I've seen that a lot of people in the notebook section proposed a CNN model, my first though for this competition was to use a recurrent NN. What are the advantages of the CNN for this competition?
", "messages": [ { "id": "1029945", "postDate": "09/28/2020 09:40:32", "content": "I've seen that a lot of people in the notebook section proposed a CNN model, my first though for this competition was to use a recurrent NN. What are the advantages of the CNN for this competition?
", "rawMarkdown": "I've seen that a lot of people in the notebook section proposed a CNN model, my first though for this competition was to use a recurrent NN. What are the advantages of the CNN for this competition?", "votes": null }, { "id": "1030609", "postDate": "09/28/2020 19:10:53", "content": "You can use them both together (like ConvLSTM2D)
", "rawMarkdown": "You can use them both together (like ConvLSTM2D)", "votes": null }, { "id": "1031763", "postDate": "09/29/2020 17:10:38", "content": "I think the approach when trying to use CNN is that the game of life algorithm modifies the value of a cell based on neighboring cells using a logic that involves a 3x3 convolution, so a CNN is a good idea to solve the problem.
\nAn RNN is also a good idea considering that the model varies with the number of iterations.
\nI am not sure if there is an advantage in one or the other option and as ML_Expert_Silicon_Valley says it is possible to use both.
", "rawMarkdown": "I think the approach when trying to use CNN is that the game of life algorithm modifies the value of a cell based on neighboring cells using a logic that involves a 3x3 convolution, so a CNN is a good idea to solve the problem.\n\nAn RNN is also a good idea considering that the model varies with the number of iterations.\n\nI am not sure if there is an advantage in one or the other option and as [ML_Expert_Silicon_Valley ](https://www.kaggle.com/eladwar)says it is possible to use both.", "votes": null }, { "id": "1031786", "postDate": "09/29/2020 17:28:22", "content": "I think it could have some use, but it would certainly work better with more batches/transitions between growth cycles. I have more interest in the use of graph networks to solve this challenge.
", "rawMarkdown": "I think it could have some use, but it would certainly work better with more batches/transitions between growth cycles. I have more interest in the use of graph networks to solve this challenge.", "votes": null }, { "id": "1031863", "postDate": "09/29/2020 18:32:00", "content": "I have not experimented with CNN for this challenge, I am focusing on the genetic algorithms.
\nIt would be interesting to implement a GNN to solve this problem.
", "rawMarkdown": "I have not experimented with CNN for this challenge, I am focusing on the genetic algorithms.\n\nIt would be interesting to implement a GNN to solve this problem.", "votes": null } ], "comments": [ { "id": 1031763, "author_name": "desareca", "author_url": "", "post_date": "09/29/2020 17:10:38", "content": "I think the approach when trying to use CNN is that the game of life algorithm modifies the value of a cell based on neighboring cells using a logic that involves a 3x3 convolution, so a CNN is a good idea to solve the problem.
\nAn RNN is also a good idea considering that the model varies with the number of iterations.
\nI am not sure if there is an advantage in one or the other option and as ML_Expert_Silicon_Valley says it is possible to use both.
", "votes": null, "replies": [ { "id": 1031786, "author_name": "eladwar", "author_url": "", "post_date": "09/29/2020 17:28:22", "content": "I think it could have some use, but it would certainly work better with more batches/transitions between growth cycles. I have more interest in the use of graph networks to solve this challenge.
", "votes": null, "replies": [] }, { "id": 1031863, "author_name": "desareca", "author_url": "", "post_date": "09/29/2020 18:32:00", "content": "I have not experimented with CNN for this challenge, I am focusing on the genetic algorithms.
\nIt would be interesting to implement a GNN to solve this problem.
", "votes": null, "replies": [] } ] }, { "id": 1030609, "author_name": "eladwar", "author_url": "", "post_date": "09/28/2020 19:10:53", "content": "You can use them both together (like ConvLSTM2D)
", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1029945": "I've seen that a lot of people in the notebook section proposed a CNN model, my first though for this competition was to use a recurrent NN. What are the advantages of the CNN for this competition?", "1030609": "You can use them both together (like ConvLSTM2D)", "1031763": "I think the approach when trying to use CNN is that the game of life algorithm modifies the value of a cell based on neighboring cells using a logic that involves a 3x3 convolution, so a CNN is a good idea to solve the problem.\n\nAn RNN is also a good idea considering that the model varies with the number of iterations.\n\nI am not sure if there is an advantage in one or the other option and as [ML_Expert_Silicon_Valley ](https://www.kaggle.com/eladwar)says it is possible to use both.", "1031786": "I think it could have some use, but it would certainly work better with more batches/transitions between growth cycles. I have more interest in the use of graph networks to solve this challenge.", "1031863": "I have not experimented with CNN for this challenge, I am focusing on the genetic algorithms.\n\nIt would be interesting to implement a GNN to solve this problem." }, "source": "meta" }