{ "id": 238108, "title": "Previous Kaggle Competitions for inspirations", "url": "/competitions/seti-breakthrough-listen/discussion/238108", "author_name": "Athar Sayed", "post_date": "2021-05-11T08:41:30.793000", "votes": 21, "comment_count": 0, "views": 0, "content": "

Looks really interesting competition and it feels awesome to see Kaggle being applied to SciFi stuff , anyways , even if the problem is looking completely new , it can be reduced to detecting anamolies in Spectograms !
\nThis is the idea adopted by most of Audio Detection Competitions ! It would be interesting to try out some of the ideas from those competitions here , Kaggle has lot of Audio Competitions launched previously there are great resource sharing and Solutions sharing at the end which can be used for building and improving upon baselines :) .

\n

For building robust solution my initial thoughts are that we need to take inspirations from both Audio Competitions and competitions based on Signal processing as pointed out in description.

\n

1. Previous Audio Competition

\n
    \n

  1. Rainforest Connection Species ,

    2. \n

  2. Freesound Audio Tagging

    3. \n

  3. Cornell Birdcall Recognition

    4. \n
\n

Highlights : Many people used pretrained CNN and Finetuned them on the Audio Classification task , along with that some winning solutions used SED based models

\n

2. Signal Processing Competitions

\n
    \n

  1. VSB Power Line Fault Detection : This data involves signals with lot of noise and task is kind of anomaly detection from labelled data and hence the winning solutions tips and tricks may be useful.

    2. \n

  2. LANL Earthquake Prediction , This competition was a regression based competition in which we had to predict time before an earthquake hits , interestingly this competition also involves synthetic data

    3. \n
\n

3 . Camera Model Identification competition , a relatively old competition still there can be lot of learnings from winning solutions as well .

\n

It is really interesting and Fun problem to solve !
\nHappy Kaggling !

", "messages": [ { "id": 1301782, "postDate": "2021-05-11T08:41:30.793Z", "content": "

Looks really interesting competition and it feels awesome to see Kaggle being applied to SciFi stuff , anyways , even if the problem is looking completely new , it can be reduced to detecting anamolies in Spectograms !
\nThis is the idea adopted by most of Audio Detection Competitions ! It would be interesting to try out some of the ideas from those competitions here , Kaggle has lot of Audio Competitions launched previously there are great resource sharing and Solutions sharing at the end which can be used for building and improving upon baselines :) .

\n

For building robust solution my initial thoughts are that we need to take inspirations from both Audio Competitions and competitions based on Signal processing as pointed out in description.

\n

1. Previous Audio Competition

\n
    \n

  1. Rainforest Connection Species ,

    2. \n

  2. Freesound Audio Tagging

    3. \n

  3. Cornell Birdcall Recognition

    4. \n
\n

Highlights : Many people used pretrained CNN and Finetuned them on the Audio Classification task , along with that some winning solutions used SED based models

\n

2. Signal Processing Competitions

\n
    \n

  1. VSB Power Line Fault Detection : This data involves signals with lot of noise and task is kind of anomaly detection from labelled data and hence the winning solutions tips and tricks may be useful.

    2. \n

  2. LANL Earthquake Prediction , This competition was a regression based competition in which we had to predict time before an earthquake hits , interestingly this competition also involves synthetic data

    3. \n
\n

3 . Camera Model Identification competition , a relatively old competition still there can be lot of learnings from winning solutions as well .

\n

It is really interesting and Fun problem to solve !
\nHappy Kaggling !

", "rawMarkdown": "Looks really interesting competition and it feels awesome to see Kaggle being applied to SciFi stuff , anyways , even if the problem is looking completely new , it can be reduced to detecting anamolies in Spectograms ! \nThis is the idea adopted by most of Audio Detection Competitions ! It would be interesting to try out some of the ideas from those competitions here , Kaggle has lot of Audio Competitions launched previously there are great resource sharing and Solutions sharing at the end which can be used for building and improving upon baselines :) .\n\nFor building robust solution my initial thoughts are that we need to take inspirations from both **Audio Competitions** and competitions based on **Signal processing** as pointed out in description.\n\n**1. Previous Audio Competition**\n1. [Rainforest Connection Species ](https://www.kaggle.com/c/rfcx-species-audio-detection) ,\n\n2. [Freesound Audio Tagging](https://www.kaggle.com/c/freesound-audio-tagging-2019)\n\n3. [Cornell Birdcall Recognition](https://www.kaggle.com/c/birdsong-recognition)\n\n**Highlights : Many people used pretrained CNN and Finetuned them on the Audio Classification task , along with that some winning solutions used [SED](https://paperswithcode.com/task/sound-event-detection) based models**\n\n**2. Signal Processing Competitions**\n1. [VSB Power Line Fault Detection ](https://www.kaggle.com/c/vsb-power-line-fault-detection) : This data involves signals with lot of noise and task is kind of anomaly detection from labelled data and hence the winning solutions tips and tricks may be useful.\n\n2. [LANL Earthquake Prediction](https://www.kaggle.com/c/LANL-Earthquake-Prediction) , This competition was a regression based competition in which we had to predict time before an earthquake hits , **interestingly this competition also involves synthetic data**\n\n3 . [Camera Model Identification competition ](https://www.kaggle.com/c/sp-society-camera-model-identification) , a relatively old competition still there can be lot of learnings from winning solutions as well .\n\nIt is really interesting and Fun problem to solve !\nHappy Kaggling !", "votes": 21 } ], "comments": [], "raw_markdown_by_id": { "1301782": "Looks really interesting competition and it feels awesome to see Kaggle being applied to SciFi stuff , anyways , even if the problem is looking completely new , it can be reduced to detecting anamolies in Spectograms ! \nThis is the idea adopted by most of Audio Detection Competitions ! It would be interesting to try out some of the ideas from those competitions here , Kaggle has lot of Audio Competitions launched previously there are great resource sharing and Solutions sharing at the end which can be used for building and improving upon baselines :) .\n\nFor building robust solution my initial thoughts are that we need to take inspirations from both **Audio Competitions** and competitions based on **Signal processing** as pointed out in description.\n\n**1. Previous Audio Competition**\n1. [Rainforest Connection Species ](https://www.kaggle.com/c/rfcx-species-audio-detection) ,\n\n2. [Freesound Audio Tagging](https://www.kaggle.com/c/freesound-audio-tagging-2019)\n\n3. [Cornell Birdcall Recognition](https://www.kaggle.com/c/birdsong-recognition)\n\n**Highlights : Many people used pretrained CNN and Finetuned them on the Audio Classification task , along with that some winning solutions used [SED](https://paperswithcode.com/task/sound-event-detection) based models**\n\n**2. Signal Processing Competitions**\n1. [VSB Power Line Fault Detection ](https://www.kaggle.com/c/vsb-power-line-fault-detection) : This data involves signals with lot of noise and task is kind of anomaly detection from labelled data and hence the winning solutions tips and tricks may be useful.\n\n2. [LANL Earthquake Prediction](https://www.kaggle.com/c/LANL-Earthquake-Prediction) , This competition was a regression based competition in which we had to predict time before an earthquake hits , **interestingly this competition also involves synthetic data**\n\n3 . [Camera Model Identification competition ](https://www.kaggle.com/c/sp-society-camera-model-identification) , a relatively old competition still there can be lot of learnings from winning solutions as well .\n\nIt is really interesting and Fun problem to solve !\nHappy Kaggling !" } }