Pub Date : 2022-12-01DOI: 10.1109/CogMI56440.2022.00023
Gabriele Maurina, Hajar Homayouni, Sudipto Ghosh, I. Ray, G. Duggan
Accurate energy consumption prediction is critical for proper resource allocation, meeting energy demand, and energy supply security. This work aims at developing a methodology for accurately modeling and predicting electricity consumption during abnormal long-lasting events, such as COVID-19 pandemic, which considerably affect consumption patterns in different types of premises. The proposed methodology involves three steps: (A) selects among multiple models the most accurate one in energy consumption prediction under normal conditions, (B) uses the selected model to analyze the impact of a specific abnormal event on energy consumption for various classes of premises, and (C) investigates which features contribute most to energy consumption prediction for abnormal conditions and which features can be added to improve such predictions.We use COVID-19 as a case study with datasets obtained from Fort Collins Utilities, which contain energy consumption data for residential and different sizes of commercial and industrial premises in the city of Fort Collins, Colorado, USA. We also use temperature records from NOAA and COVID-19 public orders from Larimer County.We validate the methodology by demonstrating that the methodology can help design a model suited for the pandemic situation using representative features, and as a result, accurately predict the energy consumption. Our results show that the MLP model selected by our methodology performs better than the other models even when they all use the COVID-related features. We also demonstrate that the methodology can help measure the impacts of the pandemic on the energy consumption.
{"title":"A Methodology for Energy Usage Prediction in Long-Lasting Abnormal Events","authors":"Gabriele Maurina, Hajar Homayouni, Sudipto Ghosh, I. Ray, G. Duggan","doi":"10.1109/CogMI56440.2022.00023","DOIUrl":"https://doi.org/10.1109/CogMI56440.2022.00023","url":null,"abstract":"Accurate energy consumption prediction is critical for proper resource allocation, meeting energy demand, and energy supply security. This work aims at developing a methodology for accurately modeling and predicting electricity consumption during abnormal long-lasting events, such as COVID-19 pandemic, which considerably affect consumption patterns in different types of premises. The proposed methodology involves three steps: (A) selects among multiple models the most accurate one in energy consumption prediction under normal conditions, (B) uses the selected model to analyze the impact of a specific abnormal event on energy consumption for various classes of premises, and (C) investigates which features contribute most to energy consumption prediction for abnormal conditions and which features can be added to improve such predictions.We use COVID-19 as a case study with datasets obtained from Fort Collins Utilities, which contain energy consumption data for residential and different sizes of commercial and industrial premises in the city of Fort Collins, Colorado, USA. We also use temperature records from NOAA and COVID-19 public orders from Larimer County.We validate the methodology by demonstrating that the methodology can help design a model suited for the pandemic situation using representative features, and as a result, accurately predict the energy consumption. Our results show that the MLP model selected by our methodology performs better than the other models even when they all use the COVID-related features. We also demonstrate that the methodology can help measure the impacts of the pandemic on the energy consumption.","PeriodicalId":211430,"journal":{"name":"2022 IEEE 4th International Conference on Cognitive Machine Intelligence (CogMI)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116308643","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-01DOI: 10.1109/CogMI56440.2022.00011
Stefan Nastic, Philipp Raith, Alireza Furutanpey, Thomas W. Pusztai, S. Dustdar
Serverless computing has been establishing itself as a compelling paradigm for the development and of modern cloud-native applications. Serverless represents the next step in the evolution of cloud programming models, services and platforms, which is especially appealing due to its low management overhead, easy deployment, scale-to-zero and the promise of optimized costs. Recently, due to the advantages it offers, the serverless paradigm has been growing beyond traditional clouds, making its way to the Edge. The natural evolutionary step for serverless computing is to unify the Edge and the Cloud into what we refer to as Edge-Cloud Continuum. In this paper, we outline our vision of the Serverless Computing Fabric (SCF) for the Edge-Cloud continuum. We introduce the reference architecture for the SCF and show how it unlocks the full potential of the Edge-Cloud continuum. We also discuss main opportunities and challenges, which need to be overcome in order to achieve the vision of the Serverless Computing Fabric. Finally, we introduce key design principles together with core enabling runtime mechanisms, which are intended to serve as a research road map towards the Serverless Computing Fabric for Edge-Cloud continuum.
