An official website of the United States government
Federated learning (FL) enables multiple participants to train a global machine learning model without sharing their private training data. Peer-to-peer (P2P) FL advances existing centralized FL paradigms by eliminating the server that aggregates local models from participants and then updates the global model. However, P2P FL is vulnerable to (i) honest-but-curious participants whose objective is to infer private training data of other participants, and (ii) Byzantine participants who can transmit arbitrarily manipulated local models to corrupt the learning process. P2P FL schemes that simultaneously guarantee Byzantine resilience and preserve privacy have been less studied. In this paper, we develop Brave, a protocol that ensures Byzantine Resilience And priVacy-prEserving property for P2P FL in the presence of both types of adversaries. We show that Brave preserves privacy by establishing that any honest-but-curious adversary cannot infer other participants’ private data by observing their models. We further prove that Brave is Byzantine-resilient, which guarantees that all benign participants converge to an identical model that deviates from a global model trained without Byzantine adversaries by a bounded distance. We evaluate Brave against three state-of-the-art adversaries on a P2P FL for image classification tasks on benchmark datasets CIFAR10 and MNIST. Our results show that global models learned with Brave in the presence of adversaries achieve comparable classification accuracy to global models trained in the absence of any adversary.
more »
« less
AccountNet: Accountable Data Propagation Using Verifiable Peer Shuffling
Collecting evidence of data that software systems produce and consume can provide critical information for reconstructing erratic behavior, tracing the origin of faults, and thus holding the responsible party accountable for particular consequences. However, when data propagates across trust boundaries with conflicting interests, they can be tempted to make evidence unprovable in order to avoid potential liability. Hence, we present a data propagation protocol that makes such attempts either detectable or ineffective by having data transfers witnessed by other network participants that collectively act as the prover for the data propagation process. The protocol builds an unstructured peer-to-peer overlay and is designed to disincentivize collusion among a malicious coalition of nodes by enforcing network participants to constantly exchange their partial views on the network in a random yet verifiable manner. A data producer or consumer who does not faithfully follow the protocol ends up having fewer witnesses from their side, making the network resistant to collusion with high probability. We derive the conditions under which this property holds and demonstrate the practicality and cost of our approach through the implementation of a distributed application built on top of the proposed protocol.
more »
« less
- Award ID(s):
- 2302610
- PAR ID:
- 10510868
- Publisher / Repository:
- IEEE
- Date Published:
- ISBN:
- 979-8-3503-3986-4
- Page Range / eLocation ID:
- 48 to 61
- Format(s):
- Medium: X
- Location:
- Hong Kong, Hong Kong
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Collusion rings pose a significant threat to peer review. In these rings, reviewers who are also authors coordinate to manipulate paper assignments, often by strategically bidding on each other’s papers. A promising solution is to detect collusion through these manipulated bids, enabling conferences to take appropriate action. However, while methods exist for detecting other types of fraud, no research has yet shown that identifying collusion rings is feasible. In this work, we consider the question of whether it is feasible to detect collusion rings from the paper bidding. We conduct an empirical analysis of two realistic conference bidding datasets and evaluate existing algorithms for fraud detection in other applications. We find that collusion rings can achieve considerable success at manipulating the paper assignment while remaining hidden from detection: for example, in one dataset, undetected colluders are able to achieve assignment to up to 30% of the papers authored by other colluders. In addition, when 10 colluders bid on all of each other’s papers, no detection algorithm outputs a group of reviewers with more than 31% overlap with the true colluders. These results suggest that collusion cannot be effectively detected from the bidding using popular existing tools, demonstrating the need to develop more complex detection algorithms as well as those that leverage additional metadata (e.g., reviewer-paper text-similarity scores).more » « less
-
