Pub Date : 2022-12-01DOI: 10.1109/ICMLA55696.2022.00102
Samuel Blad, Martin Längkvist, Franziska Klügl-Frohnmeyer, A. Loutfi
Q-Networks are used in Reinforcement Learning to model the expected return from every action at a given state. When training Q-Networks, external reward signals are propagated to the previously performed actions leading up to each reward. If many actions are required before experiencing a reward, the reward signal is distributed across all those actions, where some actions may have greater impact on the reward than others. As the number of significant actions between rewards increases, the relative importance of each action decreases. If actions have too small importance, their impact might be overshadowed by noise in a deep neural network model, potentially causing convergence issues. In this work, we empirically test the limits of increasing the number of actions leading up to a reward in a simple grid-world environment. We show in our experiments that even though the training error surpasses the reward signal attributed to each action, the model is still able to learn a smooth enough value representation.
{"title":"Empirical analysis of the convergence of Double DQN in relation to reward sparsity","authors":"Samuel Blad, Martin Längkvist, Franziska Klügl-Frohnmeyer, A. Loutfi","doi":"10.1109/ICMLA55696.2022.00102","DOIUrl":"https://doi.org/10.1109/ICMLA55696.2022.00102","url":null,"abstract":"Q-Networks are used in Reinforcement Learning to model the expected return from every action at a given state. When training Q-Networks, external reward signals are propagated to the previously performed actions leading up to each reward. If many actions are required before experiencing a reward, the reward signal is distributed across all those actions, where some actions may have greater impact on the reward than others. As the number of significant actions between rewards increases, the relative importance of each action decreases. If actions have too small importance, their impact might be overshadowed by noise in a deep neural network model, potentially causing convergence issues. In this work, we empirically test the limits of increasing the number of actions leading up to a reward in a simple grid-world environment. We show in our experiments that even though the training error surpasses the reward signal attributed to each action, the model is still able to learn a smooth enough value representation.","PeriodicalId":128160,"journal":{"name":"2022 21st IEEE International Conference on Machine Learning and Applications (ICMLA)","volume":"78 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127233843","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/ICMLA55696.2022.00202
Y. Manyari, P. Callet, Laurent Dollé
Teaching robots behavioral skills by leveraging examples provided by an expert, also referred to as Imitation Learning from Observation (IfO or ILO), is a promising approach for learning novel tasks without requiring a task-specific reward function to be engineered. We propose a RL-based framework to teach robots manipulation tasks given expert observation-only demonstrations. First, a representation model is trained to extract spatial and temporal features from demonstrations. Graph Neural Networks (GNNs) are used to encode spatial patterns, while LSTMs and Transformers are used to encode temporal features. Second, based on an off-the-shelf RL algorithm, the demonstrations are leveraged through the trained representation to guide the policy training towards solving the task demonstrated by the expert. We show that our approach compares favorably to state-of-the-art IfO algorithms with a 99% success rate and transfers well to the real world.
{"title":"Imitation from Observation using RL and Graph-based Representation of Demonstrations","authors":"Y. Manyari, P. Callet, Laurent Dollé","doi":"10.1109/ICMLA55696.2022.00202","DOIUrl":"https://doi.org/10.1109/ICMLA55696.2022.00202","url":null,"abstract":"Teaching robots behavioral skills by leveraging examples provided by an expert, also referred to as Imitation Learning from Observation (IfO or ILO), is a promising approach for learning novel tasks without requiring a task-specific reward function to be engineered. We propose a RL-based framework to teach robots manipulation tasks given expert observation-only demonstrations. First, a representation model is trained to extract spatial and temporal features from demonstrations. Graph Neural Networks (GNNs) are used to encode spatial patterns, while LSTMs and Transformers are used to encode temporal features. Second, based on an off-the-shelf RL algorithm, the demonstrations are leveraged through the trained representation to guide the policy training towards solving the task demonstrated by the expert. We show that our approach compares favorably to state-of-the-art IfO algorithms with a 99% success rate and transfers well to the real world.","PeriodicalId":128160,"journal":{"name":"2022 21st IEEE International Conference on Machine Learning and Applications (ICMLA)","volume":"13 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127356364","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/ICMLA55696.2022.00025
H. Mahyar, Peter Tulala, E. Ghalebi, R. Grosu
A certain amount of process deviations characterizes semiconductor manufacturing processes. Automated detection of these production issues followed by an automated root cause analysis has the potential to increase the effectiveness of semiconductor production. Manufacturing defects exhibit typical patterns in measured wafer test data, e.g., rings, spots, repetitive textures, or scratches. Recognizing these patterns is an essential step for finding the root cause of production issues. This paper demonstrates that combining Information Maximizing Generative Adversarial Network (InfoGAN) and Wasserstein GAN (WGAN) with a new loss function is suitable for extracting the most characteristic features from extensive real-world sensory wafer test data, which in various aspects outperforms traditional unsupervised techniques. These features are then used in subsequent clustering tasks to group wafers into clusters according to their exhibit patterns. The primary outcome of this work is a statistical generative model for recognizing spatial wafermaps patterns using deep adversarial neural networks. We experimentally evaluate the performance of the proposed approach over a real dataset.
