skip to main content


Title: Improving Crowdsourcing-Based Image Classification Through Expanded Input Elicitation and Machine Learning
This work investigates how different forms of input elicitation obtained from crowdsourcing can be utilized to improve the quality of inferred labels for image classification tasks, where an image must be labeled as either positive or negative depending on the presence/absence of a specified object. Five types of input elicitation methods are tested: binary classification (positive or negative); the ( x, y )-coordinate of the position participants believe a target object is located; level of confidence in binary response (on a scale from 0 to 100%); what participants believe the majority of the other participants' binary classification is; and participant's perceived difficulty level of the task (on a discrete scale). We design two crowdsourcing studies to test the performance of a variety of input elicitation methods and utilize data from over 300 participants. Various existing voting and machine learning (ML) methods are applied to make the best use of these inputs. In an effort to assess their performance on classification tasks of varying difficulty, a systematic synthetic image generation process is developed. Each generated image combines items from the MPEG-7 Core Experiment CE-Shape-1 Test Set into a single image using multiple parameters (e.g., density, transparency, etc.) and may or may not contain a target object. The difficulty of these images is validated by the performance of an automated image classification method. Experiment results suggest that more accurate results can be achieved with smaller training datasets when both the crowdsourced binary classification labels and the average of the self-reported confidence values in these labels are used as features for the ML classifiers. Moreover, when a relatively larger properly annotated dataset is available, in some cases augmenting these ML algorithms with the results (i.e., probability of outcome) from an automated classifier can achieve even higher performance than what can be obtained by using any one of the individual classifiers. Lastly, supplementary analysis of the collected data demonstrates that other performance metrics of interest, namely reduced false-negative rates, can be prioritized through special modifications of the proposed aggregation methods.  more » « less
Award ID(s):
1850355
NSF-PAR ID:
10336705
Author(s) / Creator(s):
; ; ; ; ;
Date Published:
Journal Name:
Frontiers in Artificial Intelligence
Volume:
5
ISSN:
2624-8212
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Kamar, Ece ; Luther, Kurt (Ed.)
    This study investigates how different forms of input elicitation obtained from crowdsourcing can be utilized to improve the quality of inferred labels for image classification tasks, where an image must be labeled as either positive or negative depending on the presence/absence of a specified object. Three types of input elicitation methods are tested: binary classification (positive or negative); level of confidence in binary response (on a scale from 0-100%); and what participants believe the majority of the other participants’ binary classification is. We design a crowdsourcing experiment to test the performance of the proposed input elicitation methods and use data from over 200 participants. Various existing voting and machine learning (ML) methods are applied and others developed to make the best use of these inputs. In an effort to assess their performance on classification tasks of varying difficulty, a systematic synthetic image generation process is developed. Each generated image combines items from the MPEG-7 Core Experiment CE-Shape-1 Test Set into a single image using multiple parameters (e.g., density, transparency, etc.) and may or may not contain a target object. The difficulty of these images is validated by the performance of an automated image classification method. Experimental results suggest that more accurate classifications can be achieved when using the average of the self-reported confidence values as an additional attribute for ML algorithms relative to what is achieved with more traditional approaches. Additionally, they demonstrate that other performance metrics of interest, namely reduced false-negative rates, can be prioritized through special modifications of the proposed aggregation methods that leverage the variety of elicited inputs. 
    more » « less
  2. In this work we explore confidence elicitation methods for crowdsourcing "soft" labels, e.g., probability estimates, to reduce the annotation costs for domains with ambiguous data. Machine learning research has shown that such "soft" labels are more informative and can reduce the data requirements when training supervised machine learning models. By reducing the number of required labels, we can reduce the costs of slow annotation processes such as audio annotation. In our experiments we evaluated three confidence elicitation methods: 1) "No Confidence" elicitation, 2) "Simple Confidence" elicitation, and 3) "Betting" mechanism for confidence elicitation, at both individual (i.e., per participant) and aggregate (i.e., crowd) levels. In addition, we evaluated the interaction between confidence elicitation methods, annotation types (binary, probability, and z-score derived probability), and "soft" versus "hard" (i.e., binarized) aggregate labels. Our results show that both confidence elicitation mechanisms result in higher annotation quality than the "No Confidence" mechanism for binary annotations at both participant and recording levels. In addition, when aggregating labels at the recording level, results indicate that we can achieve comparable results to those with 10-participant aggregate annotations using fewer annotators if we aggregate "soft" labels instead of "hard" labels. These results suggest that for binary audio annotation using a confidence elicitation mechanism and aggregating continuous labels we can obtain higher annotation quality, more informative labels, with quality differences more pronounced with fewer participants. Finally, we propose a way of integrating these confidence elicitation methods into a two-stage, multi-label annotation pipeline. 
