Pub Date : 2022-12-01DOI: 10.1109/CogMI56440.2022.00027
Lingshu Hu, Can Li, Wenbo Wang, Bin Pang, Yi Shang
Recently deep learning methods have achieved great success in understanding and analyzing text messages. In real-world applications, however, labeled text data are often small-sized and imbalanced in classes due to the high cost of data collection and human annotation, limiting the performance of deep learning classifiers. Therefore, this study explores an understudied area—how sample sizes and imbalance ratios influence the performance of deep learning models and augmentation methods—and provides a solution to this problem. Specifically, this study examines the performance of BERT, Word2Vec, and WordNet augmentation methods with BERT fine-tuning on datasets of sizes 500, 1,000, and 2,000 and imbalance ratios of 4:1 and 9:1. Experimental results show that BERT augmentation improves the performance of BERT in detecting the minority class, and the improvement is most significantly (15.6–40.4% F1 increase compared to the base model and 2.8%–10.4% F1 increase compared to the model with the oversampling method) when the data size is small (e.g., 500 training documents) and highly imbalanced (e.g., 9:1). When the data size increases or the imbalance ratio decreases, the improvement generated by the BERT augmentation becomes smaller or insignificant. Moreover, BERT augmentation plus BERT fine-tuning achieves the best performance compared to other models and methods, demonstrating a promising solution for small-sized, highly imbalanced text classification tasks.
{"title":"Performance Evaluation of Text Augmentation Methods with BERT on Small-sized, Imbalanced Datasets","authors":"Lingshu Hu, Can Li, Wenbo Wang, Bin Pang, Yi Shang","doi":"10.1109/CogMI56440.2022.00027","DOIUrl":"https://doi.org/10.1109/CogMI56440.2022.00027","url":null,"abstract":"Recently deep learning methods have achieved great success in understanding and analyzing text messages. In real-world applications, however, labeled text data are often small-sized and imbalanced in classes due to the high cost of data collection and human annotation, limiting the performance of deep learning classifiers. Therefore, this study explores an understudied area—how sample sizes and imbalance ratios influence the performance of deep learning models and augmentation methods—and provides a solution to this problem. Specifically, this study examines the performance of BERT, Word2Vec, and WordNet augmentation methods with BERT fine-tuning on datasets of sizes 500, 1,000, and 2,000 and imbalance ratios of 4:1 and 9:1. Experimental results show that BERT augmentation improves the performance of BERT in detecting the minority class, and the improvement is most significantly (15.6–40.4% F1 increase compared to the base model and 2.8%–10.4% F1 increase compared to the model with the oversampling method) when the data size is small (e.g., 500 training documents) and highly imbalanced (e.g., 9:1). When the data size increases or the imbalance ratio decreases, the improvement generated by the BERT augmentation becomes smaller or insignificant. Moreover, BERT augmentation plus BERT fine-tuning achieves the best performance compared to other models and methods, demonstrating a promising solution for small-sized, highly imbalanced text classification tasks.","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":"126747571","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/CogMI56440.2022.00012
Francesca Naretto, A. Monreale, F. Giannotti
In recent years we are witnessing the diffusion of AI systems based on powerful Machine Learning models which find application in many critical contexts such as medicine, financial market and credit scoring. In such a context it is particularly important to design Trustworthy AI systems while guaranteeing transparency, with respect to their decision reasoning and privacy protection. Although many works in the literature addressed the lack of transparency and the risk of privacy exposure of Machine Learning models, the privacy risks of explainers have not been appropriately studied. This paper presents a methodology for evaluating the privacy exposure raised by interpretable global explainers able to imitate the original black-box classifier. Our methodology exploits the well-known Membership Inference Attack. The experimental results highlight that global explainers based on interpretable trees lead to an increase in privacy exposure.