一定数量的工艺偏差是半导体制造过程的特征。自动化检测这些生产问题，然后进行自动化的根本原因分析，有可能提高半导体生产的效率。在测量的晶圆测试数据中，制造缺陷表现出典型的模式，例如，环、斑点、重复纹理或划痕。识别这些模式是找到生产问题的根本原因的必要步骤。本文证明了将信息最大化生成对抗网络(InfoGAN)和Wasserstein GAN (WGAN)与一种新的损失函数相结合，适用于从大量真实感官晶片测试数据中提取最具特征的特征，在许多方面优于传统的无监督技术。然后在后续的集群任务中使用这些特性，根据晶圆的显示模式将其分组到集群中。这项工作的主要成果是一个使用深度对抗神经网络识别空间晶圆图模式的统计生成模型。我们通过实验评估了在真实数据集上提出的方法的性能。
{"title":"DeepWafer: A Generative Wafermap Model with Deep Adversarial Networks","authors":"H. Mahyar, Peter Tulala, E. Ghalebi, R. Grosu","doi":"10.1109/ICMLA55696.2022.00025","DOIUrl":"https://doi.org/10.1109/ICMLA55696.2022.00025","url":null,"abstract":"A certain amount of process deviations characterizes semiconductor manufacturing processes. Automated detection of these production issues followed by an automated root cause analysis has the potential to increase the effectiveness of semiconductor production. Manufacturing defects exhibit typical patterns in measured wafer test data, e.g., rings, spots, repetitive textures, or scratches. Recognizing these patterns is an essential step for finding the root cause of production issues. This paper demonstrates that combining Information Maximizing Generative Adversarial Network (InfoGAN) and Wasserstein GAN (WGAN) with a new loss function is suitable for extracting the most characteristic features from extensive real-world sensory wafer test data, which in various aspects outperforms traditional unsupervised techniques. These features are then used in subsequent clustering tasks to group wafers into clusters according to their exhibit patterns. The primary outcome of this work is a statistical generative model for recognizing spatial wafermaps patterns using deep adversarial neural networks. We experimentally evaluate the performance of the proposed approach over a real dataset.","PeriodicalId":128160,"journal":{"name":"2022 21st IEEE International Conference on Machine Learning and Applications (ICMLA)","volume":"73 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130621903","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/ICMLA55696.2022.00021
Yifan Zhao, Xian Yang, Ranga Raju Vatsavai
For more than five decades, remote sensing imagery has been providing critical information for many applications such as crop monitoring, disaster assessment, and urban planning. Unfortunately, more than 50% of optical remote sensing images are contaminated by clouds severely affecting the object identification. However, thanks to recent advances in remote sensing instruments and increase in number of operational satellites, we now have petabytes of multi-sensor observations covering the globe. Historically cloud imputation techniques were designed for single sensor images, thus existing benchmarks were mostly limited to single sensor images, which precludes design and validation of cloud imputation techniques on multi-sensor data. In this paper, we introduce a new benchmark data set consisting of images from two widely used and publicly available satellite images, Landsat-8 and Sentinel-2, and a new multi-stream deep residual network (MDRN). This newly introduced benchmark dataset fills an important gap in the existing benchmark datasets, which allows exploitation of multi-resolution spectral information from the cloud-free regions of temporally nearby images, and the MDRN algorithm addresses imputation using the multi-resolution data. Both quantitative and qualitative experiments show that the utility of our benchmark dataset and as well as efficacy of our MDRN architecture in cloud imputation. The MDRN outperforms the closest competing method by 14.1%.