{"title":"A Serverless Computing Fabric for Edge & Cloud","authors":"Stefan Nastic, Philipp Raith, Alireza Furutanpey, Thomas W. Pusztai, S. Dustdar","doi":"10.1109/CogMI56440.2022.00011","DOIUrl":"https://doi.org/10.1109/CogMI56440.2022.00011","url":null,"abstract":"Serverless computing has been establishing itself as a compelling paradigm for the development and of modern cloud-native applications. Serverless represents the next step in the evolution of cloud programming models, services and platforms, which is especially appealing due to its low management overhead, easy deployment, scale-to-zero and the promise of optimized costs. Recently, due to the advantages it offers, the serverless paradigm has been growing beyond traditional clouds, making its way to the Edge. The natural evolutionary step for serverless computing is to unify the Edge and the Cloud into what we refer to as Edge-Cloud Continuum. In this paper, we outline our vision of the Serverless Computing Fabric (SCF) for the Edge-Cloud continuum. We introduce the reference architecture for the SCF and show how it unlocks the full potential of the Edge-Cloud continuum. We also discuss main opportunities and challenges, which need to be overcome in order to achieve the vision of the Serverless Computing Fabric. Finally, we introduce key design principles together with core enabling runtime mechanisms, which are intended to serve as a research road map towards the Serverless Computing Fabric for Edge-Cloud continuum.","PeriodicalId":211430,"journal":{"name":"2022 IEEE 4th International Conference on Cognitive Machine Intelligence (CogMI)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125707939","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-01DOI: 10.1109/CogMI56440.2022.00016
R. Baeza-Yates, Pablo Villoslada
In this essay we compare human and artificial intelligence from two points of view: computational and neuroscience. We discuss the differences and limitations of AI with respect to our intelligence, ending with three challenging areas that are already with us: neural technologies, responsible AI, and hybrid AI systems.
{"title":"Human vs. Artificial Intelligence","authors":"R. Baeza-Yates, Pablo Villoslada","doi":"10.1109/CogMI56440.2022.00016","DOIUrl":"https://doi.org/10.1109/CogMI56440.2022.00016","url":null,"abstract":"In this essay we compare human and artificial intelligence from two points of view: computational and neuroscience. We discuss the differences and limitations of AI with respect to our intelligence, ending with three challenging areas that are already with us: neural technologies, responsible AI, and hybrid AI systems.","PeriodicalId":211430,"journal":{"name":"2022 IEEE 4th International Conference on Cognitive Machine Intelligence (CogMI)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125601052","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-01DOI: 10.1109/CogMI56440.2022.00015
A. Raglin, Sharon Sputz, Andrew Smyth
Army Research Laboratory’s Content Understanding Branch, Artificial Reasoning Team research objective is to enable systems to reason given existing and future information supporting shared understanding and providing enhanced capabilities for choices and decisions. Various reasoning approaches are used to form the “best” hypothesis from multiple modalities of data generating use cases and assessing their impact on decisions given multiple criteria. The NSF Engineering Research Center for Smart Streetscapes (CS3) convergent research is inspired by potential streetscape applications. Thus, real-time understanding of complex streetscapes correspondingly requires progress in fundamental engineering knowledge and enables exciting opportunities for deploying technology: A “smart streetscape” could instantly sense human behavior and safely guide individual within the environment, amplify emergency services, and protect people against threats and dangers. The ARL and CS3 collaboration centers around the overlapping challenge for situational awareness in complex environments and how the joint research efforts can generate potential capabilities. This paper will present concepts from existing research and ideas for new research to address these common questions and challenges.
