Mixed-initiative nested classification by optimal thresholding

Hyun, Baro; Faied, Mariam; Kabamba, Pierre T.; Girard, Anouck

doi:10.1109/cdc.2011.6160633

Cited by 6 publications

(5 citation statements)

References 21 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…This implies that if the workload-independent classifiers (machines) have better performances than the workloaddependent classifiers (human operators), there is no optimal solution for n. Similarly, it can be shown that for σ 1 < σ 2 , 2 Although we do not present a full proof for this claim due to page restriction, it can be easily shown with a similar argument presented in [2].…”

Section: A Analytical Propertiesmentioning

confidence: 76%

“…In our previous work [2], [3], we considered a mixedinitiative nested classification where two heterogeneous classifiers are composed in a nested architecture. Figure 3 shows the concept.…”

Section: B Mixed-initiative Nested Classificationmentioning

confidence: 99%

“…Due to page limitation, we guide the readers to [2] for the derivation. The global objective of the nested team architecture is to minimize the probability of misclassification by choosing the threshold variables for the primary trichotomous classifier, i.e., min…”

Section: B Mixed-initiative Nested Classificationmentioning

confidence: 99%

“…Previously, we considered mixed-initiative nested classification with two classifiers: a primary workload-independent trichotomous classifier and a secondary workload-dependent dichotomous classifier [2], [3]. We adopt workload independence and dependence as "first-order" models to capture the features of machines and humans, respectively.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Mixed-initiative nested classification for n team members

Hyun

Faied

Kabamba

et al. 2012

2012 IEEE 51st IEEE Conference on Decision and Control (CDC)

Self Cite

View full text Add to dashboard Cite

We consider the problem of finding the optimal human-to-machine ratio for classification tasks, where humans and machines are abstracted as workload dependent and independent classifiers, respectively. The contribution is twofold: 1. We generalize the mixed-initiative nested thresholding, i.e., a classification architecture that uses a primary workloadindependent classifier and a secondary workload-dependent classifier, for a general n number of classifiers in the architecture, 2. We identify the optimal ratio of the mixedinitiative team members, the corresponding minimal probability of misclassification, and the individual workload applied to the workload-dependent classifier as a function of the total workload applied to the architecture.

show abstract

Section: A Analytical Propertiesmentioning

confidence: 76%

“…In our previous work [2], [3], we considered a mixedinitiative nested classification where two heterogeneous classifiers are composed in a nested architecture. Figure 3 shows the concept.…”

Section: B Mixed-initiative Nested Classificationmentioning

confidence: 99%

Section: B Mixed-initiative Nested Classificationmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Mixed-initiative nested classification for n team members

Hyun

Faied

Kabamba

et al. 2012

2012 IEEE 51st IEEE Conference on Decision and Control (CDC)

Self Cite

View full text Add to dashboard Cite

show abstract

“…The optimal threshold is computed by minimizing the probability of misclassification (theory and examples of thresholding for classification can be found in [31]), The second order condition for a minimum is indeed satisfied:…”

Section: B Probability Of Misclassificationmentioning

confidence: 99%

Optimal Configuration of Alarm Sensors for Monitoring Mobile Ergodic Markov Phenomena on Arbitrary Graphs

2015

Self Cite

View full text Add to dashboard Cite

Motivated by persistent monitoring tasks, this paper considers the placement of alarm sensors incapable of long-distance communication on arbitrary graphs and the selection of the rates of their revisits (by an external agent) to monitor a mobile phenomenon whose movements occur on a graph and are modeled as an ergodic Markov chain. The alarm sensors can be placed on nodes and edges in the graph and act as both sensors and classifiers (i.e., they make a classification decision about the presence of the phenomenon based on a measurement of their surroundings). An approach to design the classifier for each alarm sensor is provided and methods to fuse the measurements of colocated alarm sensors are given. Sensor placement problems to optimize Fisher information, probability of misclassification, or the penalty incurred by poor detections, missed detections, and false alarms are formulated. Approaches to solve the formulated problems for the different optimization criteria are provided. Characteristics of these approaches are described and their merits are discussed. The sensors' revisit rates are selected by matching the recurrence times of the phenomenon at the sensor locations. The different approaches are illustrated through simulations.Index Terms-Sensor placement, Fisher information, probability of misclassification, revisit deadlines.

show abstract