In this evidence-based practice paper, we discuss design rationale, implementation and evidence from a professional development program for emerging education researchers (PEER). Many STEM faculty, trained only in disciplinary research, transition into research on the teaching and learning side of their discipline, with transitions occurring after typical formal training opportunities (e.g. grad school, postdocs) are over. There are limited opportunities for professional development when starting education research, and options are highly dependent on home institution type, department priorities, and faculty career stage. The PEER program helps faculty at any institution jumpstart their transition into discipline-based education research. Our goal is to help foster the next generation of STEM education researchers. PEER participants develop quality research projects, engage in targeted experiential work to develop their projects and skills, and collaborate and form a long-term support community of peers, mentors and collaborators. Over the last 8 years, more than 1000 participants have engaged in PEER field schools worldwide. In this paper, we lay out the guiding principles of PEER: collaboration, responsiveness, communication, and playfulness. We situate the program within existing models for faculty professional development and describe the available modalities of PEER field schools: extended introductory in-person field schools (3-5 days), online or in-person gateway workshops (1.5 hours), and the new advanced in-person field schools (5 days). Each of these modalities is built off collaborative work among participants, blending development of foundational skills in education research with individual progress in their own specific education research projects. Drawing on evidence from interviews and surveys with STEM participants, we demonstrate the impact of different module activities on their professional skills, identity, and self-efficacy. We discuss the affordances and constraints of different formats and implications for faculty professional development. We prefer to present this work through a roundtable discussion, but we are also open to a lightning talk or a poster.more » « less
-
Research on smart home monitoring for older adults has predominantly focused on systems whose data and alerts are directed towards family members, caregivers, or healthcare providers. Older adults have expressed interest in engaging with these systems by seeing and using their data, but they are often limited to a passive role as subjects of monitoring. This paper presents qualitative results of a longitudinal smart home project with older adults living independently in the community. Based on interviews conducted throughout the 2.5-year study with 12 participants, we report on their lived experiences of having the monitoring system in their homes and on how they reflected on the data collected by the system. The results show how participants were able to extract meaningful information from the monitoring data without finding the system invasive or intrusive. Specifically, older adults exhibited interest in data that they found indicative of living an active lifestyle, such as time spent outside the home. Drawing from critical literature on active aging, we discuss implications for incorporating peer comparisons to support reflection on personal health data without reinforcing a deficit narrative of aging.more » « less
-
Abstract BackgroundBeing socially excluded has detrimental effects, with prolonged exclusion linked to loneliness and social isolation. Social disconnection interventions that do not require direct support actions (e.g., “how can I help?”) offer promise in mitigating the affective and cognitive consequences of social exclusion. We examine how various social disconnection interventions involving friends and unknown peers might mitigate social exclusion by buffering (interveningbefore) and by promoting recovery (interveningafter). MethodsWe present an integrative data analysis (IDA) of five studies (N = 664) that systematically exposed participants to exclusion (vs. inclusion) social dynamics. Using a well-validated paradigm, participants had a virtual interaction with two other people. Unbeknownst to participants, the other people’s behavior was programmed to either behave inclusively toward the participant or for one to behave exclusively. Critically, our social disconnection interventions experimentally manipulated whether a friend was present (vs. an unknown peer vs. being alone), the nature of interpersonal engagement (having a face-to-face conversation vs. a reminder of an upcoming interaction vs. mere presence), and the timing of the intervention in relation to the social dynamic (before vs. during vs. after). We then assessed participants’ in-the-moment affective and cognitive responses, which included mood, feelings of belonging, sense of control, and social comfort. ResultsExperiencing exclusion (vs. inclusion) led to negative affective and cognitive consequences. However, engaging in a face-to-face conversation with a friendbeforethe exclusion lessened its impact (p < .001). Moreover, a face-to-face conversation with a friendafterexclusion, and even a reminder of an upcoming interaction with a friend, sped-up recovery (ps < .001). There was less conclusive evidence that a face-to-face conversation with an unknown peer, or that the mere presence of a friend or unknown peer, conferred protective benefits. ConclusionsThe findings provide support for the effectiveness of social disconnection interventions that involve actual (i.e., face-to-face) or symbolic (i.e., reminders) interactions with friends. These interventions target momentary vulnerabilities that arise from social exclusion by addressing negative affect and cognitions before or after they emerge. As such, they offer a promising approach to primary prevention prior to the onset of loneliness and social isolation.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1109/ICDCS57875.2023.00050