{"title":"Evaluating the Privacy Exposure of Interpretable Global Explainers","authors":"Francesca Naretto, A. Monreale, F. Giannotti","doi":"10.1109/CogMI56440.2022.00012","DOIUrl":"https://doi.org/10.1109/CogMI56440.2022.00012","url":null,"abstract":"In recent years we are witnessing the diffusion of AI systems based on powerful Machine Learning models which find application in many critical contexts such as medicine, financial market and credit scoring. In such a context it is particularly important to design Trustworthy AI systems while guaranteeing transparency, with respect to their decision reasoning and privacy protection. Although many works in the literature addressed the lack of transparency and the risk of privacy exposure of Machine Learning models, the privacy risks of explainers have not been appropriately studied. This paper presents a methodology for evaluating the privacy exposure raised by interpretable global explainers able to imitate the original black-box classifier. Our methodology exploits the well-known Membership Inference Attack. The experimental results highlight that global explainers based on interpretable trees lead to an increase in privacy exposure.","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":"122314550","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/CogMI56440.2022.00013
Zonyin Shae, Kun-Yi Chen, Chi-Yu Chang, Yuan-Yu Tsai, C. Chou, William I. Baskett, Chi-Ren Shyu, J. J. Tsai
We share the common hypothesis/belief that the more aggregated good quality training data, the better the performance that can be attained by the resulting Artificial Intelligence (AI) model. However, this common belief, in general, is not true in the medical area, since healthcare data sets sourced from different hospitals are often not identically distributed (Non-IID). This imposes severe technical challenges for effectively aggregating the individual hospital data sets together. In this vision paper, instead of offering complete solutions, we will discuss some questions and food for thought with the goal of aiding effective data aggregation and improving federated learning (FL) AI model performance: (1) benchmark and measure the Non-IID degree of medical data sets. (2) include the Non-IID degree metrics in the FL data aggregation mechanism. (3) search for the optimal global model creation strategy among a group of many medical data sets. (4) investigate FL performance better than the centralized learning. This paper will discuss these questions by outlining a visionary approach for exploring a medical blockchain FL mechanism to effectively aggregate medical data across multiple healthcare systems to serve large populations with broad demographics.
{"title":"Thoughts on Non-IID Data Impact in Healthcare with Federated Learning Medical Blockchain","authors":"Zonyin Shae, Kun-Yi Chen, Chi-Yu Chang, Yuan-Yu Tsai, C. Chou, William I. Baskett, Chi-Ren Shyu, J. J. Tsai","doi":"10.1109/CogMI56440.2022.00013","DOIUrl":"https://doi.org/10.1109/CogMI56440.2022.00013","url":null,"abstract":"We share the common hypothesis/belief that the more aggregated good quality training data, the better the performance that can be attained by the resulting Artificial Intelligence (AI) model. However, this common belief, in general, is not true in the medical area, since healthcare data sets sourced from different hospitals are often not identically distributed (Non-IID). This imposes severe technical challenges for effectively aggregating the individual hospital data sets together. In this vision paper, instead of offering complete solutions, we will discuss some questions and food for thought with the goal of aiding effective data aggregation and improving federated learning (FL) AI model performance: (1) benchmark and measure the Non-IID degree of medical data sets. (2) include the Non-IID degree metrics in the FL data aggregation mechanism. (3) search for the optimal global model creation strategy among a group of many medical data sets. (4) investigate FL performance better than the centralized learning. This paper will discuss these questions by outlining a visionary approach for exploring a medical blockchain FL mechanism to effectively aggregate medical data across multiple healthcare systems to serve large populations with broad demographics.","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":"128495506","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/CogMI56440.2022.00025
Alicia Jiayun Law, Ruoyu Hu, Lisa Alazraki, A. Gopalan, Neophytos Polydorou, A. Edalat
In this work, we propose a computational framework that leverages existing out-of-language data to create a conversational agent for the delivery of Self-Attachment Technique (SAT) in Mandarin. Our framework does not require large-scale human translations, yet it achieves a comparable performance whilst also maintaining safety and reliability. We propose two different methods of augmenting available response data through empathetic rewriting. We evaluate our chatbot against a previous, English-only SAT chatbot through non-clinical human trials (N = 42), each lasting five days, and quantitatively show that we are able to attain a comparable level of performance to the English SAT chatbot. We provide qualitative analysis on the limitations of our study and suggestions with the aim of guiding future improvements.