{"title":"Artificial Reasoning in the Streetscape","authors":"A. Raglin, Sharon Sputz, Andrew Smyth","doi":"10.1109/CogMI56440.2022.00015","DOIUrl":"https://doi.org/10.1109/CogMI56440.2022.00015","url":null,"abstract":"Army Research Laboratory’s Content Understanding Branch, Artificial Reasoning Team research objective is to enable systems to reason given existing and future information supporting shared understanding and providing enhanced capabilities for choices and decisions. Various reasoning approaches are used to form the “best” hypothesis from multiple modalities of data generating use cases and assessing their impact on decisions given multiple criteria. The NSF Engineering Research Center for Smart Streetscapes (CS3) convergent research is inspired by potential streetscape applications. Thus, real-time understanding of complex streetscapes correspondingly requires progress in fundamental engineering knowledge and enables exciting opportunities for deploying technology: A “smart streetscape” could instantly sense human behavior and safely guide individual within the environment, amplify emergency services, and protect people against threats and dangers. The ARL and CS3 collaboration centers around the overlapping challenge for situational awareness in complex environments and how the joint research efforts can generate potential capabilities. This paper will present concepts from existing research and ideas for new research to address these common questions and challenges.","PeriodicalId":211430,"journal":{"name":"2022 IEEE 4th International Conference on Cognitive Machine Intelligence (CogMI)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115046286","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-01DOI: 10.1109/CogMI56440.2022.00020
Javed Mostafa, W. Ke
Robotic communities are increasingly important in executing operations in a wide variety of industries. Before designing and deploying such robots it is important to determine and carefully plan the configuration, knowledge composition, and coordination strategies. Multi-agent simulation modeling offers a malleable and powerful way to conduct such planning and elucidate key parameters and their interactions associated with collaboration dynamics. The paper offers motivations, an adaptive learning scheme, and empirical evidence drawn from a few case studies. Among the key findings one is that complex tasks can be conducted effectively and efficiently over billions of robots without relying on a singular source of global knowledge. Another interesting finding is that through collaboration and emergent learning, robots can create communication channels among dominant players and less dominant intermediaries that are critical connectors across network overlays (representing clusters of specialists).
{"title":"Collaboration, Self-Reflection, and Adaptation in Robot Communities: Using Multi-Agent Distributed Learning for Coordination Planning","authors":"Javed Mostafa, W. Ke","doi":"10.1109/CogMI56440.2022.00020","DOIUrl":"https://doi.org/10.1109/CogMI56440.2022.00020","url":null,"abstract":"Robotic communities are increasingly important in executing operations in a wide variety of industries. Before designing and deploying such robots it is important to determine and carefully plan the configuration, knowledge composition, and coordination strategies. Multi-agent simulation modeling offers a malleable and powerful way to conduct such planning and elucidate key parameters and their interactions associated with collaboration dynamics. The paper offers motivations, an adaptive learning scheme, and empirical evidence drawn from a few case studies. Among the key findings one is that complex tasks can be conducted effectively and efficiently over billions of robots without relying on a singular source of global knowledge. Another interesting finding is that through collaboration and emergent learning, robots can create communication channels among dominant players and less dominant intermediaries that are critical connectors across network overlays (representing clusters of specialists).","PeriodicalId":211430,"journal":{"name":"2022 IEEE 4th International Conference on Cognitive Machine Intelligence (CogMI)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124940828","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-01DOI: 10.1109/CogMI56440.2022.00017
D. Rawat
Artificial Intelligence (AI) enabled systems have shown tremendous impact in our national defense and in our society due to recent advances in artificial neural networks, deep learning, machine learning, and Internet of Things, big data, computing and communications. New AI capabilities can improve efficiency, trust, and efficacy for mission critical applications for tactical autonomy with minimal supervision from human operators in multi-domain battlefield (MDB) environments that are complex, contested and unpredictable. Although AI-enabled tools have been responsive to people and complementary to human capabilities, in order to realize its full potential in tactical applications, there are several challenges to be addressed for making trustworthy, ethical, fair, real-time explainable AI-enabled autonomous systems. Collaborations between platforms/systems as well as joint human-machine learning/teaming could address many of these issues to provide trusted and shared understanding and delivering cost-effective and adaptive systems to assist operations across military domains (space, air, land, maritime, and cyber) at combat speed using a shared set of resources. In this paper, we present some challenges and perspectives for AI enabled tactical autonomy.
