The Xyggy Discovery Engine is a different kind of beast from conventional machine learning and information retrieval.
The Engine integrates deep learning and Bayesian machine learning with realtime learning. The result is discovery with meaning.
Uses include intentional discovery, personal discovery, autonomous discovery, feeds, recommendation systems and probably a lot more.
Also, query interactively; add or update data at any time; for all data types; and scale.
Conventional machine learning delivers realtime scoring (against a static data model) but not realtime learning (on static or changing data) which is its Achilles heel.
It is like learning to eat spaghetti for the first time by watching someone else do it. But, you can’t put it into practice until the next day after your brain has been updated with the knowledge while sleeping. Imagine if real life was like that.
A real world example demonstrates this - Inside Facebook's Biggest AI Project Ever
“To do this Facebook run tens of trillions queries per day to make about six million predictions per second. Facebook trains the algorithms that power its News Feed within hours, using trillions of data points. The company updates its learning models every 15 minutes to two hours so that it can react quickly to current events.”
Each time a model is re-trained and re-tested, the cost in time and resources accumulate. It also demonstrates the data model is out of sync with the real-world.
The Engine automatically learns as it goes along.
There are no training and re-training cycles.
The world changes.
With conventional AI, when new or update data arrives, the model is re-trained and re-tested. Again, the cost in time and resources accumulate.
The Engine is the analogical equivalent of an in-memory SQL DBMS with equivalent CRUD operations and query statements all available through the API.
Add, update and delete data items individually or in bulk at any time.
Design a UI/UX for a responsive and interactive AI with the API.
Options includes multiple-items-per-query, more-like-these, less-like-these, relevance feedback, serendipity, anomaly detection and active learning.
A query is formed with one or more items and the Engine will find other relevant items in ranked order. A query can be modified mid-flight.
Deep learning vectors can be of any length.
Query results are not based on Euclidean or cosine distance metrics (which do not exhibit any intelligence) but realtime learning (inference) which also means the Engine doesn’t suffer from scalability or accuracy problems.
For all data types including text, images, audio and composite. Structured and unstructured data.
Multi-modal learning is possible.
Scales effortlessly on a single server and loosly-coupled distributed machines.
Personal discovery with privacy. Today we have intentional discovery (“I want this”) and social discovery (“My friends think I want that”) yet personal discovery (“I don’t know what I want but I will when I see it”) remains largely an unsolved problem.
Transform relational databases into intelligent databases.
Job seekers search a corpus of job descriptions with their resume. Conversely, employers search for candidates with a ‘fake’ resume or job description (or both) as the query.
... and many more.
Email Hello Xyggy