{"title":"Multi-stream Deep Residual Network for Cloud Imputation Using Multi-resolution Remote Sensing Imagery","authors":"Yifan Zhao, Xian Yang, Ranga Raju Vatsavai","doi":"10.1109/ICMLA55696.2022.00021","DOIUrl":"https://doi.org/10.1109/ICMLA55696.2022.00021","url":null,"abstract":"For more than five decades, remote sensing imagery has been providing critical information for many applications such as crop monitoring, disaster assessment, and urban planning. Unfortunately, more than 50% of optical remote sensing images are contaminated by clouds severely affecting the object identification. However, thanks to recent advances in remote sensing instruments and increase in number of operational satellites, we now have petabytes of multi-sensor observations covering the globe. Historically cloud imputation techniques were designed for single sensor images, thus existing benchmarks were mostly limited to single sensor images, which precludes design and validation of cloud imputation techniques on multi-sensor data. In this paper, we introduce a new benchmark data set consisting of images from two widely used and publicly available satellite images, Landsat-8 and Sentinel-2, and a new multi-stream deep residual network (MDRN). This newly introduced benchmark dataset fills an important gap in the existing benchmark datasets, which allows exploitation of multi-resolution spectral information from the cloud-free regions of temporally nearby images, and the MDRN algorithm addresses imputation using the multi-resolution data. Both quantitative and qualitative experiments show that the utility of our benchmark dataset and as well as efficacy of our MDRN architecture in cloud imputation. The MDRN outperforms the closest competing method by 14.1%.","PeriodicalId":128160,"journal":{"name":"2022 21st IEEE International Conference on Machine Learning and Applications (ICMLA)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131002729","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/ICMLA55696.2022.00037
Brian Koga, Theresa VanderWeide, Xinghui Zhao, Xuechen Zhang
Deep neural networks (DNNs) have recently gained unprecedented success in various domains. In resource-constrained edge systems (e.g., mobile devices and IoT devices) QoS-aware DNNs are required to meet latency and memory/storage requirements of mission-critical deep learning applications. However, none of the existing DNNs has been de-signed to satisfy both latency and memory bounds simultaneously as specified by end-users in the resource-constrained systems. This paper proposes a runtime system, BlinkNet, which can guarantee both latency and memory/storage bounds for one or multiple DNNs via efficient QoS-aware per-layer approximation. We implement BlinkNet in Apache TVM and evaluate it using CaffeNet, CIFAR-10-quick, and VGG16 network models on both CPU and GPU platforms. Our experimental results show that BlinkNet can enforce various latency and memory bounds set by end-users with real-world datasets.
