the CR interpolant and the divergence is locally the best-approximation for any regularity. Numerical experiments support the findings and suggest that the other errors converge optimally even for the lowest regularity solutions and a crack-problem, as long as the crack is resolved by the mesh.
We consider a linear model which can have a large number of explanatory variables, the errors with an asymmetric distribution or some values of the explained variable are missing at random. In order to take in account these several situations, we consider the non parametric empirical likelihood (EL) estimation method. Because a constraint in EL contains an indicator function then a smoothed function instead of the indicator will be considered. Two smoothed expectile maximum EL methods are proposed, one of which will automatically select the explanatory variables. For each of the methods we obtain the convergence rate of the estimators and their asymptotic normality. The smoothed expectile empirical log-likelihood ratio process follow asymptotically a chi-square distribution and moreover the adaptive LASSO smoothed expectile maximum EL estimator satisfies the sparsity property which guarantees the automatic selection of zero model coefficients. In order to implement these methods, we propose four algorithms.
We present new adaptive learning rates that can be used with any momentum method. To showcase our new learning rates we develop MoMo and MoMo-Adam, which are SGD with momentum (SGDM) and Adam together with our new adaptive learning rates. Our MoMo methods are motivated through model-based stochastic optimization, wherein we use momentum estimates of the batch losses and gradients sampled at each iteration to build a model of the loss function. Our model also makes use of any known lower bound of the loss function by using truncation. Indeed most losses are bounded below by zero. We then approximately minimize this model at each iteration to compute the next step. For losses with unknown lower bounds, we develop new on-the-fly estimates of the lower bound that we use in our model. Numerical experiments show that our MoMo methods improve over SGDM and Adam in terms of accuracy and robustness to hyperparameter tuning for training image classifiers on MNIST, CIFAR10, CIFAR100, Imagenet32, DLRM on the Criteo dataset, and a transformer model on the translation task IWSLT14.
Tracking Cartesian motion with end~effectors is a fundamental task in robot control. For motion that is not known in advance, the solvers must find fast solutions to the inverse kinematics (IK) problem for discretely sampled target poses. On joint control level, however, the robot's actuators operate in a continuous domain, requiring smooth transitions between individual states. In this work, we present a boost to the well-known Jacobian transpose method to address this goal, using the mass matrix of a virtually conditioned twin of the manipulator. Results on the UR10 show superior convergence and quality of our dynamics-based solver against the plain Jacobian method. Our algorithm is straightforward to implement as a controller, using common robotics libraries.
Access to individual-level health data is essential for gaining new insights and advancing science. In particular, modern methods based on artificial intelligence rely on the availability of and access to large datasets. In the health sector, access to individual-level data is often challenging due to privacy concerns. A promising alternative is the generation of fully synthetic data, i.e. data generated through a randomised process that have similar statistical properties as the original data, but do not have a one-to-one correspondence with the original individual-level records. In this study, we use a state-of-the-art synthetic data generation method and perform in-depth quality analyses of the generated data for a specific use case in the field of nutrition. We demonstrate the need for careful analyses of synthetic data that go beyond descriptive statistics and provide valuable insights into how to realise the full potential of synthetic datasets. By extending the methods, but also by thoroughly analysing the effects of sampling from a trained model, we are able to largely reproduce significant real-world analysis results in the chosen use case.
The Nash Equilibrium (NE) estimation in bidding games of electricity markets is the key concern of both generation companies (GENCOs) for bidding strategy optimization and the Independent System Operator (ISO) for market surveillance. However, existing methods for NE estimation in emerging modern electricity markets (FEM) are inaccurate and inefficient because the priori knowledge of bidding strategies before any environment changes, such as load demand variations, network congestion, and modifications of market design, is not fully utilized. In this paper, a Bayes-adaptive Markov Decision Process in FEM (BAMDP-FEM) is therefore developed to model the GENCOs' bidding strategy optimization considering the priori knowledge. A novel Multi-Agent Generative Adversarial Imitation Learning algorithm (MAGAIL-FEM) is then proposed to enable GENCOs to learn simultaneously from priori knowledge and interactions with changing environments. The obtained NE is a Bayesian Nash Equilibrium (BNE) with priori knowledge transferred from the previous environment. In the case study, the superiority of this proposed algorithm in terms of convergence speed compared with conventional methods is verified. It is concluded that the optimal bidding strategies in the obtained BNE can always lead to more profits than NE due to the effective learning from the priori knowledge. Also, BNE is more accurate and consistent with situations in real-world markets.
To address increasing societal concerns regarding privacy and climate, the EU adopted the General Data Protection Regulation (GDPR) and committed to the Green Deal. Considerable research studied the energy efficiency of software and the accuracy of machine learning models trained on anonymised data sets. Recent work began exploring the impact of privacy-enhancing techniques (PET) on both the energy consumption and accuracy of the machine learning models, focusing on k-anonymity. As synthetic data is becoming an increasingly popular PET, this paper analyses the energy consumption and accuracy of two phases: a) applying privacy-enhancing techniques to the concerned data set, b) training the models on the concerned privacy-enhanced data set. We use two privacy-enhancing techniques: k-anonymisation (using generalisation and suppression) and synthetic data, and three machine-learning models. Each model is trained on each privacy-enhanced data set. Our results show that models trained on k-anonymised data consume less energy than models trained on the original data, with a similar performance regarding accuracy. Models trained on synthetic data have a similar energy consumption and a similar to lower accuracy compared to models trained on the original data.
Invariant approaches have been remarkably successful in tackling the problem of domain generalization, where the objective is to perform inference on data distributions different from those used in training. In our work, we investigate whether it is possible to leverage domain information from the unseen test samples themselves. We propose a domain-adaptive approach consisting of two steps: a) we first learn a discriminative domain embedding from unsupervised training examples, and b) use this domain embedding as supplementary information to build a domain-adaptive model, that takes both the input as well as its domain into account while making predictions. For unseen domains, our method simply uses few unlabelled test examples to construct the domain embedding. This enables adaptive classification on any unseen domain. Our approach achieves state-of-the-art performance on various domain generalization benchmarks. In addition, we introduce the first real-world, large-scale domain generalization benchmark, Geo-YFCC, containing 1.1M samples over 40 training, 7 validation, and 15 test domains, orders of magnitude larger than prior work. We show that the existing approaches either do not scale to this dataset or underperform compared to the simple baseline of training a model on the union of data from all training domains. In contrast, our approach achieves a significant improvement.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191