Management of Training Data for Deep Learning
We are looking for a technology to enrich training data automatically that may let us get better results of deep learning efficiently.
Why this is important
Fujitsu Japan (FJJ) develops natural language processing technology (conversation) and object recognition using deep learning or other machine learning technologies. With these, FJJ is planning to enter the customer service market including call center and service robot.
We need training data of high quality to make AI better. However, in reality, existing data requires some kind of formatting or compensation for use. FJJ seeks for a technology to format and compensate data efficiently.
Description of the needWe are looking for:
- Technology or tool that analyzes, categorizes, and formats content of the text data – (1)
- Technology which triggers training when text data is updated or added –(2).
- Technology that can write the actions (1) and (2) as a procedure to reuse for other different data
- E.g. Recognizing the status of each text files based on their contents: First the AI extracts the words that represent status (complete/in progress/resumed) from a group of text files and then matches each text files with the correct status as attributes.
- Uses new theories such as neural networks.
To be considered, submissions must include
- Contact information for the potential technology partner
- Description of the technology, its stage of development, and its capabilities
- Information on any patents granted or applied for (i.e., information in the public domain)