{"title":"A Multilingual Virtual Guide for Self-Attachment Technique","authors":"Alicia Jiayun Law, Ruoyu Hu, Lisa Alazraki, A. Gopalan, Neophytos Polydorou, A. Edalat","doi":"10.1109/CogMI56440.2022.00025","DOIUrl":"https://doi.org/10.1109/CogMI56440.2022.00025","url":null,"abstract":"In this work, we propose a computational framework that leverages existing out-of-language data to create a conversational agent for the delivery of Self-Attachment Technique (SAT) in Mandarin. Our framework does not require large-scale human translations, yet it achieves a comparable performance whilst also maintaining safety and reliability. We propose two different methods of augmenting available response data through empathetic rewriting. We evaluate our chatbot against a previous, English-only SAT chatbot through non-clinical human trials (N = 42), each lasting five days, and quantitatively show that we are able to attain a comparable level of performance to the English SAT chatbot. We provide qualitative analysis on the limitations of our study and suggestions with the aim of guiding future improvements.","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":"125318678","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/CogMI56440.2022.00019
Stacey Truex, Giorgi Alavidze
In March 2020, the World Health Organization (WHO) declared the novel coronavirus (COVID-19) a global pandemic. Globally the rapid spread of COVID-19 ground economies to a halt with stay at home orders and took the lives of millions of people. Therefore when vaccines became available as a tool to slow the spread of the COVID-19 virus, governments world-wide were looking to incentivize their populations to get vaccinated. Included in this effort, the government of Georgia created a lottery initiative to monetarily reward citizens who were vaccinated and encourage participation from those who were hesitant to get vaccinated. The Georgian lottery system that developed out of this initiative included a website displaying lottery winner data leading to serious privacy leakage. In this paper, we develop of an attack framework that allows adversaries with minimal background knowledge to re-identify STOPCOV Lottery winners and deploying our system against a subpopulation vulnerable to attack. We then propose our privacy-enhanced alternative, PSLotto, which simultaneously preserves the functionalities of the existing STOPCOV Lottery system and protects the privacy of lottery winners.
{"title":"PSLotto: A Privacy-Enhanced COVID Lottery System","authors":"Stacey Truex, Giorgi Alavidze","doi":"10.1109/CogMI56440.2022.00019","DOIUrl":"https://doi.org/10.1109/CogMI56440.2022.00019","url":null,"abstract":"In March 2020, the World Health Organization (WHO) declared the novel coronavirus (COVID-19) a global pandemic. Globally the rapid spread of COVID-19 ground economies to a halt with stay at home orders and took the lives of millions of people. Therefore when vaccines became available as a tool to slow the spread of the COVID-19 virus, governments world-wide were looking to incentivize their populations to get vaccinated. Included in this effort, the government of Georgia created a lottery initiative to monetarily reward citizens who were vaccinated and encourage participation from those who were hesitant to get vaccinated. The Georgian lottery system that developed out of this initiative included a website displaying lottery winner data leading to serious privacy leakage. In this paper, we develop of an attack framework that allows adversaries with minimal background knowledge to re-identify STOPCOV Lottery winners and deploying our system against a subpopulation vulnerable to attack. We then propose our privacy-enhanced alternative, PSLotto, which simultaneously preserves the functionalities of the existing STOPCOV Lottery system and protects the privacy of lottery winners.","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":"115579790","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/CogMI56440.2022.00018
Robert K. L. Kennedy, Zahra Salekshahrezaee, T. Khoshgoftaar
Typical fraud datasets lack consistent and accurate labels and, as such, are typically highly imbalanced with non-fraud examples greatly outnumbering the fraudulent ones. This presents significant challenges to machine learning researchers and practitioners. Due to these challenges, an effective approach in identifying fraudulent data points needs to handle highly-imbalanced datasets and be robust to class labeling. This paper introduces a novel unsupervised procedure for learning from imbalanced datasets without class labels by iteratively cleaning the training dataset. Our methodology uses an autoencoder as an underlying learner. We describe its fraud detection performance and compare it to a baseline unsupervised fraud detection learner. Our results show that our procedure significantly outperforms the baseline, in both AUC and TPR, when testing on a publicly available highly-imbalanced credit card fraud detection dataset.