{"title":"Artificial Intelligence Meets Tactical Autonomy: Challenges and Perspectives","authors":"D. Rawat","doi":"10.1109/CogMI56440.2022.00017","DOIUrl":"https://doi.org/10.1109/CogMI56440.2022.00017","url":null,"abstract":"Artificial Intelligence (AI) enabled systems have shown tremendous impact in our national defense and in our society due to recent advances in artificial neural networks, deep learning, machine learning, and Internet of Things, big data, computing and communications. New AI capabilities can improve efficiency, trust, and efficacy for mission critical applications for tactical autonomy with minimal supervision from human operators in multi-domain battlefield (MDB) environments that are complex, contested and unpredictable. Although AI-enabled tools have been responsive to people and complementary to human capabilities, in order to realize its full potential in tactical applications, there are several challenges to be addressed for making trustworthy, ethical, fair, real-time explainable AI-enabled autonomous systems. Collaborations between platforms/systems as well as joint human-machine learning/teaming could address many of these issues to provide trusted and shared understanding and delivering cost-effective and adaptive systems to assist operations across military domains (space, air, land, maritime, and cyber) at combat speed using a shared set of resources. In this paper, we present some challenges and perspectives for AI enabled tactical autonomy.","PeriodicalId":211430,"journal":{"name":"2022 IEEE 4th International Conference on Cognitive Machine Intelligence (CogMI)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121976177","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-01DOI: 10.1109/CogMI56440.2022.00022
Junlin Wang, Wenbo Wang, Yingzi Shang, Dong Xu
Computational protein structure prediction is an important problem in bioinformatics and the ability to accurately evaluating the quality of predicted protein models is of significant interest. In this paper, three new single-model quality assessment (QA) methods, MMQA-1 MMQA-2 and MMQA-HE, are proposed based on two-stage machine learning and hierarchical ensemble techniques. MMQA-1 and MMQA-2 train different machine learning models in two separate stages. They divide the entire feature set into two groups and uses completely different feature sets and training data in each stage to train a predictive model. MMQA-HE is an ensemble method that combines individual models not only at the tree level, but also at the forest level. In CASP14, MMQA-1 ranked No. 2 in terms of average GDT-TS difference. MMQA-2 and MMQA-HE improve MMQA-1 and outperform existing state-of-the-art QA methods across multiple QA performance metrics.
{"title":"New Heuristic Methods for Protein Model Quality Assessment via Two-Stage Machine Learning and Hierarchical Ensemble","authors":"Junlin Wang, Wenbo Wang, Yingzi Shang, Dong Xu","doi":"10.1109/CogMI56440.2022.00022","DOIUrl":"https://doi.org/10.1109/CogMI56440.2022.00022","url":null,"abstract":"Computational protein structure prediction is an important problem in bioinformatics and the ability to accurately evaluating the quality of predicted protein models is of significant interest. In this paper, three new single-model quality assessment (QA) methods, MMQA-1 MMQA-2 and MMQA-HE, are proposed based on two-stage machine learning and hierarchical ensemble techniques. MMQA-1 and MMQA-2 train different machine learning models in two separate stages. They divide the entire feature set into two groups and uses completely different feature sets and training data in each stage to train a predictive model. MMQA-HE is an ensemble method that combines individual models not only at the tree level, but also at the forest level. In CASP14, MMQA-1 ranked No. 2 in terms of average GDT-TS difference. MMQA-2 and MMQA-HE improve MMQA-1 and outperform existing state-of-the-art QA methods across multiple QA performance metrics.","PeriodicalId":211430,"journal":{"name":"2022 IEEE 4th International Conference on Cognitive Machine Intelligence (CogMI)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128480613","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-12-01DOI: 10.1109/CogMI56440.2022.00014
D. Verma
In order to create AI/ML based solutions that will be trusted during production, issues that hamper usage of AI models in practical solutions needs to be addressed. Despite a significant interest in the area of AI/ML, the primary focus of the research community has been on the training of AI models, including their performance, trustworthiness, explainability and scalability. Training, however, is only one half of the work required to create an AI-based solution. The other half, using the trained model for inference during operations, is mistakenly considered a relatively mundane task. As a result, challenges arising in model inference time has received comparatively scant attention. Inference is when AI model is put into practice, resulting in many challenges that are worth the attention of the research community. Despite the existence of several pre-trained models on many Internet sites, anyone trying to build an AI/ML based solution would be hard-pressed to find a model that is useful, trustworthy and reliable, or suitable for the task. Even when a custom model is trained, the solution often falters because the use of model fails to account for the differences in the training and inference environment. In this paper, we identify those challenges and discuss how we can design a generic inference server for trustworthy AI/ML based solutions.
