Aaron M. Hosford

Machine Learning Researcher and Engineer

Bio

I'm a machine learning researcher and entrepreneur with over 30 years of focused, self-directed learning, design, and experimentation with novel machine learning algorithms. I was interested in algorithms before I even knew what one was, inventing a maze generation algorithm in 3rd grade, which allowed me to generate mazes with a guaranteed single solution without knowing what the solution was, so I could still enjoy solving them. I got my start with programming at the age of 12, when I discovered that my parents' 286 came with a programming manual for BASIC, and promptly read it front-to-back, trying out all the example code.

In my mid teens, I found a book on neural networks and another on formal symbolic logic, and my undying interest in machine learning and AI was born. (To be clear, "AI" does not mean ChatGPT and other large language models. The field is so much broader than that.) I began experimenting with all sorts of ML algorithms and designing my own. This has continued to be my "special interest" - like a hobby, but we autistics do it way more intensely - for the entirety of my adult life. As a result, I developed a tremendously broad and deep understanding of ML algorithms. I tend to invent a new ML algorithm about once a month, on average. Many of them don't work, but many of them do, often better than existing algorithms, which led me to co-found a deep-tech startup in AI/ML. We are currently Transparent AI, but we are in the process of rebranding, as we have expanded in scope from novel interpretable ML algorithms to cognitive architectures as well.

For the last ~20 years, I've also worked various computing-related day jobs, in addition to the self-driven projects, mostly in heavily regulated areas like banking and telco. I have helped to build a major NLP application serving tens of millions of customers, where I was a contributing author for a patent in natural language processing. I've also architected and implemented multiple predictive analytics pipelines, one of which was a demand forecasting system projected to save the company millions of dollars. I found I really enjoy working with a team of engineers to solve hard problems in creative ways. There's an energy that fills the room when you're plotting out how to meet the unique challenges that stakeholders need to be addressed.

Select Open-Source Projects

XCS

XCS (Accuracy-based Classifier System) is an algorithm invented by Stewart W. Wilson, Ph.D. XCS is a type of Learning Classifier System (LCS), a machine learning algorithm that utilizes a genetic algorithm acting on a rule-based system, to solve a reinforcement learning problem. You can check out my open-sourced implementation of the algorithm here.

Pyramids Semantic Parser

The Pyramids Semantic Parser is a pure-Python parser I built for the purpose of extracting semantic information from natural language. I use it in my NLU system, NPC. It is a work in progress, but can already handle a broad cross section of natural language inputs.

Less Naive Bayes

Less Naive Bayes is a machine learning classification algorithm related to Naive Bayes, but capable of handling input spaces that are not linearly separable. It works by successively training new sub-classifiers to predict both the target classification and the mis-classifications of earlier classifiers. The feature space is successively subdivided into linearly separable subspaces which the sub-classifiers are better capable of distinguishing. Eventually the feature space is transformed sufficiently that the most recent layer is fully capable of correctly classifying all samples with distinguishable features.