{"title":"A Novel Approach for Unsupervised Learning of Highly-Imbalanced Data","authors":"Robert K. L. Kennedy, Zahra Salekshahrezaee, T. Khoshgoftaar","doi":"10.1109/CogMI56440.2022.00018","DOIUrl":"https://doi.org/10.1109/CogMI56440.2022.00018","url":null,"abstract":"Typical fraud datasets lack consistent and accurate labels and, as such, are typically highly imbalanced with non-fraud examples greatly outnumbering the fraudulent ones. This presents significant challenges to machine learning researchers and practitioners. Due to these challenges, an effective approach in identifying fraudulent data points needs to handle highly-imbalanced datasets and be robust to class labeling. This paper introduces a novel unsupervised procedure for learning from imbalanced datasets without class labels by iteratively cleaning the training dataset. Our methodology uses an autoencoder as an underlying learner. We describe its fraud detection performance and compare it to a baseline unsupervised fraud detection learner. Our results show that our procedure significantly outperforms the baseline, in both AUC and TPR, when testing on a publicly available highly-imbalanced credit card fraud detection dataset.","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":"122491726","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/CogMI56440.2022.00029
Bander Allogmany, D. Josyula
In recent years, personalized learning systems have garnered significant academic research attention in the field of education. In a personalized learning system, learners receive a customized learning style that is tailored to their unique needs, goals, and abilities. Thus, students can achieve their objectives faster than with the traditional method of learning. Rapid advancements in artificial intelligence technologies enable tracking and influencing each student’s learning process. Machine learning algorithms facilitate the determination of students’ learning styles, abilities, and progress throughout the learning process. One of the major challenges to effective personalization is the resistance of machine learning models to adapt to non-stationary data streams. Machine learning models for personalized learning systems are susceptible to the concept drift phenomenon, in which the models’ performance deteriorates over time due to changes in data distribution. For successful personalization, it is critical for the underlying predictive and classification models to adapt to data distribution changes. In this paper, we propose an approach to address concept drifts in personalized learning systems, and evaluate the approach on the OULAD dataset infused with concept drift. The proposed method comprises training utilizing sequential features extracted automatically.
{"title":"An approach to dealing with incremental concept drift in personalized learning systems","authors":"Bander Allogmany, D. Josyula","doi":"10.1109/CogMI56440.2022.00029","DOIUrl":"https://doi.org/10.1109/CogMI56440.2022.00029","url":null,"abstract":"In recent years, personalized learning systems have garnered significant academic research attention in the field of education. In a personalized learning system, learners receive a customized learning style that is tailored to their unique needs, goals, and abilities. Thus, students can achieve their objectives faster than with the traditional method of learning. Rapid advancements in artificial intelligence technologies enable tracking and influencing each student’s learning process. Machine learning algorithms facilitate the determination of students’ learning styles, abilities, and progress throughout the learning process. One of the major challenges to effective personalization is the resistance of machine learning models to adapt to non-stationary data streams. Machine learning models for personalized learning systems are susceptible to the concept drift phenomenon, in which the models’ performance deteriorates over time due to changes in data distribution. For successful personalization, it is critical for the underlying predictive and classification models to adapt to data distribution changes. In this paper, we propose an approach to address concept drifts in personalized learning systems, and evaluate the approach on the OULAD dataset infused with concept drift. The proposed method comprises training utilizing sequential features extracted automatically.","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":"116664090","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/CogMI56440.2022.00026
Yang Zhang, Shiqi Wang, Zhenduo Zhai, Y. Shang, Reid Viegut, Elisabeth Webb, A. Raedeke, J. Sartwell
Monitoring waterfowl populations and distribution is important for conservation. This paper presents our recent work on creating new aerial image datasets collected by drones and applying and evaluating state-of-the-art deep learning models for waterfowl detection and classification. We collected thousands of aerial images from 10 conservation areas in Missouri, labeled around 600 images with close to 300,000 bird labels, and created 9 datasets with different properties for training and evaluating deep neural network models. Among the models, YOLOv5 performed the best, outperforming Faster R-CNN and RetinaNet. To reduce the amount of labeled data needed for model training, we applied Soft Teacher, a semi-supervised learning method, and obtained slightly better detection performance than supervised learning methods, with just half of the labeled training examples. We trained generic detection models using all datasets containing diverse images and obtained accurate detection results in most cases. For waterfowl classification, we created a dataset of images containing individual waterfowl by cropping them from raw aerial images. We applied several deep learning models to the dataset and obtained promising results.
