Wow, there is so much to learn in this Competition. Forget any prize. learning is the focus.
\nOne of the Fine-grained categorization (Human-in-the-loop) is \"Embedding human experts’ knowledge into computational models\" Let's see what we can get Googling it:
\nAuthors: Masahiro Mitsuhara, Hiroshi Fukui, Yusuke Sakashita, Takanori Ogata, Tsubasa Hirakawa, Takayoshi Yamashita, Hironobu Fujiyoshi
\nCitation: arXiv:1905.03540 [cs.CV]
\n\"In this work, the authors aimed to realize a method for embedding human knowledge into deep neural networks. While the conventional method to embed human knowledge has been applied for non-deep machine learning, it is challenging to apply it for deep learning models due to the enormous number of model parameters.\"
\n\"In this paper, they proposed a fine-tuning method that utilizes a single-channel attention map which is manually edited by a human expert. Their fine-tuning method can train a network so that the output attention map corresponds to the edited ones. As a result, the fine-tuned network can output an attention map that takes into account human knowledge.\"
\n\"Experimental results with ImageNet, CUB-200-2010, and IDRiD demonstrated that it is possible to obtain a clear attention map for a visual explanation and improve the classification performance. Their findings can be a novel framework for optimizing networks through human intuitive editing via a visual interface and suggest new possibilities for human-machine cooperation in addition to the improvement of visual explanations.\"
\nhttps://arxiv.org/abs/1905.03540
\nAuthors: Changyu Deng ; Xunbi Ji; Colton Rainey; Jianyu Zhang ; Wei Lu
\nhttps://doi.org/10.1016/j.isci.2020.101656
\nHIGHLIGHTS
\nIntegrating knowledge into machine learning delivers superior performance
\nKnowledge is categorized and its representations are presented
\nVarious methods to bridge human knowledge and machine learning are shown
\nSuggestions on approaches and perspectives on future research directions are provided
\n\"Multitask Learning\"
\n\"Humans do not learn individual tasks in a linear sequence, but they learn several tasks simultaneously. This efficient behavior is replicated in machine learning with multitask learning (MTL). MTL shares knowledge between tasks so they are all learned simultaneously with higher overall performance .\"
\nRead all these in that paper ( Model Assumptions, Preliminaries of Probabilistic ML, Deterministic Assumptions, Network Architecture, Sy,mmetry of Convolutional Neural Networks, Design of Neuron Connections, Data Augmentation, Simulation, Reinforcement Learning,Active Learning, Interactive Visual Analytics, Parameter Initialization, Transfer Learning.
\n\"As humans create faster and more accurate knowledge-based models to simulate the world, using simulations to acquire a large amount of data becomes an increasingly efficient method for machine learning. The primary advantage of simulations is the ability to gather a large amount of data when experimentally would be costly, time consuming, or even dangerous\".
\n\"If using neural networks, tailor the architecture to be suitable for the tasks. If possible, incorporate some known properties, such as Symmetry of Convolutional Neural Networks. Logic, equations, and temporal nature can be, respectively, reflected in the structure of networks by, for instance, combining with symbolic AI, designing special layers/architectures, and using RNNs.\"
\n\"Design algorithms to include humans in the loop. The interaction between machine and environment can be modeled and optimized in Reinforcement Learning . Humans can be asked to label data or provide distribution (Active Learning). Interactive Visual Analytics can be used to help humans understand machine learning results and then adjust models during or after training.\"
\n\"The authors can regulate the intermediate results or network layers to produce models more understandable and controllable by humans.\"
\n\"Designing and implementing machine learning algorithms is an iterative process. This requires humans to analyze the models and knowledge integration to take advantage of human understanding of the real world. That review may help current and prospective users of machine learning to understand these fields and inspire them to build more efficient models.\"
\nhttps://www.sciencedirect.com/science/article/pii/S2589004220308488
", "messages": [ { "id": 1694999, "postDate": "2022-02-17T22:05:14.043Z", "content": "Wow, there is so much to learn in this Competition. Forget any prize. learning is the focus.
