** Neural Network Console Challenge ** (NNC-Challenge) is an AI development tool developed by SONY ** Neural Network Console ** that challenges deep learning using data provided by supporting companies. The contest has been held twice so far.
Here, I will leave a list of the 1st and 2nd prize-winning works as a memorandum for reference in future Challenges.
The sponsoring company of 2nd sells BGM and sound effects for video production, events, sound effects, etc. This is Audiostock. The data provided was over 10,000 BGM data and was held from 2020.09.16 to 2020.10.19.
** Grand Prize **: Achieve similar song search The theme is to search for similar songs that I really wanted to find in Audiostock's UX. As a result, we are developing an effective and highly accurate similar song search algorithm.
** Ledge.ai Award **: Automatically select BGM according to the content of the conversation Developed an engine that processes everyday conversations in natural language using BERT, automatically determines the similarity to any of "rock", "pop", and "ballad", and recommends songs.
** NNC Award **: Voice data classification and recommendations The music is quantified on the scale of emotions and emotions, and the songs with high similarity are recommended. The recommendation accuracy was high. It also presents a new search method called image recommendations that match the music.
** Special Encouragement Award 1 **: [Automatically classify musical instruments "piano" and "electric guitar"](https://www.slideshare.net/yoshi1966/neural-network-console-contest-report-2020-by -kitagawa) An algorithm that automatically classifies whether BGM is using a piano / electric guitar. High accuracy is achieved by a learning process of 4 steps in total.
** Special Encouragement Award 2 **: Recommend restaurants from BGM Create an algorithm to classify recommended shops (lunch) from BGM. Furthermore, the algorithm is used to verify what kind of shops are recommended from the voice data that the author often hears.
The sponsoring company of 1st is PIXTA who sells commercial photographs. .. The data provided was over 10,000 image data and was held from 2020.03.04 to 2020.03.31.
** Grand Prize **: Smiley smile classifier with laughing face and smiling face Using the Neural Network Console from the facial expression image of a person, we created a smile classifier that classifies into three types: "smiles that seem to make you laugh", "smiles that smile with a smile", and "other than smiles".
** Ledge.ai Award **: Select keywords from Ogai Mori's novels and classify them Morphological analysis was performed based on the impressions and reviews sent to the main story and the novel from Ogai Mori's short story "Cup", and images were classified by keywords extracted using the Neural Network Console.
** NNC Award **: Classified by the number of people in the image The Neural Network Console Windows version was used to classify the number of people in the image data of various categories, including one, two, and groups (three or more), and then the Cloud version was used to classify the images.
Recommended Posts