{"title":"BlinkNet: Software-Defined Deep Learning Analytics with Bounded Resources","authors":"Brian Koga, Theresa VanderWeide, Xinghui Zhao, Xuechen Zhang","doi":"10.1109/ICMLA55696.2022.00037","DOIUrl":"https://doi.org/10.1109/ICMLA55696.2022.00037","url":null,"abstract":"Deep neural networks (DNNs) have recently gained unprecedented success in various domains. In resource-constrained edge systems (e.g., mobile devices and IoT devices) QoS-aware DNNs are required to meet latency and memory/storage requirements of mission-critical deep learning applications. However, none of the existing DNNs has been de-signed to satisfy both latency and memory bounds simultaneously as specified by end-users in the resource-constrained systems. This paper proposes a runtime system, BlinkNet, which can guarantee both latency and memory/storage bounds for one or multiple DNNs via efficient QoS-aware per-layer approximation. We implement BlinkNet in Apache TVM and evaluate it using CaffeNet, CIFAR-10-quick, and VGG16 network models on both CPU and GPU platforms. Our experimental results show that BlinkNet can enforce various latency and memory bounds set by end-users with real-world datasets.","PeriodicalId":128160,"journal":{"name":"2022 21st IEEE International Conference on Machine Learning and Applications (ICMLA)","volume":"45 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125373340","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/ICMLA55696.2022.00011
N. Nafi, Creighton Glasscock, W. Hsu
In this work, we introduce Attention-based Partially Decoupled Actor-Critic (APDAC), an actor-critic architecture for generalization in reinforcement learning, which partially separates the policy and the value functions. To learn directly from images, traditional actor-critic architectures use a shared network to represent the policy and value functions. While a shared representation allows parameter and feature sharing, it can also lead to overfitting that catastrophically damages generalization performance. On the other hand, two separate networks for policy and value can help to avoid overfitting and reduce the generalization gap, but at the cost of added complexity both in terms of architecture design and computation time. APDAC is a hybrid architecture that builds upon the combined strengths of both architectures by sharing initial layer blocks of the network and separating the later ones for policy and value. APDAC incorporates an attention mechanism to enable robust representation learning. We present meaningful visualization of the policy and value that explains the perception of the trained agent. Our empirical analysis, including an ablation study, shows that APDAC significantly outperforms the standard PPO baseline on the challenging RL generalization benchmark Procgen and achieves performance that is competitive with the recent state-of-the-art method (IDAAC) while using fewer convolutional layers and requiring less computational time. Our code is available at https://github.com/nasiknafi/apdac.
{"title":"Attention-based Partial Decoupling of Policy and Value for Generalization in Reinforcement Learning","authors":"N. Nafi, Creighton Glasscock, W. Hsu","doi":"10.1109/ICMLA55696.2022.00011","DOIUrl":"https://doi.org/10.1109/ICMLA55696.2022.00011","url":null,"abstract":"In this work, we introduce Attention-based Partially Decoupled Actor-Critic (APDAC), an actor-critic architecture for generalization in reinforcement learning, which partially separates the policy and the value functions. To learn directly from images, traditional actor-critic architectures use a shared network to represent the policy and value functions. While a shared representation allows parameter and feature sharing, it can also lead to overfitting that catastrophically damages generalization performance. On the other hand, two separate networks for policy and value can help to avoid overfitting and reduce the generalization gap, but at the cost of added complexity both in terms of architecture design and computation time. APDAC is a hybrid architecture that builds upon the combined strengths of both architectures by sharing initial layer blocks of the network and separating the later ones for policy and value. APDAC incorporates an attention mechanism to enable robust representation learning. We present meaningful visualization of the policy and value that explains the perception of the trained agent. Our empirical analysis, including an ablation study, shows that APDAC significantly outperforms the standard PPO baseline on the challenging RL generalization benchmark Procgen and achieves performance that is competitive with the recent state-of-the-art method (IDAAC) while using fewer convolutional layers and requiring less computational time. Our code is available at https://github.com/nasiknafi/apdac.","PeriodicalId":128160,"journal":{"name":"2022 21st IEEE International Conference on Machine Learning and Applications (ICMLA)","volume":"146 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123390348","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/ICMLA55696.2022.00020
Charles De Trogoff, Rim Hantach, Gisela Lechuga, P. Calvez
Currently, the need for business documents analysis, particularly invoices, is playing a vital role in companies, especially in large ones. These documents have the particularity of being visually rich, with low text quantity and many different layouts. As such, processing them with traditional techniques remains inefficient. Hence, one of the key challenge is to exploit visual patterns between entities of interest. After an overview of the state-of-the-art in this domain, we propose a graph-based model that recognizes specific text in invoices. First, an Encoder module creates a multimodal embedding for each text sequence based on textual, visual, and spatial information. This representation is then passed through a multi-layer graph attention network, before being subjected to a simple classification task. Some experimental results were conducted in order to improve the performance of the proposed approach.
