Variance in predictions across different trained models is a significant, under-explored source of error in fair binary classification. In practice, the variance on some data examples is so large that decisions can be effectively arbitrary. To investigate this problem, we take an experimental approach and make four overarching contributions. We: 1) Define a metric called self-consistency, derived from variance, which we use as a proxy for measuring and reducing arbitrariness; 2) Develop an ensembling algorithm that abstains from classification when a prediction would be arbitrary; 3) Conduct the largest to-date empirical study of the role of variance (vis-a-vis self-consistency and arbitrariness) in fair binary classification; and, 4) Release a toolkit that makes the US Home Mortgage Disclosure Act (HMDA) datasets easily usable for future research. Altogether, our experiments reveal shocking insights about the reliability of conclusions on benchmark datasets. Most fair binary classification benchmarks are close-to-fair when taking into account the amount of arbitrariness present in predictions -- before we even try to apply any fairness interventions. This finding calls into question the practical utility of common algorithmic fairness methods, and in turn suggests that we should reconsider how we choose to measure fairness in binary classification.

Labor-saving technology has already decreased employment opportunities for middle-skill workers. Experts anticipate that advances in AI and robotics will cause even more significant disruptions in the labor market over the next two decades. This paper presents three experimental studies that investigate how this profound economic change could affect mass politics. Recent observational studies suggest that workers' exposure to automation risk predicts their support not only for redistribution but also for right-wing populist policies and candidates. Other observational studies, including my own, find that workers underestimate the impact of automation on their job security. Misdirected blame towards immigrants and workers in foreign countries, rather than concerns about workplace automation, could be driving support for right-wing populism. To correct American workers' beliefs about the threats to their jobs, I conducted three survey experiments in which I informed workers about the existent and future impact of workplace automation. While these informational treatments convinced workers that automation threatens American jobs, they failed to change respondents' preferences on welfare, immigration, and trade policies. My research finds that raising awareness about workplace automation did not decrease opposition to globalization or increase support for policies that will prepare workers for future technological disruptions.

Machine learning (ML) and artificial intelligence (AI) researchers play an important role in the ethics and governance of AI, including through their work, advocacy, and choice of employment. Nevertheless, this influential group's attitudes are not well understood, undermining our ability to discern consensuses or disagreements between AI/ML researchers. To examine these researchers' views, we conducted a survey of those who published in two top AI/ML conferences (N = 524). We compare these results with those from a 2016 survey of AI/ML researchers and a 2018 survey of the US public. We find that AI/ML researchers place high levels of trust in international organizations and scientific organizations to shape the development and use of AI in the public interest; moderate trust in most Western tech companies; and low trust in national militaries, Chinese tech companies, and Facebook. While the respondents were overwhelmingly opposed to AI/ML researchers working on lethal autonomous weapons, they are less opposed to researchers working on other military applications of AI, particularly logistics algorithms. A strong majority of respondents think that AI safety research should be prioritized more and a majority that ML institutions should conduct pre-publication review to assess potential harms. Being closer to the technology itself, AI/ML researchers are well placed to highlight new risks and develop technical solutions, so this novel data has broad relevance. The findings should help to improve how researchers, private sector executives, and policymakers think about regulations, governance frameworks, guiding principles, and national and international governance strategies for AI.

Countries, companies, and universities are increasingly competing over top-tier artificial intelligence (AI) researchers. Where are these researchers likely to immigrate and what affects their immigration decisions? We conducted a survey (n = 524) of the immigration preferences and motivations of researchers that had papers accepted at one of two prestigious AI conferences: the Conference on Neural Information Processing Systems (NeurIPS) and the International Conference on Machine Learning (ICML). We find that the U.S. is the most popular destination for AI researchers, followed by the U.K., Canada, Switzerland, and France. A country's professional opportunities stood out as the most common factor that influences immigration decisions of AI researchers, followed by lifestyle and culture, the political climate, and personal relations. The destination country's immigration policies were important to just under half of the researchers surveyed, while around a quarter noted current immigration difficulties to be a deciding factor. Visa and immigration difficulties were perceived to be a particular impediment to conducting AI research in the U.S., the U.K., and Canada. Implications of the findings for the future of AI talent policies and governance are discussed.

