The potential for structural errors in emergent constraints

Studies of emergent constraints have frequently proposed that a single metric can constrain future responses of the Earth system to anthropogenic emissions. Here, we illustrate that strong relationships between observables and future climate across an ensemble can arise from common structural model assumptions with few degrees of freedom. Such cases have the potential to produce strong yet overconfident constraints when processes are represented in a common, oversimplified fashion throughout the ensemble. We consider these issues in the context of a collection of published constraints and argue that although emergent constraints are potentially powerful tools for understanding ensemble response variation and relevant observables, their naïve application to reduce uncertainties in unknown climate responses could lead to bias and overconfidence in constrained projections. The prevalence of this thinking has led to literature in which statements are made on the probability bounds of key climate variables that were confident yet inconsistent between studies. Together with statistical robustness and a mechanism, assessments of climate responses must include multiple lines of evidence to identify biases that can arise from shared, oversimplified modelling assumptions that impact both present and future climate simulations in order to mitigate against the influence of shared structural biases.

Vectorial integer bootstrapping: flexible integer estimation with application to GNSS

In this contribution, we extend the principle of integer bootstrapping (IB) to a vectorial form (VIB). The mathematical definition of the class of VIB-estimators is introduced together with their pull-in regions and other properties such as probability bounds and success rate approximations. The vectorial formulation allows sequential block-by-block processing of the ambiguities based on a user-chosen partitioning. In this way, flexibility is created, where for specific choices of partitioning, tailored VIB-estimators can be designed. This wide range of possibilities is discussed, supported by numerical simulations and analytical examples. Further guidelines are provided, as well as the possible extension to other classes of estimators.

Fine-grained Generalization Analysis of Structured Output Prediction

In machine learning we often encounter structured output prediction problems (SOPPs), i.e. problems where the output space admits a rich internal structure. Application domains where SOPPs naturally occur include natural language processing, speech recognition, and computer vision. Typical SOPPs have an extremely large label set, which grows exponentially as a function of the size of the output. Existing generalization analysis implies generalization bounds with at least a square-root dependency on the cardinality d of the label set, which can be vacuous in practice. In this paper, we significantly improve the state of the art by developing novel high-probability bounds with a logarithmic dependency on d. Furthermore, we leverage the lens of algorithmic stability to develop generalization bounds in expectation without any dependency on d. Our results therefore build a solid theoretical foundation for learning in large-scale SOPPs. Furthermore, we extend our results to learning with weakly dependent data.

A Pairing-Free Identity-Based Identification Scheme with Tight Security Using Modified-Schnorr Signatures

The security of cryptographic schemes is proven secure by reducing an attacker which breaks the scheme to an algorithm that could be used to solve the underlying hard assumption (e.g., Discrete Logarithm, Decisional Diffie–Hellman). The reduction is considered tight if it results in approximately similar probability bounds to that of solving the underlying hard assumption. Tight security is desirable as it improves security guarantees and allows the use of shorter parameters without the risk of compromising security. In this work, we propose an identity-based identification (IBI) scheme with tight security based on a variant of the Schnorr signature scheme known as TNC signatures. The proposed IBI scheme enjoys shorter parameters and key sizes as compared to existing IBI schemes without increasing the number of operations required for its identification protocol. Our scheme is suitable to be used for lightweight authentication in resource-constrained Wireless Sensor Networks (WSNs) as it utilizes the lowest amount of bandwidth when compared to other state-of-the-art symmetric key lightweight authentication schemes. Although it is costlier than its symmetric key counterparts in terms of operational costs due to its asymmetric key nature, it enjoys other benefits such as decentralized authentication and scalable key management. As a proof of concept to substantiate our claims, we perform an implementation of our scheme to demonstrate its speed and memory usage when it runs on both high and low-end devices.

On structural errors in emergent constraints

Studies of emergent constraints have frequently proposed that a single metric alone can constrain future responses of the Earth system to anthropogenic emissions. The prevalence of this thinking has led to literature and messaging which is sometimes confusing to policymakers, with a series of studies over the last decade making confident, yet contradictory, claims on the probability bounds of key climate variables. Here, we illustrate that emergent constraints are more likely to occur where the variance across an ensemble of climate models of both observable and future climate arises from common structural assumptions and few degrees of freedom. Such cases are likely to occur when processes are represented in a common, oversimplified fashion throughout the ensemble, about which we have the least confidence in performance out of sample. We consider these issues in the context of a number of published constraints, and argue that the application of emergent constraints alone to estimate uncertainties in unknown climate responses can potentially lead to bias and overconfidence in constrained projections. Together with statistical robustness and plausibility of mechanism, assessments of climate responses must include multiple lines of evidence to identify biases that arise from common oversimplified modeling assumptions which impact both present and future climate simulations in order to mitigate against the influence of common structural biases.

Deep Neural Networks for Estimation and Inference

We study deep neural networks and their use in semiparametric inference. We establish novel nonasymptotic high probability bounds for deep feedforward neural nets. These deliver rates of convergence that are sufficiently fast (in some cases minimax optimal) to allow us to establish valid second‐step inference after first‐step estimation with deep learning, a result also new to the literature. Our nonasymptotic high probability bounds, and the subsequent semiparametric inference, treat the current standard architecture: fully connected feedforward neural networks (multilayer perceptrons), with the now‐common rectified linear unit activation function, unbounded weights, and a depth explicitly diverging with the sample size. We discuss other architectures as well, including fixed‐width, very deep networks. We establish the nonasymptotic bounds for these deep nets for a general class of nonparametric regression‐type loss functions, which includes as special cases least squares, logistic regression, and other generalized linear models. We then apply our theory to develop semiparametric inference, focusing on causal parameters for concreteness, and demonstrate the effectiveness of deep learning with an empirical application to direct mail marketing.