    more » « less
  3. The overall purpose of this paper is to demonstrate how data preprocessing, training size variation, and subsampling can dynamically change the performance metrics of imbalanced text classification. The methodology encompasses using two different supervised learning classification approaches of feature engineering and data preprocessing with the use of five machine learning classifiers, five imbalanced sampling techniques, specified intervals of training and subsampling sizes, statistical analysis using R and tidyverse on a dataset of 1000 portable document format files divided into five labels from the World Health Organization Coronavirus Research Downloadable Articles of COVID-19 papers and PubMed Central databases of non-COVID-19 papers for binary classification that affects the performance metrics of precision, recall, receiver operating characteristic area under the curve, and accuracy. One approach that involves labeling rows of sentences based on regular expressions significantly improved the performance of imbalanced sampling techniques verified by performing statistical analysis using a t-test documenting performance metrics of iterations versus another approach that automatically labels the sentences based on how the documents are organized into positive and negative classes. The study demonstrates the effectiveness of ML classifiers and sampling techniques in text classification datasets, with different performance levels and class imbalance issues observed in manual and automatic methods of data processing. 
    more » « less
  4. Koyejo, S. ; Mohamed, S. ; Agarwal, A. ; Belgrave, D. ; Cho, K. ; Oh, A. (Ed.)
    Machine Learning (ML) research has focused on maximizing the accuracy of predictive tasks. ML models, however, are increasingly more complex, resource intensive, and costlier to deploy in resource-constrained environments. These issues are exacerbated for prediction tasks with sequential classification on progressively transitioned stages with “happens-before” relation between them.We argue that it is possible to “unfold” a monolithic single multi-class classifier, typically trained for all stages using all data, into a series of single-stage classifiers. Each single- stage classifier can be cascaded gradually from cheaper to more expensive binary classifiers that are trained using only the necessary data modalities or features required for that stage. UnfoldML is a cost-aware and uncertainty-based dynamic 2D prediction pipeline for multi-stage classification that enables (1) navigation of the accuracy/cost tradeoff space, (2) reducing the spatio-temporal cost of inference by orders of magnitude, and (3) early prediction on proceeding stages. UnfoldML achieves orders of magnitude better cost in clinical settings, while detecting multi- stage disease development in real time. It achieves within 0.1% accuracy from the highest-performing multi-class baseline, while saving close to 20X on spatio- temporal cost of inference and earlier (3.5hrs) disease onset prediction. We also show that UnfoldML generalizes to image classification, where it can predict different level of labels (from coarse to fine) given different level of abstractions of a image, saving close to 5X cost with as little as 0.4% accuracy reduction. 
    more » « less
  5. Deep learning on graph structures has shown exciting results in various applications. However, few attentions have been paid to the robustness of such models, in contrast to numerous research work for image or text adversarial attack and defense. In this paper, we focus on the adversarial attacks that fool deep learning models by modifying the combinatorial structure of data. We first propose a reinforcement learning based attack method that learns the generalizable attack policy, while only requiring prediction labels from the target classifier. We further propose attack methods based on genetic algorithms and gradient descent in the scenario where additional prediction confidence or gradients are available. We use both synthetic and real-world data to show that, a family of Graph Neural Network models are vulnerable to these attacks, in both graph-level and node-level classification tasks. We also show such attacks can be used to diagnose the learned classifiers. 
    more » « less