Search for: All records

We initiate the study of a fundamental question concerning adversarial noise models in statistical problems where the algorithm receives i.i.d. draws from a distribution D. The definitions of these adversaries specify the {\sl type} of allowable corruptions (noise model) as well as {\sl when} these corruptions can be made (adaptivity); the latter differentiates between oblivious adversaries that can only corrupt the distribution D and adaptive adversaries that can have their corruptions depend on the specific sample S that is drawn from D. We investigate whether oblivious adversaries are effectively equivalent to adaptive adversaries, across all noise models studied in the literature, under a unifying framework that we introduce. Specifically, can the behavior of an algorithm A in the presence of oblivious adversaries always be well-approximated by that of an algorithm A′ in the presence of adaptive adversaries? Our first result shows that this is indeed the case for the broad class of {\sl statistical query} algorithms, under all reasonable noise models. We then show that in the specific case of {\sl additive noise}, this equivalence holds for {\sl all} algorithms. Finally, we map out an approach towards proving this statement in its fullest generality, for all algorithms and under all reasonable noise models. 

Using the framework of boosting, we prove that all impurity-based decision tree learning algorithms, including the classic ID3, C4.5, and CART, are highly noise tolerant. Our guarantees hold under the strongest noise model of nasty noise, and we provide near-matching upper and lower bounds on the allowable noise rate. We further show that these algorithms, which are simple and have long been central to everyday machine learning, enjoy provable guarantees in the noisy setting that are unmatched by existing algorithms in the theoretical literature on decision tree learning. Taken together, our results add to an ongoing line of research that seeks to place the empirical success of these practical decision tree algorithms on firm theoretical footing. 

The Gulf of Alaska is rimmed by a Mesozoic- Cenozoic accretionary wedge complex comprised of the Chugach and the Prince William terranes. This study focuses on understanding the history and relationship between turbidites of the Chugach and Prince William terranes in the Prince William Sound area: the Campanian-Maastrichtian Valdez Group and the Paleocene-Eocene Orca Group. Critical in understanding this system is the Contact fault system, a poorly understood collection of fault strands that has traditionally been viewed as the terrane boundary between the two (Winkler and Plafker 1975; Plafker et al. 1977; Tysdal and Case 1979; Dumoulin 1988; Bol and Gibbons 1992; Bol and Roeske 1993; Arkle et al. 2013). It was thought to be an original accretionary fault that separates terranes (Plafker et al. 1977; Nokleberg et al., 1986; Dumoulin, 1988); however, more recent work has characterized it as a sequence of faults that may be related to internal deformation of the accretionary complex (Bol and Gibbons, 1992; Bol and Roeske, 1993). The main goal of this study is to determine whether the Contact fault functions as a terrane boundary across Prince William Sound by presenting new detrital zircon U-Pb ages from either side of the fault system (Fig. 1). First, we specifically focus on understanding the age relationships between the Valdez Group and the Orca Group on either side of the Jack Bay fault and the Landlock fault in Valdez Arm, northeastern Prince William Sound (Fig. 2). Second, we focus on strands of the Contact fault system in three other locations: Unakwik Inlet, Kings Bay, and Seward (Fig. 3). This second goal of this study is to determine the nature of the Landlock block, an area currently defined by the convergence of the Jack Bay and Landlock faults (Fig. 2). Early maps portrayed the Landlock block as Valdez Group (Moffit, 1954; Winkler and Plafker, 1975), but the current location of the Contact fault in Jack Bay includes the block as part of the Orca Group (Dumoulin, 1998; Bol and Roeske, 1993). Therefore, an important question is the affinity of the rocks of the Landlock block, whether they are Orca Group or Valdez Group. 

