Deep Learning

Keywords: Feature Generation, Classification, Neural Network, Autoencoder, Fingerprint

At Fact AI, we love deep neural networks (DNNs) and use them both for feature generation and classification. During feature generation, our state-of-the-art DNNs create rich and compact fingerprints from products or services. Those fingerprints can be seen as the DNA of a product and are later used for further machine learning applications such as classification or segmentation. One classification application might be the matching of product fingerprints with customer needs. Our WCFM project shows our neural networks in action.


Anomaly Detection

Keywords: One-class classification, Neural Network, Autoencoder


Anomaly or outlier detection is a method used to identify unusual patterns that do not conform to expected behavior. Such unusual patterns usually correspond to suspicious items or events and are typically related to some kind of problem, which could, for example, be an error state of a production machine in industry. A machine learning model for anomaly detection, typically analyzes a huge amount of faultless data extracting the essential characteristics of it. During production, the model will detect suspicious behavior by the anomalous patterns. What anomalous means in this context, learns the model by itself during training, therefore, no rules have to be predefined.

Natural Language Processing

Keywords: Information Retrieval, Sentiment Analysis, Topic Detection, Recurrent Neural Network (RNN)

Companies are often faced with a huge amount of unstructured, textual data produced by various stakeholders when in need of quantitative, business-related information. Together with the Salzburg University of Applied Sciences, we created a product intelligence tool combining state-of-the-art technologies into a natural language processing (NLP) engine capable of collecting and storing product-related online data, extracting metadata and analyzing aspect-based sentiments. [2] Our expertise in the field of NLP, gives us the opportunity to include textual data at anytime during feature generation, making sure that valuable information in an unstructured format such as continuous text is not lost.