We currently offer these services for text and images:

1. Deep vector embeddings
   Gixante uses several deep representation models, as well as open source models like Word2Vec, GloVe and fastText. They are continuously trained on curated open data to map any text into arrays of numerical features. The combination of different approaches extracts more information, while dimensionality reduction tailored to your data removes redundancy.
2. Entity recognition
   Using annotated corpora and news articles, Gixante updates a vast library of entities. These are extracted from your data and filtered automatically to control noise and tractability.
3. Query understanding (text only)
   A few words tell us so much - Gixante’s module for query understanding translates them for your models, depending on the context and on you data.
4. Automatic data labelling
   Gixante can label ‘dark’ data of any size, starting from a brief, a set of rules, a set of entities or a set of examples, even if vague. We use other modules to transfer pre-existing knowledge to your data (distant supervision) and a generative model to fill in the gaps. The latter is semi-supervised by a human to ensure correct stratification while training and thresholded to the maximum accuracy. You can even be the human-in-the-loop if you wish.
5. Supervised learners
   Gixante can naturally close the loop with classifers and regressors. In conjunction with other services, this is particularly powerful because it allows to create human interfaces that refine from usage and can query any type of database.