Community engagement efforts have become an important avenue for raising public interest and know-how related to engineering. These efforts draw the young and the diverse into seeing engineering as a worthwhile profession. One such effort at the national level in the U.S. is the “National Engineers Week”. This is a week-long celebration held every February that consists of numerous events and activities organized for the general public with a focus towards students, women, and under-represented groups. In this paper, we examined this effort through the lens of social media and analyzed Twitter data collected for two hashtags used during the National Engineers Week 2017: “#eweek2017” and “#engineersweek”. Our dataset consisted of 6,583 original tweets and 10,885 retweets. To study the impact of the outreach we used three analytical approaches: descriptive analysis, content analysis, and network analysis. We found that the Twitter campaign participation was dominated by engineering companies and individual users followed by a limited participation of educational institutions, professional engineering associations, and non-profits. As opposed to other popular hashtag campaigns, not a single news media organization was identified as a participating user signaling a lower new media-driven propagation of the campaign among the public. From a content perspective, the tweets can be categorized as event promotion, showcasing employees of engineering companies, or encouraging and inspiring public (especially women and children) towards engineering. With the growing popularity of social media, community engagement efforts need to strategically leverage hashtags and other media elements for a broader impact. 

Workforce development in engineering is a high priority to keep pace with innovation and change within engineering disciplines and also within organizations. Increasingly, workforce development requires more retraining and retooling of employees than ever before as information technology has accelerated both the creation of a new body of knowledge and also the skills required to perform the work. In this paper we present a field study of a highly dynamic workplace – a cybersecurity firm – undertaken to better understand how engineers keep up with the pace of knowledge that is needed for their work. Fifteen professionals, with a wide range of experience and educational background, were interviewed. Data were analyzed iteratively and interpretively. The findings from the study suggest that over time some well-defined ways of learning had developed in the workplace we studied. These learning practices combined in-person and online interactions and resources. We also found that learning was triggered largely by the need to solve a problem or by the interests of the engineers to learn more in order to be prepared for new knowledge in the field. Depending on the problem they faced, the engineers mapped the requirements of what was needed to solve the problem, identified the resources that were available, and then selected the optimal resource. Often, as is common with problem solving, our participants had to try out multiple options. Theoretically, our study contributes by integrating an information seeking perspective with situated cognition to inform future studies of learning in information rich engineering and technology workplaces. 

Each year, significant investment of time and resources is made to improve diversity within engineering across a range of federal and state agencies, private/not-for-profit organizations, and foundations. In spite of decades of investments, efforts have not yielded desired returns - participation by minorities continues to lag at a time when STEM workforce requirements are increasing. In recent years a new stream of data has emerged - online social networks, including Twitter, Facebook, and Instagram - that act as a key sensor of social behavior and attitudes of the public. Almost 87% of the American population now participates in some form of social media activity. Consequently, social networking sites have become powerful indicators of social action and social media data has shown significant promise for studying many issues including public health communication, political campaign, humanitarian crisis, and, activism. We argue that social media data can likewise be leveraged to better understand and improve engineering diversity. As a case study to illustrate the viability of the approach, we present findings from a campaign, #ILookLikeAnEngineer (using Twitter data – 19,354 original tweets and 29,529 retweets), aimed at increasing gender diversity in the engineering workplace. The campaign provided a continuous momentum to the overall effort to increase diversity and novel ways of connecting with relevant audience. Our analysis demonstrates that diversity initiatives related to STEM attract voices from various entities including individuals, large corporations, media outlets, and community interest groups. 

We present a study that examines how a social media activism campaign aimed at improving gender diversity within engineering gained and maintained momentum in its early period. We examined over 50,000 Tweets posted over the first ~75 days of the #ILookLikeAnEngineer campaign and found that diverse participation – of types of users – increased activity at crucial moments. We categorize these triggers into four types: 1) Event-Driven: Alignment of the campaign with offline events related to the issue (Diversity SFO, Disrupt, etc.); 2) Media-Driven: News coverage of the events in the media (TechCrunch, CNN, BBC, etc.); 3) Industry-Driven: Web participation in the campaign by large organizations (Microsoft, Tesla, GE, Cisco, etc.); and 4) Personality-Driven: Alignment of the events with popular and/or known personalities (e.g. Isis Anchalee; Michelle Sun; Ada Lovelace.) This study illustrates how one mechanism – triggering – supports connective action in social media campaign.