Gamalon’s Bayesian Program Synthesis Revolutionizes Machine Learning

Gamalon Inc. is a technology startup in Cambridge, Massachusetts led by CEO Ben Vigoda. Gamalon Inc. is currently developing technology which improves the machine learning of AI. This technology is referred to as Bayesian Program Synthesis (BPS) and has a huge advantage over typical Deep Learning techniques.

Deep Learning typically involves using millions of data points for the artificial intelligence to recognize specific objects. This approach has a few problems accordingly to Gamalon. In addition to being time-consuming to gather millions of data points, Deep Learning can get confused by additional related objects.

Gamalon uses Google’s “Quick, Draw!” app as an example of these challenges. The team at Gamalon demonstrate Deep Learning’s shortcomings by drawing a floor lamp and a chair to go with it, but Google’s Deep Learning AI assumes that they drew something completely different from the original idea of the “floor lamp” and therefore they should try again. Gamalon demonstrated BPS using the same example and the software was able to identify the floor lamp and the chair as 2 separate objects. BPS understands the drawn art regardless of different sizes, positions or even different styles.

Gamalon’s Bayesian Program Synthesis (BPS) uses human-readable code to learn and relearn based on the previous information provided. In the example regarding the lamp and the chair, Gamalon started teaching BPS about lines, and then rectangles (made up of 4 lines) then progressed to a seat (which is a combination of rectangles). At each step, BPS can be “taught” new concepts by any number of people and their drawings with a simple button click. Because of the human-readable code, Gamalon has the option of removing any taught concept with a simple click of a button. This is important to filter and the information that BPS learns without a complete codebase rewrite. The huge benefit of Gamalon’s Bayesian Program Synthesis (BPS) is the reduced costs. Unlike normal Deep Learning, BPS can run off of the power of a single tablet and learn from a few examples, compared to hundreds of servers and millions of data points.

Gamalon is currently testing commercial opportunities for BPS. They have created Gamalon Structure and Gamalon Match which have saved some clients a large amount of time and money. Gamalon Structure converts paragraphs of text to clean, data rows for use in business and Gamalon Match remove duplicates and/or link certain rows of data. A current client of Gamalon used to spend four million dollars and 9 months to sort through their data with a large team of people every year. Now BPS accomplishes that same task in a few minutes with over twice the accuracy.

Learn More:

Be the first to comment

Leave a Reply