Machine learning (ML) and artificial intelligence (AI) researchers play an important role in the ethics and governance of AI, including through their work, advocacy, and choice of employment. Nevertheless, this influential group's attitudes are not well understood, undermining our ability to discern consensuses or disagreements between AI/ML researchers. To examine these researchers' views, we conducted a survey of those who published in two top AI/ML conferences (N = 524). We compare these results with those from a 2016 survey of AI/ML researchers (Grace et al., 2018) and a 2018 survey of the US public (Zhang & Dafoe, 2020). We find that AI/ML researchers place high levels of trust in international organizations and scientific organizations to shape the development and use of AI in the public interest; moderate trust in most Western tech companies; and low trust in national militaries, Chinese tech companies, and Facebook. While the respondents were overwhelmingly opposed to AI/ML researchers working on lethal autonomous weapons, they are less opposed to researchers working on other military applications of AI, particularly logistics algorithms. A strong majority of respondents think that AI safety research should be prioritized and that ML institutions should conduct pre-publication review to assess potential harms. Being closer to the technology itself, AI/ML researchers are well placed to highlight new risks and develop technical solutions, so this novel attempt to measure their attitudes has broad relevance. The findings should help to improve how researchers, private sector executives, and policymakers think about regulations, governance frameworks, guiding principles, and national and international governance strategies for AI. This article appears in the special track on AI & Society.

Artificial intelligence (AI) has widespread societal implications, yet social scientists are only beginning to study public attitudes toward the technology. Existing studies find that the public's trust in institutions can play a major role in shaping the regulation of emerging technologies. Using a large-scale survey (N=2000), we examined Americans' perceptions of 13 AI governance challenges as well as their trust in governmental, corporate, and multistakeholder institutions to responsibly develop and manage AI. While Americans perceive all of the AI governance issues to be important for tech companies and governments to manage, they have only low to moderate trust in these institutions to manage AI applications.

Traffic Engineering for Failure Recovery (TEFR) optimizes the rerouted traffic during network failure. Using Linear Program (LP) to solve this optimization problem is too computationally intensive. Thus, we propose a novel Universal Single-link Traffic Rerouting (USTR) approach based on the maximum flow method. Having the same level of optimality as LP, USTR has a lower time complexity by orders of magnitude than LP, so that all the links in network can be protected. Since it is the first time to use maximum flow method to solve the traffic rerouting problem, we rigorously prove the correctness and the complexity of USTR. We evaluate USTR on real network topologies with thousands of links and nodes, and the running time is quite acceptable. Specifically, we use SDN to implement a prototype of USTR on the OpenDaylight (ODL) controller which works in OpenFlow, MPLS and IP networks.

Advances in artificial intelligence (AI) will transform modern life by reshaping transportation, health, science, finance, and the military. To adapt public policy, we need to better anticipate these advances. Here we report the results from a large survey of machine learning researchers on their beliefs about progress in AI. Researchers predict AI will outperform humans in many activities in the next ten years, such as translating languages (by 2024), writing high-school essays (by 2026), driving a truck (by 2027), working in retail (by 2031), writing a bestselling book (by 2049), and working as a surgeon (by 2053). Researchers believe there is a 50% chance of AI outperforming humans in all tasks in 45 years and of automating all human jobs in 120 years, with Asian respondents expecting these dates much sooner than North Americans. These results will inform discussion amongst researchers and policymakers about anticipating and managing trends in AI. This article is part of the special track on AI and Society.

To speed up the recovery from network failures, an extensive list of methods have been proposed. Many failurerecovery methods are proposed based on tunneling or marking, which increase the packet processing burden on routers and consume extra bandwidth. With neither tunneling nor marking, existing methods guarantee recovery from any singlelink failure if a detour for the failed link exists, but they generate long traffic detours that will degrade the network performance, and even increase the operational cost, which is undesirable to network operators. Therefore, in this paper, we propose a Failure Inference approach to shortening Traffic Detours named as FITD, which works in OSPF/IS-IS networks. FITD does not use explicit failure notification, and can infer which link fails based on traffic information. FITD guarantees recovery from any single-link failure if a detour for the failed link exists. In particular, for networks with symmetric link weights, FITD guarantees to generate shortest detours for any single-link failure.

The decoupled architecture and the fine-grained flow-control feature limit the scalability of a flow-based software-defined network (SDN). In order to address this problem, some studies construct a flat control plane architecture; others build a hierarchical control plane architecture to improve the scalability of an SDN. However, the two kinds of structure still have unresolved issues: A flat control plane structure cannot solve the superlinear computational complexity growth of the control plane when the SDN scales to a large size, and the centralized abstracted hierarchical control plane structure brings a path stretch problem. To address these two issues, we propose Orion, a hybrid hierarchical control plane for large-scale networks. Orion can effectively reduce the computational complexity of an SDN control plane by several orders of magnitude. We also design an abstracted hierarchical routing method to solve the path stretch problem. Furthermore, we propose a hierarchical fast reroute method to illustrate how to achieve fast rerouting in the proposed hybrid hierarchical control plane. Orion is implemented to verify the feasibility of the hybrid hierarchical approach. Finally, we verify the effectiveness of Orion from both the theoretical and experimental aspects.