{"title":"Automatic Key Information Extraction from Visually Rich Documents","authors":"Charles De Trogoff, Rim Hantach, Gisela Lechuga, P. Calvez","doi":"10.1109/ICMLA55696.2022.00020","DOIUrl":"https://doi.org/10.1109/ICMLA55696.2022.00020","url":null,"abstract":"Currently, the need for business documents analysis, particularly invoices, is playing a vital role in companies, especially in large ones. These documents have the particularity of being visually rich, with low text quantity and many different layouts. As such, processing them with traditional techniques remains inefficient. Hence, one of the key challenge is to exploit visual patterns between entities of interest. After an overview of the state-of-the-art in this domain, we propose a graph-based model that recognizes specific text in invoices. First, an Encoder module creates a multimodal embedding for each text sequence based on textual, visual, and spatial information. This representation is then passed through a multi-layer graph attention network, before being subjected to a simple classification task. Some experimental results were conducted in order to improve the performance of the proposed approach.","PeriodicalId":128160,"journal":{"name":"2022 21st IEEE International Conference on Machine Learning and Applications (ICMLA)","volume":"32 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121302281","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/ICMLA55696.2022.00063
Abdullah Alsaedi, S. Thomason, F. Grasso, Phillip Brooker
Social emotion prediction is concerned with the prediction of the reader’s emotion when exposed to a text. In this paper, we propose a transfer learning approach to social emotion prediction, where the source task is writer’s emotion prediction, an area in which models are advanced due to the rich literature and availability of large and high-quality training datasets. We utilized a pre-trained writer’s emotion prediction model to predict the writer’s emotion in comments, then we aggregated the emotions and trained a classifier to predict social emotion for posts. Results show that pre-trained models for writer’s emotion prediction can improve the prediction of social emotion. Furthermore, we demonstrate that our proposed model outperforms popular models in terms of F1-score and performs similarly to the best model in terms of Acc@1.
{"title":"Transfer Learning model for Social Emotion Prediction using Writers Emotions in Comments","authors":"Abdullah Alsaedi, S. Thomason, F. Grasso, Phillip Brooker","doi":"10.1109/ICMLA55696.2022.00063","DOIUrl":"https://doi.org/10.1109/ICMLA55696.2022.00063","url":null,"abstract":"Social emotion prediction is concerned with the prediction of the reader’s emotion when exposed to a text. In this paper, we propose a transfer learning approach to social emotion prediction, where the source task is writer’s emotion prediction, an area in which models are advanced due to the rich literature and availability of large and high-quality training datasets. We utilized a pre-trained writer’s emotion prediction model to predict the writer’s emotion in comments, then we aggregated the emotions and trained a classifier to predict social emotion for posts. Results show that pre-trained models for writer’s emotion prediction can improve the prediction of social emotion. Furthermore, we demonstrate that our proposed model outperforms popular models in terms of F1-score and performs similarly to the best model in terms of Acc@1.","PeriodicalId":128160,"journal":{"name":"2022 21st IEEE International Conference on Machine Learning and Applications (ICMLA)","volume":"43 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126924424","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/ICMLA55696.2022.00177
Indrajeet Ghosh, Avijoy Chakma, S. R. Ramamurthy, Nirmalya Roy, Nicholas R. Waytowich
Intelligent and complex human motion analysis can help design the next generation IoT and AR/VR systems for automated human performance assessment. Such an automated system can help advocate the interpretability and translatability of complex human motions, intelligent motion feedback, and fine-grained motion skill assessment to design next-generation interactive human-machine teaming systems. Motivated by this, we design a wearable sensing framework for assessing the players’ performance and consider a live badminton game as our use case. Generally, the players on the field try to improve their performance by focusing on fast and synchronous coordination of their limbs’ reflex actions to have the ideal body postures to perform the desired shot. Learning the minute dissimilarities and distinctive traits from each limb of the players simultaneously can help assess the players’ performance and specific skillsets during a game. This paper proposes a multi-task learning framework, PerMTL to learn the shared features from each player’s limb. The PerMTL comprises a task-specific regressor output layer that helps to determine the dissimilarities and distinctive traits between the player’s limbs for collective inference in a body sensor network (BSN) environment. We evaluate the PerMTL framework using publicly available Badminton Activity Recognition (BAR) and Daily and Sports Activities (DSA) datasets. Empirical results indicate that PerMTL achieves R2 Score of ≈ 82% in predicting the players’ performance.