\nOne of the Fine-grained categorization (Human-in-the-loop) is \"Embedding human experts’ knowledge into computational models\" Let's see what we can get Googling it:
\nAuthors: Masahiro Mitsuhara, Hiroshi Fukui, Yusuke Sakashita, Takanori Ogata, Tsubasa Hirakawa, Takayoshi Yamashita, Hironobu Fujiyoshi
\nCitation: arXiv:1905.03540 [cs.CV]
\n\"In this work, the authors aimed to realize a method for embedding human knowledge into deep neural networks. While the conventional method to embed human knowledge has been applied for non-deep machine learning, it is challenging to apply it for deep learning models due to the enormous number of model parameters.\"
\n\"In this paper, they proposed a fine-tuning method that utilizes a single-channel attention map which is manually edited by a human expert. Their fine-tuning method can train a network so that the output attention map corresponds to the edited ones. As a result, the fine-tuned network can output an attention map that takes into account human knowledge.\"
\n\"Experimental results with ImageNet, CUB-200-2010, and IDRiD demonstrated that it is possible to obtain a clear attention map for a visual explanation and improve the classification performance. Their findings can be a novel framework for optimizing networks through human intuitive editing via a visual interface and suggest new possibilities for human-machine cooperation in addition to the improvement of visual explanations.\"
\nhttps://arxiv.org/abs/1905.03540
\nAuthors: Changyu Deng ; Xunbi Ji; Colton Rainey; Jianyu Zhang ; Wei Lu
\nhttps://doi.org/10.1016/j.isci.2020.101656
\nHIGHLIGHTS
\nIntegrating knowledge into machine learning delivers superior performance
\nKnowledge is categorized and its representations are presented
\nVarious methods to bridge human knowledge and machine learning are shown
\nSuggestions on approaches and perspectives on future research directions are provided
\n\"Multitask Learning\"
\n\"Humans do not learn individual tasks in a linear sequence, but they learn several tasks simultaneously. This efficient behavior is replicated in machine learning with multitask learning (MTL). MTL shares knowledge between tasks so they are all learned simultaneously with higher overall performance .\"
\nRead all these in that paper ( Model Assumptions, Preliminaries of Probabilistic ML, Deterministic Assumptions, Network Architecture, Sy,mmetry of Convolutional Neural Networks, Design of Neuron Connections, Data Augmentation, Simulation, Reinforcement Learning,Active Learning, Interactive Visual Analytics, Parameter Initialization, Transfer Learning.
\n\"As humans create faster and more accurate knowledge-based models to simulate the world, using simulations to acquire a large amount of data becomes an increasingly efficient method for machine learning. The primary advantage of simulations is the ability to gather a large amount of data when experimentally would be costly, time consuming, or even dangerous\".
\n\"If using neural networks, tailor the architecture to be suitable for the tasks. If possible, incorporate some known properties, such as Symmetry of Convolutional Neural Networks. Logic, equations, and temporal nature can be, respectively, reflected in the structure of networks by, for instance, combining with symbolic AI, designing special layers/architectures, and using RNNs.\"
\n\"Design algorithms to include humans in the loop. The interaction between machine and environment can be modeled and optimized in Reinforcement Learning . Humans can be asked to label data or provide distribution (Active Learning). Interactive Visual Analytics can be used to help humans understand machine learning results and then adjust models during or after training.\"
\n\"The authors can regulate the intermediate results or network layers to produce models more understandable and controllable by humans.\"
\n\"Designing and implementing machine learning algorithms is an iterative process. This requires humans to analyze the models and knowledge integration to take advantage of human understanding of the real world. That review may help current and prospective users of machine learning to understand these fields and inspire them to build more efficient models.\"
\nhttps://www.sciencedirect.com/science/article/pii/S2589004220308488
