An official website of the United States government
Ensemble learning, in its simplest form, entails the training of multiple models with the same training set. In a standard supervised setting, the training set can be viewed as a 'teacher' with an unbounded capacity of interactions with a single group of 'trainee' models. One can then ask the following broad question: How can we train an ensemble if the teacher has a bounded capacity of interactions with the trainees? Towards answering this question we consider how humans learn in peer groups. The problem of how to group individuals in order to maximize outcomes via cooperative learning has been debated for a long time by social scientists and policymakers. More recently, it has attracted research attention from an algorithmic standpoint which led to the design of grouping policies that appear to result in better aggregate learning in experiments with human subjects. Inspired by human peer learning, we hypothesize that using partially trained models as teachers to other less accurate models, i.e.~viewing ensemble learning as a peer process, can provide a solution to our central question. We further hypothesize that grouping policies, that match trainer models with learner models play a significant role in the overall learning outcome of the ensemble. We present a formalization and through extensive experiments with different types of classifiers, we demonstrate that: (i) an ensemble can reach surprising levels of performance with little interaction with the training set (ii) grouping policies definitely have an impact on the ensemble performance, in agreement with previous intuition and observations in human peer learning.
more »
« less
Peer Learning Through Targeted Dynamic Groups Formation
Peer groups leverage the presence of knowledgeable individuals in order to increase the knowledge level of other participants. The `smart' formation of peer groups can thus play a crucial role in educational settings, including online social networks and learning platforms. Indeed, the targeted groups formation problem, where the objective is to maximize a measure of aggregate knowledge, has received considerable attention in recent literature. In this paper we initiate a dynamic variant of the problem that, unlike previous works, allows the change of group composition over time while still targeting to maximize the aggregated knowledge level. The problem is studied in a principled way, using a realistic learning gain function and for two different interaction modes among the group members. On the algorithmic side, we present DyGroups, a generic algorithmic framework that is greedy in nature and highly scalable. We present non-trivial proofs to demonstrate theoretical guarantees for DyGroups in a special case. We also present real peer learning experiments with humans, and perform synthetic data experiments to demonstrate the effectiveness of our proposed solutions by comparing against multiple appropriately selected baseline algorithms.
more »
« less
- Award ID(s):
- 2039863
- PAR ID:
- 10310837
- Date Published:
- Journal Name:
- 2021 IEEE 37th International Conference on Data Engineering (ICDE)
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Background. Middle school students’ math anxiety and low engagement have been major issues in math education. In order to reduce their anxiety and support their math learning, game-based learning (GBL) has been implemented. GBL research has underscored the role of social dynamics to facilitate a qualitative understanding of students’ knowledge. Whereas students’ peer interactions have been deemed a social dynamic, the relationships among peer interaction, task efficiency, and learning engagement were not well understood in previous empirical studies. Method. This mixed-method research implemented E-Rebuild, which is a 3D architecture game designed to promote students’ math problem-solving skills. We collected a total of 102 50-minutes gameplay sessions performed by 32 middle school students. Using video-captured and screen-recorded gameplaying sessions, we implemented behavior observations to measure students’ peer interaction efficiency, task efficiency, and learning engagement. We used association analyses, sequential analysis, and thematic analysis to explain how peer interaction promoted students’ task efficiency and learning engagement. Results. Students’ peer interactions were negatively related to task efficiency and learning engagement. There were also different gameplay patterns by students’ learning/task-relevant peer-interaction efficiency (PIE) level. Students in the low PIE group tended to progress through game tasks more efficiently than those in the high PIE group. The results of qualitative thematic analysis suggested that the students in the low PIE group showed more reflections on game-based mathematical problem solving, whereas those with high PIE experienced distractions during gameplay. Discussion. This study confirmed that students’ peer interactions without purposeful and knowledge-constructive collaborations led to their low task efficiency, as well as low learning engagement. The study finding shows further design implications: (1) providing in-game prompts to stimulate students’ math-related discussions and (2) developing collaboration contexts that legitimize students’ interpersonal knowledge exchanges with peers.more » « less
-
null (Ed.)Algorithmic fairness is a major concern in recent years as the influence of machine learning algorithms becomes more widespread. In this paper, we investigate the issue of algorithmic fairness from a network-centric perspective. Specifically, we introduce a novel yet intuitive function known as fairness perception and provide an axiomatic approach to analyze its properties. Using a peer-review network as a case study, we also examine its utility in terms of assessing the perception of fairness in paper acceptance decisions. We show how the function can be extended to a group fairness metric known as fairness visibility and demonstrate its relationship to demographic parity. We also discuss a potential pitfall of the fairness visibility measure that can be exploited to mislead individuals into perceiving that the algorithmic decisions are fair. We demonstrate how the problem can be alleviated by increasing the local neighborhood size of the fairness perception function.more » « less
-
Transfer students, who are disproportionately URM and first-generation, are a target population for boosting engineering representation. Transfer students in mechanical, aerospace and civil engineering at [the institution] take thermodynamics, a required gateway course, in their first or second term. This paper outlines the results from an observational study to determine how students interact in a peer-led learning environment. The PEERSIST (Peer-led, Student Instructed, Study group) model promotes academic competence through peer dialogue, in which disciplinary knowledge is socially co-constructed and refined over successive sessions. In order to help demonstrate that student interactions are the main source of learning in Peer-Led Study Groups (PLSGs), interactions between students were recorded and compared to those in traditional TA-led recitations using the observation protocol. Results show that students in PLSGs interact with their peers significantly more than students in the TA-led control group. The study also compares peer interactions by incoming course preparedness and finds a non-significant relationship between incoming GPA and peer-to-peer interactions. In contrast, the study finds a negative relationship between the rate at which students ask for and receive help and incoming GPA.more » « less
-
Disagreement is often perceived negatively, yet it can be beneficial for learning and scientific inquiry. However, students tend to avoid engaging in disagreement. Peer critique activities offer a promising way to encourage students to embrace disagreement, which supports learning as students articulate their ideas, making them available for discussion, revision, and refinement. This study aims to better understand how students express disagreement during peer critique within small groups and how that affects moving their inquiry forward. It explores 5th-grade students’ management of disagreement within a computer-supported collaborative modeling environment. Using conversation analysis, we identified various forms of disagreements employed by students when engaging with different audiences. We observed a tendency for students to disagree softly; that is, disagreement was implied and/or mitigated. Students’ resolution of both direct and soft disagreements effectively promoted their collective knowledge advancement, including building shared scientific understanding and improving their models, while maintaining a positive socio-emotional climate. These findings have implications for designing CSCL environments with respect to supporting students in providing and responding to peer critiques at the group level.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- https://doi.org/10.1109/ICDE51399.2021.00018
