Types Of Uncertainty Research Articles

ANEDL: Adaptive Negative Evidential Deep Learning for Open-Set Semi-supervised Learning

Semi-supervised learning (SSL) methods assume that labeled data, unlabeled data and test data are from the same distribution. Open-set semi-supervised learning (Open-set SSL) con- siders a more practical scenario, where unlabeled data and test data contain new categories (outliers) not observed in labeled data (inliers). Most previous works focused on out- lier detection via binary classifiers, which suffer from insufficient scalability and inability to distinguish different types of uncertainty. In this paper, we propose a novel framework, Adaptive Negative Evidential Deep Learning (ANEDL) to tackle these limitations. Concretely, we first introduce evidential deep learning (EDL) as an outlier detector to quantify different types of uncertainty, and design different uncertainty metrics for self-training and inference. Furthermore, we propose a novel adaptive negative optimization strategy, making EDL more tailored to the unlabeled dataset containing both inliers and outliers. As demonstrated empirically, our proposed method outperforms existing state-of-the-art methods across four datasets.

Probabilistic seismic hazard analysis using reliability methods

By considering uncertainties in the input parameters (e.g., magnitude, location, wave path, etc.), the Probabilistic Seismic Hazard Analysis (PSHA) aims to compute annual rate of various exceeding ground motions at a site or a map of sites of all anticipated given earthquakes. Uncertainties may be originated due to inherent randomness of the phenomena or variability in the mean values of di erent models parameters, mainly due to use of nite-sample size of observations. The rst, in literature reviews, is commonly named aleatory uncertainty; the second is known as epistemic uncertainty. The total probability numerical integration, generally employed to calculate PSHA, only considers aleatory uncertainties, and variability in the models' parameters is neglected to simplify calculation. In this paper, as an alternative to the total probability numerical integration, matured and standard reliability methods tailored to e ortlessly consider both types ofuncertainties are put forward to compute site-speci c PSHA. Then, as an application study, the peak ground acceleration hazard curve for the site, at which a historical bridge is located, is developed and compared with those obtained from the total probability numerical integration.

Dealing With Uncertainty in Policy Analysis and Policymaking

‘‘Penetrating so many secrets,we cease to believe in the unknowable.But there it sits nevertheless,calmly licking its chops.’’– H.L. Mencken, Minority Report, 1956Policymaking is about the future. If we were able to predictthe future accurately, preferred policies could be identified(at least in principle) by simply examining the future thatwould follow from the implementation of each possiblepolicy and picking the one that produced the most favorableoutcomes. However, for most systems of interest today(particularly social and economic systems), such predictionis not possible, due to their increasing complexity, theirincreasing interrelationships with other systems, and theincreasing uncertainty of developments external to thesystem that have important effects on the system. Wheneven the best model cannot reliably predict the details of asystem’s behavior, the classical approach of choosing apolicy based on the outcomes from a best estimate model isno longer credible. Such policies are ‘best’ for a future thatmost certainly will not occur, and have implications for thefuture that actually occurs that are typically not examined inthe course of policy design and analysis. Current approachesto policy analysis have serious difficulties in dealing withproblems characterized by complexity or disequilibriumbehavior, systems undergoing significant organizational andstructural change, and systems that can only be influencedrather than controlled. Yet, these characteristics haveincreasingly become staple characteristics of the world inwhich we live. Such systems are fundamentally unpredict-able. Yet, rapid economic, political, and social changes are areality, and public policies must be devised in spite ofprofound uncertainties about the future.Even though the future cannot be predicted, it is possibleto prepare for it. If, in the face of massive uncertainties,public policies are to be useful and credible, new approacheswill be needed for dealing with uncertainty. This specialissue of Integrated Assessment is a first step in filling thisneed. The papers in this issue are drawn from a sessionentitled ‘‘Dealing With Uncertainty in Policy Analysis andPolicymaking’’ that was part of the 5th InternationalConference on Technology, Policy and Innovation, whichwas held in The Hague in June 2001.There are five papers in this collection. They can bedivided into two categories:1. How can uncertainty in policy analysis and policymakingbe characterized (what is it? how can it be placed in ahistorical context? how can we classify different types ofuncertainties?)?2. How can policy analysts and policymakers deal withuncertainties (i.e., how can policies be developed thathave a good chance of succeeding in spite of enormousuncertainties about the future?)?1. CHARACTERIZING UNCERTAINTYThat uncertainties exist in practically all policymakingsituations is generally understood by policymakers andpolicy analysts. But there is little appreciation for the factthat there are many different types of uncertainty, and thereis a lack of understanding about their relative magnitudesand the different tools that are appropriate to use for dealingwith the different types. Even within the community ofpolicy analysts who deal with uncertainty in their work,there is no commonly shared terminology, and no agreementon a typology of uncertainties. The first paper (by Walker,Harrem€ooes, Rotmans, van der Sluijs, van Asselt, Janssen,and von Krauss) aims to provide a conceptual basis for thesystematic classification of uncertainty in model-baseddecision support activities, such as policy analysis, inte-grated assessment, and risk assessment. As van Asselt [1]notes, any typology of uncertainties is context dependent. Infact, according to her, uncertainty type by definition ‘‘refersto the way in which uncertainty manifests itself in aparticular context.’’ The context for the typology ofuncertainty presented in this paper is model-based decisionsupport. The authors first define uncertainty in model-based

Uncertain reasoning about agents' beliefs and reasoning

Reasoning about mental states and processes is important in varioussubareas of the legal domain. A trial lawyer might need to reason andthe beliefs, reasoning and other mental states and processes of membersof a jury; a police officer might need to reason about the conjecturedbeliefs and reasoning of perpetrators; a judge may need to consider adefendant's mental states and processes for the purposes of sentencing;and so on. Further, the mental states in question may themselves beabout the mental states and processes of other people. Therefore, if AIsystems are to assist with reasoning tasks in law, they may need to beable to reason about mental states and processes. Such reasoning isriddled with uncertainty, and this is true in particular in the legaldomain. The article discusses how various different types ofuncertainty arise, and shows how they greatly complicate the task ofreasoning about mental states and processes. The article concentrates onthe special case of states of belief and processes of reasoning, andsketches an implemented, prototype computer program (ATT-Meta) thatcopes with the various types of uncertainty in reasoning about beliefsand reasoning. In particular, the article outlines the system'sfacilities for handling conflict between different lines of argument,especially when these lie within the reasoning of different people. Thesystem's approach is illustrated by application to a real-life muggingexample.