", "rawMarkdown": "Wow, there is so much to learn in this Competition. Forget any prize. learning is the focus.\n\nOne of the Fine-grained categorization (Human-in-the-loop) is \"Embedding human experts’ knowledge into computational models\" Let's see what we can get Googling it:\n\n#Embedding Human Knowledge into Deep Neural Network via Attention Map\n\nAuthors: Masahiro Mitsuhara, Hiroshi Fukui, Yusuke Sakashita, Takanori Ogata, Tsubasa Hirakawa, Takayoshi Yamashita, Hironobu Fujiyoshi\n\nCitation: \tarXiv:1905.03540 [cs.CV]\n\n\"In this work, the authors aimed to realize a method for embedding human knowledge into deep neural networks. While the conventional method to embed human knowledge has been applied for non-deep machine learning, it is challenging to apply it for deep learning models due to the enormous number of model parameters.\"\n\n\"In this paper, they proposed a fine-tuning method that utilizes a single-channel attention map which is manually edited by a human expert. Their fine-tuning method can train a network so that the output attention map corresponds to the edited ones. As a result, the fine-tuned network can output an attention map that takes into account human knowledge.\"\n\n\"Experimental results with ImageNet, CUB-200-2010, and IDRiD demonstrated that it is possible to obtain a clear attention map for a visual explanation and improve the classification performance. Their findings can be a novel framework for optimizing networks through human intuitive editing via a visual interface and suggest new possibilities for human-machine cooperation in addition to the improvement of visual explanations.\"\n\nhttps://arxiv.org/abs/1905.03540\n\n#Integrating Machine Learning with Human Knowledge (Leonardo da Vinci would loved that)\n\nAuthors: Changyu Deng ; Xunbi Ji; Colton Rainey; Jianyu Zhang ; Wei Lu\n\nhttps://doi.org/10.1016/j.isci.2020.101656\n\nHIGHLIGHTS\n\nIntegrating knowledge into machine learning delivers superior performance\n\nKnowledge is categorized and its representations are presented\n\nVarious methods to bridge human knowledge and machine learning are shown\n\nSuggestions on approaches and perspectives on future research directions are provided\n\n\"Multitask Learning\"\n\n\"Humans do not learn individual tasks in a linear sequence, but they learn several tasks simultaneously. This efficient behavior is replicated in machine learning with multitask learning (MTL). MTL shares knowledge between tasks so they are all learned simultaneously with higher overall performance .\"\n\nRead all these in that paper ( Model Assumptions, Preliminaries of Probabilistic ML, Deterministic Assumptions, Network Architecture, Sy,mmetry of Convolutional Neural Networks, Design of Neuron Connections, Data Augmentation, Simulation, Reinforcement Learning,Active Learning, Interactive Visual Analytics, Parameter Initialization, Transfer Learning.\n\n\"As humans create faster and more accurate knowledge-based models to simulate the world, using simulations to acquire a large amount of data becomes an increasingly efficient method for machine learning. The primary advantage of simulations is the ability to gather a large amount of data when experimentally would be costly, time consuming, or even dangerous\". \n\n\"If using neural networks, tailor the architecture to be suitable for the tasks. If possible, incorporate some known properties, such as Symmetry of Convolutional Neural Networks. Logic, equations, and temporal nature can be, respectively, reflected in the structure of networks by, for instance, combining with symbolic AI, designing special layers/architectures, and using RNNs.\"\n\n\"Design algorithms to include humans in the loop. The interaction between machine and environment can be modeled and optimized in Reinforcement Learning . Humans can be asked to label data or provide distribution (Active Learning). Interactive Visual Analytics can be used to help humans understand machine learning results and then adjust models during or after training.\"\n\n\n\"The authors can regulate the intermediate results or network layers to produce models more understandable and controllable by humans.\"\n\n\"Designing and implementing machine learning algorithms is an iterative process. This requires humans to analyze the models and knowledge integration to take advantage of human understanding of the real world. That review may help current and prospective users of machine learning to understand these fields and inspire them to build more efficient models.\"\n\nhttps://www.sciencedirect.com/science/article/pii/S2589004220308488", "votes": 3 } ], "comments": [], "raw_markdown_by_id": { "1694999": "Wow, there is so much to learn in this Competition. Forget any prize. learning is the focus.