{"title":"PerMTL: A Multi-Task Learning Framework for Skilled Human Performance Assessment","authors":"Indrajeet Ghosh, Avijoy Chakma, S. R. Ramamurthy, Nirmalya Roy, Nicholas R. Waytowich","doi":"10.1109/ICMLA55696.2022.00177","DOIUrl":"https://doi.org/10.1109/ICMLA55696.2022.00177","url":null,"abstract":"Intelligent and complex human motion analysis can help design the next generation IoT and AR/VR systems for automated human performance assessment. Such an automated system can help advocate the interpretability and translatability of complex human motions, intelligent motion feedback, and fine-grained motion skill assessment to design next-generation interactive human-machine teaming systems. Motivated by this, we design a wearable sensing framework for assessing the players’ performance and consider a live badminton game as our use case. Generally, the players on the field try to improve their performance by focusing on fast and synchronous coordination of their limbs’ reflex actions to have the ideal body postures to perform the desired shot. Learning the minute dissimilarities and distinctive traits from each limb of the players simultaneously can help assess the players’ performance and specific skillsets during a game. This paper proposes a multi-task learning framework, PerMTL to learn the shared features from each player’s limb. The PerMTL comprises a task-specific regressor output layer that helps to determine the dissimilarities and distinctive traits between the player’s limbs for collective inference in a body sensor network (BSN) environment. We evaluate the PerMTL framework using publicly available Badminton Activity Recognition (BAR) and Daily and Sports Activities (DSA) datasets. Empirical results indicate that PerMTL achieves R2 Score of ≈ 82% in predicting the players’ performance.","PeriodicalId":128160,"journal":{"name":"2022 21st IEEE International Conference on Machine Learning and Applications (ICMLA)","volume":"52 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115284275","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/ICMLA55696.2022.00087
Susana Lavado, Miguel Mateus, Leid Zejnilovic
Predicting students’ retention risk is one of the major trends in machine learning applications in education. While early identification of at-risk students allows timely planning and implementation of measures to prevent adverse outcomes, there is a trade-off between the predictive model’s performance and the prediction window size, or model performance and its actionability. In this study, we used a dataset of 83,596 unique Portuguese students in grades 5th to 9th to predict retention at or before the end of 9th grade. We explored how different prediction window sizes impact the predictive model’s performance, the feature importance, and the models’ bias. The models with the shorter prediction window performed better in terms of precision, but the model with the largest prediction window showed a higher lift over the existing rule-based model. Prediction window size impacted the importance of demographic features and model’s fairness. Our results contribute to the extant discussion on predicting retention, by adding empirical evidence about the models’ added value in performance versus the existing practice, suggesting types of data to collect and use, and discussing education-specific challenges of responsible data science.
{"title":"The Performance-Actionability Trade-Off in Retention Prediction at Middle School","authors":"Susana Lavado, Miguel Mateus, Leid Zejnilovic","doi":"10.1109/ICMLA55696.2022.00087","DOIUrl":"https://doi.org/10.1109/ICMLA55696.2022.00087","url":null,"abstract":"Predicting students’ retention risk is one of the major trends in machine learning applications in education. While early identification of at-risk students allows timely planning and implementation of measures to prevent adverse outcomes, there is a trade-off between the predictive model’s performance and the prediction window size, or model performance and its actionability. In this study, we used a dataset of 83,596 unique Portuguese students in grades 5th to 9th to predict retention at or before the end of 9th grade. We explored how different prediction window sizes impact the predictive model’s performance, the feature importance, and the models’ bias. The models with the shorter prediction window performed better in terms of precision, but the model with the largest prediction window showed a higher lift over the existing rule-based model. Prediction window size impacted the importance of demographic features and model’s fairness. Our results contribute to the extant discussion on predicting retention, by adding empirical evidence about the models’ added value in performance versus the existing practice, suggesting types of data to collect and use, and discussing education-specific challenges of responsible data science.","PeriodicalId":128160,"journal":{"name":"2022 21st IEEE International Conference on Machine Learning and Applications (ICMLA)","volume":"23 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125228784","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：