{"title":"Inference for Trustworthy Machine Intelligence: Challenges and Solutions","authors":"D. Verma","doi":"10.1109/CogMI56440.2022.00014","DOIUrl":"https://doi.org/10.1109/CogMI56440.2022.00014","url":null,"abstract":"In order to create AI/ML based solutions that will be trusted during production, issues that hamper usage of AI models in practical solutions needs to be addressed. Despite a significant interest in the area of AI/ML, the primary focus of the research community has been on the training of AI models, including their performance, trustworthiness, explainability and scalability. Training, however, is only one half of the work required to create an AI-based solution. The other half, using the trained model for inference during operations, is mistakenly considered a relatively mundane task. As a result, challenges arising in model inference time has received comparatively scant attention. Inference is when AI model is put into practice, resulting in many challenges that are worth the attention of the research community. Despite the existence of several pre-trained models on many Internet sites, anyone trying to build an AI/ML based solution would be hard-pressed to find a model that is useful, trustworthy and reliable, or suitable for the task. Even when a custom model is trained, the solution often falters because the use of model fails to account for the differences in the training and inference environment. In this paper, we identify those challenges and discuss how we can design a generic inference server for trustworthy AI/ML based solutions.","PeriodicalId":211430,"journal":{"name":"2022 IEEE 4th International Conference on Cognitive Machine Intelligence (CogMI)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125814139","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-09-19DOI: 10.1109/CogMI56440.2022.00024
K. Lai, V. Shmerko, S. Yanushkevich
In this paper, we compare the performance and fairness metrics on visual and thermal images of faces, including the synthetic images of human subjects with face masks. The comparative experiment is performed on two datasets: the SpeakingFace and Thermal-Mask dataset. We assess fairness on real images and show how the same process can be applied to synthetic images. The chosen fairness metrics include demographic parity difference and equalized odds difference. While the demographic parity difference is assessed as 1.24 for random guessing in the process of face identification, it reaches 5.0 when both the precision and recall rate approach 99.99%. These results confirm that inherently biased datasets significantly impact the fairness of any biometric system. For biometric-enabled systems, fairness is related to the adequacy of the data to represent different groups of human subjects. In this paper, we focus on three demographic groups: age, gender, and ethnicity. A primary cause of biases with respect to these groups is the class imbalance introduced through the data collection process. To address the imbalanced datasets, the classes with fewer samples can be augmented with synthetic images to generate a more balanced dataset, resulting in less bias when training a machine learning system. The study shows that fairness is correlated to the performance of the system rather than to the genesis of the images (real or synthetic). The experiment on a simple 3-Block CNN with a precision and recall rate of 99.99% using the demographic parity difference as an estimate of fairness showed that among gender, ethnicity, and age, the latter is an attribute that is the most sensitive while age is the least one.