In today's Internet, users can send packets with forged source IP addresses, called IP Spoofing, which causes many severe problems to the Internet, especially the security problem. Extensive methods have been proposed to prevent IP spoofing. One category of anti-spoofing methods is extra-information-based. The other category of anti-spoofing methods is existing-information-based. The existing-information-based anti-spoofing methods are much easier to deploy than the extra-information-based ones. Thus, we focus on the existing-information-based anti-spoofing methods in this paper. However, the existing existing-information-based anti-spoofing methods perform poorly on preventing IP spoofing. Therefore, in this paper we propose a new existing-information-based anti-spoofing method, named Link-state-based Anti-Spoofing (LAS), which works in a link-state-protocol-based network. LAS provides good effectiveness on preventing IP spoofing as our simulation results show that even if LAS is only deployed on 10% of routers in a link-state-protocol-based network, 80%~99% of spoofing cases in the network can be prevented in general.

No abstract available.

It is of great significance for network operators and researchers to obtain accurate AS-level traceroute paths, for which mapping IP addresses to correct AS numbers is critical. Thus, there have been a lot of efforts to improve the original IP-to-AS mapping table, which was extracted from BGP routing tables. One of these efforts is called pair matching, which refines the original mapping table by maximizing the number of matched pairs of traceroute and BGP AS paths. However, the existing pair-matching-based methods refine the original IP-to-AS mapping table only with the prefix granularity, i.e., IP addresses in the same /24 prefix are mapped to the same AS or the same set of ASes, which does not fit reality. In this paper, we attempt to refine the IP-to-AS mapping table with the IP address granularity, i.e., allowing IP addresses in the same prefix to be mapped to different ASes. The results show that our fine-grained method can produce a more accurate IP-to-AS mapping table. In addition, this paper also provides a better understanding for the pair-matching-based methods.

In today's IP network, if a link or router fails, packets that traverse the failed link or router will be lost until the network re-converges even if there exists another path bypassing the failure. We call such a single failure a valid single failure. Therefore the IETF published a framework called IP Fast ReRoute (IPFRR) which aims to provide protection against such valid single failures before the network re-converges. Based on the IPFRR framework, a lot of methods have been proposed. One category of IPFRR methods is IP-tunnel-based. The IP-tunnel-based IPFRR methods impose extra cost such as the encapsulation cost, on the traffic delivery. In this paper we focus on the otber category of IPFRR methods, no-tunnel-based IPFRR methods, which do not impose any extra cost on the traffic delivery. However, the existing no-tunnel-based IPFRR methods cannot provide complete protection against all the valid single failures. Therefore in this paper we propose a new kind of no-tunnel-based IPERR method named with RPFP which can provide complete protection against all the valid single failures.

No abstract available.

The traditional IP routing protocols transfer traffic only according to destination, which is not fine-grained enough to satisfy the significant demands of quality-of-service (QoS) and traffic engineering, so fine-grained flow control techniques, such as Multiprotocol Label Switching (MPLS) [1], emerge. MPLS is the most widely used technique to do fine-grained flow control.

Internet core routers are facing challenges brought by the ever-increasing transit bandwidth and routing scale. In order to meet the requirements of highly efficient packet forwarding, some solutions propose to load a small portion of the BGP RIB entries into the FIB. Therefore the most popular prefixes, which contribute major traffic loads, need to be cached in the FIB as long as possible. In this paper, we try to propose a prediction based method to cache those popular prefixes in the FIB. The cache strategy is guided by the traffic prediction of a grey model. We also apply FIB aggregation techniques to suppress the number of overlapped sub-prefixes of the popular prefixes on cache/route updates. We evaluate our method with real traffic traces and find that our prediction-based cache replacement strategy outperforms other cache strategies and matches Internet traffic dynamics very well.

Although traceroute has the potential to discover AS links that are invisible to existing BGP monitors, it is well known that the common approach for mapping router IP addresses to AS numbers based on BGP routing tables is highly error-prone. We develop a systematic framework to quantify the potential errors of traceroute measurement in AS-level topology inference. In comparing traceroute-derived AS paths with BGP AS paths, we take a novel approach to identifying mismatched path segments and then inferring the causes of these mismatches through a set of tests. Our results show that about 60% of mismatches are due to routers using IP addresses belonging to peering neighbors. This result helps settle a debate in previous works regarding the major cause of errors in traceroute measurement. With the approximate ground truth of the ASes with BGP monitors inside, we identify the inaccuracy of publicly available traceroute-derived topology datasets and find that between 8% and 42% of AS adjacencies on the monitored ASes are false. With a new method to characterize AS links, we show that the derived (false) links between Tier-1/large ISPs and their customers' customers appear more frequently than real links do.