{"title":"Development of New Aerial Image Datasets and Deep Learning Methods for Waterfowl Detection and Classification","authors":"Yang Zhang, Shiqi Wang, Zhenduo Zhai, Y. Shang, Reid Viegut, Elisabeth Webb, A. Raedeke, J. Sartwell","doi":"10.1109/CogMI56440.2022.00026","DOIUrl":"https://doi.org/10.1109/CogMI56440.2022.00026","url":null,"abstract":"Monitoring waterfowl populations and distribution is important for conservation. This paper presents our recent work on creating new aerial image datasets collected by drones and applying and evaluating state-of-the-art deep learning models for waterfowl detection and classification. We collected thousands of aerial images from 10 conservation areas in Missouri, labeled around 600 images with close to 300,000 bird labels, and created 9 datasets with different properties for training and evaluating deep neural network models. Among the models, YOLOv5 performed the best, outperforming Faster R-CNN and RetinaNet. To reduce the amount of labeled data needed for model training, we applied Soft Teacher, a semi-supervised learning method, and obtained slightly better detection performance than supervised learning methods, with just half of the labeled training examples. We trained generic detection models using all datasets containing diverse images and obtained accurate detection results in most cases. For waterfowl classification, we created a dataset of images containing individual waterfowl by cropping them from raw aerial images. We applied several deep learning models to the dataset and obtained promising results.","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":"123355984","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/CogMI56440.2022.00030
Yang Zhang, Yizhen Wang, Zhicheng Tang, Zhenduo Zhai, Y. Shang, Reid Viegut
This paper presents the results of our deep learning methods for tree detection and classification on aerial images in the Plant Recognition University Challenge sponsored by Ameren in 2021–2022. The task was to locate the trees in an aerial image and predict their family, genus, and species. For tree detection, we applied various supervised learning methods with labeled training data as well as semi-supervised learning methods with the addition of unlabeled data. Our experimental results show that the semi-supervised learning method outperformed the supervised learning methods, improving the f1-score by an average of three percent on the set of images used in the final Plant Challenge competition. For tree classification, We applied various machine learning methods and deep learning models for image classification to predict family, genus and species on the portions of images detected of trees by the detection models. By considering the relationships between family, genus and species, we developed a multi-head ResNet18-based neural network and increased mean accuracy by two percent over the baseline ResNet18. Finally, our team ranked first among all teams in the Plant Challenge competition.
{"title":"Deep Learning Methods for Tree Detection and Classification","authors":"Yang Zhang, Yizhen Wang, Zhicheng Tang, Zhenduo Zhai, Y. Shang, Reid Viegut","doi":"10.1109/CogMI56440.2022.00030","DOIUrl":"https://doi.org/10.1109/CogMI56440.2022.00030","url":null,"abstract":"This paper presents the results of our deep learning methods for tree detection and classification on aerial images in the Plant Recognition University Challenge sponsored by Ameren in 2021–2022. The task was to locate the trees in an aerial image and predict their family, genus, and species. For tree detection, we applied various supervised learning methods with labeled training data as well as semi-supervised learning methods with the addition of unlabeled data. Our experimental results show that the semi-supervised learning method outperformed the supervised learning methods, improving the f1-score by an average of three percent on the set of images used in the final Plant Challenge competition. For tree classification, We applied various machine learning methods and deep learning models for image classification to predict family, genus and species on the portions of images detected of trees by the detection models. By considering the relationships between family, genus and species, we developed a multi-head ResNet18-based neural network and increased mean accuracy by two percent over the baseline ResNet18. Finally, our team ranked first among all teams in the Plant Challenge competition.","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":"132599943","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/CogMI56440.2022.00028
DeMarcus Edwards, D. Rawat, Brian M. Sadler
Short-form video content is fast to consume, easy to digest, and for most creators, inexpensive to make. Content Creators on Video Content platforms have a vested interest in having their videos appear as high as possible in recommendations that users are shown. This paper demonstrates how content creators can manipulate video content to adversarially promote their ranking in a recommendation model that uses action classification labels as an input feature. We focus on the context of these videos in terms of action classification to extract context about these videos to then rank and adversarially promote. Our attack successfully boosted the predicted like probability in 78 percent of generated lists for our model trained with non-perturbed inputs. However, after adversarial training, our model trained with perturbed inputs was 20 percent less effective in boosting the rank of targeted videos.
{"title":"Adversarial Promotion for Video based Recommender Systems","authors":"DeMarcus Edwards, D. Rawat, Brian M. Sadler","doi":"10.1109/CogMI56440.2022.00028","DOIUrl":"https://doi.org/10.1109/CogMI56440.2022.00028","url":null,"abstract":"Short-form video content is fast to consume, easy to digest, and for most creators, inexpensive to make. Content Creators on Video Content platforms have a vested interest in having their videos appear as high as possible in recommendations that users are shown. This paper demonstrates how content creators can manipulate video content to adversarially promote their ranking in a recommendation model that uses action classification labels as an input feature. We focus on the context of these videos in terms of action classification to extract context about these videos to then rank and adversarially promote. Our attack successfully boosted the predicted like probability in 78 percent of generated lists for our model trained with non-perturbed inputs. However, after adversarial training, our model trained with perturbed inputs was 20 percent less effective in boosting the rank of targeted videos.","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":"134373178","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}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：