{"title":"Face Biometric Fairness Evaluation on Real vs Synthetic Cross-Spectral Images","authors":"K. Lai, V. Shmerko, S. Yanushkevich","doi":"10.1109/CogMI56440.2022.00024","DOIUrl":"https://doi.org/10.1109/CogMI56440.2022.00024","url":null,"abstract":"In this paper, we compare the performance and fairness metrics on visual and thermal images of faces, including the synthetic images of human subjects with face masks. The comparative experiment is performed on two datasets: the SpeakingFace and Thermal-Mask dataset. We assess fairness on real images and show how the same process can be applied to synthetic images. The chosen fairness metrics include demographic parity difference and equalized odds difference. While the demographic parity difference is assessed as 1.24 for random guessing in the process of face identification, it reaches 5.0 when both the precision and recall rate approach 99.99%. These results confirm that inherently biased datasets significantly impact the fairness of any biometric system. For biometric-enabled systems, fairness is related to the adequacy of the data to represent different groups of human subjects. In this paper, we focus on three demographic groups: age, gender, and ethnicity. A primary cause of biases with respect to these groups is the class imbalance introduced through the data collection process. To address the imbalanced datasets, the classes with fewer samples can be augmented with synthetic images to generate a more balanced dataset, resulting in less bias when training a machine learning system. The study shows that fairness is correlated to the performance of the system rather than to the genesis of the images (real or synthetic). The experiment on a simple 3-Block CNN with a precision and recall rate of 99.99% using the demographic parity difference as an estimate of fairness showed that among gender, ethnicity, and age, the latter is an attribute that is the most sensitive while age is the least one.","PeriodicalId":211430,"journal":{"name":"2022 IEEE 4th International Conference on Cognitive Machine Intelligence (CogMI)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-09-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122668013","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2022-05-20DOI: 10.1109/CogMI56440.2022.00021
Abhijit Suprem, Sanjyot Vaidya, Suma Cherkadi, Purva Singh, J. E. Ferreira, C. Pu
Machine learning models with explainable predictions are increasingly sought after, especially for real-world, mission-critical applications that require bias detection and risk mitigation. Inherent interpretability, where a model is designed from the ground-up for interpretability, provides intuitive insights and transparent explanations on model prediction and performance. In this paper, we present CoLabel, an approach to build interpretable models with explanations rooted in the ground truth. We demonstrate CoLabel in a vehicle feature extraction application in the context of vehicle make-model recognition (VMMR). By construction, CoLabel performs VMMR with a composite of interpretable features such as vehicle color, type, and make, all based on interpretable annotations of the ground truth labels. First, CoLabel performs corroborative integration to join multiple datasets that each have a subset of desired annotations of color, type, and make. Then, CoLabel uses decomposable branches to extract complementary features corresponding to desired annotations. Finally, CoLabel fuses them together for final predictions. During feature fusion, CoLabel harmonizes complementary branches so that VMMR features are compatible with each other and can be projected to the same semantic space for classification. With inherent interpretability, CoLabel achieves superior performance to the state-of-the-art black-box models, with accuracy of 0.98, 0.95, and 0.94 on CompCars, Cars196, and BoxCars116K, respectively. CoLabel provides intuitive explanations due to constructive interpretability, and subsequently achieves high accuracy and usability in mission-critical situations.
{"title":"Constructive Interpretability with CoLabel: Corroborative Integration, Complementary Features, and Collaborative Learning","authors":"Abhijit Suprem, Sanjyot Vaidya, Suma Cherkadi, Purva Singh, J. E. Ferreira, C. Pu","doi":"10.1109/CogMI56440.2022.00021","DOIUrl":"https://doi.org/10.1109/CogMI56440.2022.00021","url":null,"abstract":"Machine learning models with explainable predictions are increasingly sought after, especially for real-world, mission-critical applications that require bias detection and risk mitigation. Inherent interpretability, where a model is designed from the ground-up for interpretability, provides intuitive insights and transparent explanations on model prediction and performance. In this paper, we present CoLabel, an approach to build interpretable models with explanations rooted in the ground truth. We demonstrate CoLabel in a vehicle feature extraction application in the context of vehicle make-model recognition (VMMR). By construction, CoLabel performs VMMR with a composite of interpretable features such as vehicle color, type, and make, all based on interpretable annotations of the ground truth labels. First, CoLabel performs corroborative integration to join multiple datasets that each have a subset of desired annotations of color, type, and make. Then, CoLabel uses decomposable branches to extract complementary features corresponding to desired annotations. Finally, CoLabel fuses them together for final predictions. During feature fusion, CoLabel harmonizes complementary branches so that VMMR features are compatible with each other and can be projected to the same semantic space for classification. With inherent interpretability, CoLabel achieves superior performance to the state-of-the-art black-box models, with accuracy of 0.98, 0.95, and 0.94 on CompCars, Cars196, and BoxCars116K, respectively. CoLabel provides intuitive explanations due to constructive interpretability, and subsequently achieves high accuracy and usability in mission-critical situations.","PeriodicalId":211430,"journal":{"name":"2022 IEEE 4th International Conference on Cognitive Machine Intelligence (CogMI)","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2022-05-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131531644","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}