arXiv subjects

Camelia Simoiu

Publications and source records attributed to Camelia Simoiu.

5 recordsLinked to original sources

arXiv2022

Modeling road user response timing in naturalistic settings: a surprise-based framework

There is currently no established method for evaluating human response timing across a range of naturalistic traffic conflict types. Traditional notions derived from controlled experiments, such as perception-response time, fail to account for the situation-dependency of human responses and offer no clear way to define the stimulus in many common traffic conflict scenarios. As a result, they are not well suited for application in naturalistic settings. Our main contribution is the development of a novel framework for measuring and modeling response times in naturalistic traffic conflicts applicable to automated driving systems as well as other traffic safety domains. The framework suggests that response timing must be understood relative to the subject's current (prior) belief and is always embedded in, and dependent on, the dynamically evolving situation. The response process is modeled as a belief update process driven by perceived violations to this prior belief, that is, by surprising stimuli. The framework resolves two key limitations with traditional notions of response time when applied in naturalistic scenarios: (1) The strong situation-dependence of response timing and (2) how to unambiguously define the stimulus. Resolving these issues is a challenge that must be addressed by any response timing model intended to be applied in naturalistic traffic conflicts. We show how the framework can be implemented by means of a relatively simple heuristic model fit to naturalistic human response data from real crashes and near crashes from the SHRP2 dataset and discuss how it is, in principle, generalizable to any traffic conflict scenario. We also discuss how the response timing framework can be implemented computationally based on evidence accumulation enhanced by machine learning-based generative models and the information-theoretic concept of surprise.

cs.HC↗

arXiv2021

An Empirical Analysis of HTTPS Configuration Security

It is notoriously difficult to securely configure HTTPS, and poor server configurations have contributed to several attacks including the FREAK, Logjam, and POODLE attacks. In this work, we empirically evaluate the TLS security posture of popular websites and endeavor to understand the configuration decisions that operators make. We correlate several sources of influence on sites' security postures, including software defaults, cloud providers, and online recommendations. We find a fragmented web ecosystem: while most websites have secure configurations, this is largely due to major cloud providers that offer secure defaults. Individually configured servers are more often insecure than not. This may be in part because common resources available to individual operators -- server software defaults and online configuration guides -- are frequently insecure. Our findings highlight the importance of considering SaaS services separately from individually-configured sites in measurement studies, and the need for server software to ship with secure defaults.

cs.CR↗

arXiv2017

The Problem of Infra-marginality in Outcome Tests for Discrimination

Outcome tests are a popular method for detecting bias in lending, hiring, and policing decisions. These tests operate by comparing the success rate of decisions across groups. For example, if loans made to minority applicants are observed to be repaid more often than loans made to whites, it suggests that only exceptionally qualified minorities are granted loans, indicating discrimination. Outcome tests, however, are known to suffer from the problem of infra-marginality: even absent discrimination, the repayment rates for minority and white loan recipients might differ if the two groups have different risk distributions. Thus, at least in theory, outcome tests can fail to accurately detect discrimination. We develop a new statistical test of discrimination---the threshold test---that mitigates the problem of infra-marginality by jointly estimating decision thresholds and risk distributions via a hierarchical Bayesian latent variable model. Applying our test to a dataset of 4.5 million police stops in North Carolina, we find that the problem of infra-marginality is more than a theoretical possibility, and can cause the outcome test to yield misleading results in practice.

stat.AP↗

arXiv2017

A large-scale analysis of racial disparities in police stops across the United States

To assess racial disparities in police interactions with the public, we compiled and analyzed a dataset detailing over 60 million state patrol stops conducted in 20 U.S. states between 2011 and 2015. We find that black drivers are stopped more often than white drivers relative to their share of the driving-age population, but that Hispanic drivers are stopped less often than whites. Among stopped drivers -- and after controlling for age, gender, time, and location -- blacks and Hispanics are more likely to be ticketed, searched, and arrested than white drivers. These disparities may reflect differences in driving behavior, and are not necessarily the result of bias. In the case of search decisions, we explicitly test for discrimination by examining both the rate at which drivers are searched and the likelihood searches turn up contraband. We find evidence that the bar for searching black and Hispanic drivers is lower than for searching whites. Finally, we find that legalizing recreational marijuana in Washington and Colorado reduced the total number of searches and misdemeanors for all race groups, though a race gap still persists. We conclude by offering recommendations for improving data collection, analysis, and reporting by law enforcement agencies.

stat.AP↗

arXiv2016

Unsupervised Learning For Effective User Engagement on Social Media

In this paper, we investigate the effectiveness of unsupervised feature learning techniques in predicting user engagement on social media. Specifically, we compare two methods to predict the number of feedbacks (i.e., comments) that a blog post is likely to receive. We compare Principal Component Analysis (PCA) and sparse Autoencoder to a baseline method where the data are only centered and scaled, on each of two models: Linear Regression and Regression Tree. We find that unsupervised learning techniques significantly improve the prediction accuracy on both models. For the Linear Regression model, sparse Autoencoder achieves the best result, with an improvement in the root mean squared error (RMSE) on the test set of 42% over the baseline method. For the Regression Tree model, PCA achieves the best result, with an improvement in RMSE of 15% over the baseline.

cs.LG↗