\n\nOne of the Fine-grained categorization (Human-in-the-loop) is \"Embedding human experts’ knowledge into computational models\" Let's see what we can get Googling it:\n\n#Embedding Human Knowledge into Deep Neural Network via Attention Map\n\nAuthors: Masahiro Mitsuhara, Hiroshi Fukui, Yusuke Sakashita, Takanori Ogata, Tsubasa Hirakawa, Takayoshi Yamashita, Hironobu Fujiyoshi\n\nCitation: \tarXiv:1905.03540 [cs.CV]\n\n\"In this work, the authors aimed to realize a method for embedding human knowledge into deep neural networks. While the conventional method to embed human knowledge has been applied for non-deep machine learning, it is challenging to apply it for deep learning models due to the enormous number of model parameters.\"\n\n\"In this paper, they proposed a fine-tuning method that utilizes a single-channel attention map which is manually edited by a human expert. Their fine-tuning method can train a network so that the output attention map corresponds to the edited ones. As a result, the fine-tuned network can output an attention map that takes into account human knowledge.\"\n\n\"Experimental results with ImageNet, CUB-200-2010, and IDRiD demonstrated that it is possible to obtain a clear attention map for a visual explanation and improve the classification performance. Their findings can be a novel framework for optimizing networks through human intuitive editing via a visual interface and suggest new possibilities for human-machine cooperation in addition to the improvement of visual explanations.\"\n\nhttps://arxiv.org/abs/1905.03540\n\n#Integrating Machine Learning with Human Knowledge (Leonardo da Vinci would loved that)\n\nAuthors: Changyu Deng ; Xunbi Ji; Colton Rainey; Jianyu Zhang ; Wei Lu\n\nhttps://doi.org/10.1016/j.isci.2020.101656\n\nHIGHLIGHTS\n\nIntegrating knowledge into machine learning delivers superior performance\n\nKnowledge is categorized and its representations are presented\n\nVarious methods to bridge human knowledge and machine learning are shown\n\nSuggestions on approaches and perspectives on future research directions are provided\n\n\"Multitask Learning\"\n\n\"Humans do not learn individual tasks in a linear sequence, but they learn several tasks simultaneously. This efficient behavior is replicated in machine learning with multitask learning (MTL). MTL shares knowledge between tasks so they are all learned simultaneously with higher overall performance .\"\n\nRead all these in that paper ( Model Assumptions, Preliminaries of Probabilistic ML, Deterministic Assumptions, Network Architecture, Sy,mmetry of Convolutional Neural Networks, Design of Neuron Connections, Data Augmentation, Simulation, Reinforcement Learning,Active Learning, Interactive Visual Analytics, Parameter Initialization, Transfer Learning.\n\n\"As humans create faster and more accurate knowledge-based models to simulate the world, using simulations to acquire a large amount of data becomes an increasingly efficient method for machine learning. The primary advantage of simulations is the ability to gather a large amount of data when experimentally would be costly, time consuming, or even dangerous\". \n\n\"If using neural networks, tailor the architecture to be suitable for the tasks. If possible, incorporate some known properties, such as Symmetry of Convolutional Neural Networks. Logic, equations, and temporal nature can be, respectively, reflected in the structure of networks by, for instance, combining with symbolic AI, designing special layers/architectures, and using RNNs.\"\n\n\"Design algorithms to include humans in the loop. The interaction between machine and environment can be modeled and optimized in Reinforcement Learning . Humans can be asked to label data or provide distribution (Active Learning). Interactive Visual Analytics can be used to help humans understand machine learning results and then adjust models during or after training.\"\n\n\n\"The authors can regulate the intermediate results or network layers to produce models more understandable and controllable by humans.\"\n\n\"Designing and implementing machine learning algorithms is an iterative process. This requires humans to analyze the models and knowledge integration to take advantage of human understanding of the real world. That review may help current and prospective users of machine learning to understand these fields and inspire them to build more efficient models.\"\n\nhttps://www.sciencedirect.com/science/article/pii/S2589004220308488" } }