This document provides background on my influences that I sometimes use to explain the biographical aspects that have influenced my perspective. Feel free to skip the parts that are not relevant to the article from which this linked.
I've already lived through a generation that experienced large upheaval and migration due to automation. I grew up on a farm. I moved to California because of economics. The 10x production improvements that technology brought to farming made it less economical for farmers having less than a certain critical mass of acreage.
Our family farm was considered to be medium-sized when I was young, but once it came time for me to choose a career, it suddenly seemed small. I was always interested in science and wanted to pursue that interest anyways, so it was a bit of a relief not to feel obligated to carry on the family business. My parents had bought me the World Book Encyclopedia in grade school, and I had read it A through Z, including every one of the Yearbook and Science Year supplements.
There wasn't a slot for me in the regional economy where I was born and raised so I moved west. I came for the surf and somehow ended up with an Erdős Number of 4.
How did I come to machine learning?
I had an interest in parallel processing beginning in college. I was studying dataflow architecture at USC when I encountered neural networks. I immediately knew that this was somehow more appealing than the dataflow architecture, which was fiddly and difficult to tune. Learning from data seemed like a good idea whose time had come. The field was just emerging from what had been called the AI Winter. I had myself already worked as a Lisp programmer in a role called a "knowledge engineer" developing expert systems. Little did I know I would live through another AI Winter, despite AI already being around you every second of the day.
As a member of GURU I met the future Godfather of Deep Learning, the inventor of LSTM, and other true believers of connectionism. I took a seminar on consciousness taught by Francis Crick. I did a two year internship at Hecht Nielsen Neurocomputer, which was enlightened enough to incentivize me with my very first incentive stock options. It IPO’d, allowing me to buy my first house in California. I went on to apply machine learning in myriad data-driven domains, including audio, video, medical, e-commerce, web analytics, consumer electronics, and consumer application.
How did I become interested in social psychology?
I had a keen interest in psychology as a teenager, and wanted to study it in college, but chose the more pragmatic path of electrical engineering. In my first job out of college, I worked as a software engineer, and ended up working in what back then qualified as AI -- knowledge engineering programmed in Lisp using Symbolics lisp machines. Later on, I was enrolled in the cognitive science interdisciplinary program at UCSD -- but was a bit ahead of my time. I wanted to study the underlying models (the neural networks) and not spend much if any time collecting the data. This turned out to be a lucky happenstance anyway, because I was able to dive deep into algorithm development and econometrics. At the time, connectionists and most other neural net freaks were closeted psychologists. However, it wasn't until my experience doing big data analytics on large datasets obtained from the usage of hundreds of millions of users that I realized that I was more interested in social psychology. People are individuals -- but in the aggregate, usage data turns out to be very predictive of how groups behave. I've known this for over a decade after building machine learning models that successfully predicted user demographics and psychographics.
According to Christos Papadimitriou (from whom I had the pleasure of taking the algorithms course at UCSD):
The Internet changed Computer Science (and Theoretical Computer Science) it turned it into a physical science… …and a social science. -- Algorithmic Game Theory bootcamp.
How did I come to economics?
I had an entrepreneurial interest early in my career, dating back to college where I took an introductory economics course. That consisted of the usual introduction to the only two types of economics that were taught: macro and micro. I surmised that this topic was not relevant to the path that I planned to follow and summarily dismissed it. Later on, I received a thorough introduction to econometrics by the author of the most-cited paper in economics since 1970, who had agreed to be my dissertation advisor. He also had a strong interest in neural networks.
I believe in using principles as a guiding light to illuminate career path and help with difficult choices. One of my bedrock principles was: "be data-driven". HNC Software was one of the first organizations to popularize the term "data-driven" in the 1990's, and I saw the light -- I became an ardent proponent data-driven methodology a decade before the term "data science" became popularized.
Over time, the phrase "data-driven" began to pick up multiple connotations. One was the notion of "data-driven evidence." However, others emerged, such as the notion of data-driven storytelling to explain an approach. Another is monetization -- data as a proxy for a user's identity that can be sold to parties that want to promote their brand, goods or services to that user. I eventually realized that the type of data-driven methodology that described my core values was about incentivizing humans to become better at what they do less, and less about storytelling, post-hoc justification, and monetization. A better principle emerged, namely: "incentivization." This is when I discovered my inner economist.
My dissertation research centered around creating algorithms for active learning, where a machine learner plays an active role in selecting its training examples rather than being a passive learner. The active learning algorithm I had created, which essentially does gradient descent in example space, was motivated by my own parallel research into econometrics. This algorithm is able to distill the training set for (what at the time was a challenging semi-chaotic time series known as) the Mackey-Glass time series to an economical 25 exemplars.
What I'm thinking about now
A large shift is happening in machine learning. While developments in deep learning have justifiably captured the popular imagination, a different area of machine learning is experiencing a different resurgence, one that promises to be as impactful as deep learning. Economics, econometrics, machine learning and algorithmic computational science are dovetailing into an emerging field called algorithmic game theory.
- 9,330,170 : Relating objects in different mediums
- 8,050,949 : Method and apparatus for an itinerary planner
- 8,050,948 : Method and apparatus for an itinerary planner
- 8,050,935 : Dynamic web service composition to service a user request
- 7,961,189 : Displaying artists related to an artist of interest
- 7,895,065 : Method and apparatus for an itinerary planner
- 7,844,557 : Method and system for order invariant clustering of categorical data
- 7,840,568 : Sorting media objects by similarity
- 7,774,288 : Clustering and classification of multimedia data
- 7,761,394 : Augmented dataset representation using a taxonomy which accounts for similarity and dissimilarity between each record in the dataset and a user's similarity-biased intuition
- 7,750,909 : Ordering artists by overall degree of influence
- 7,664,718 : Method and system for seed based clustering of categorical data using hierarchies
- 7,239,962 : Method and apparatus for a routing agent
- 6,473,851 : System for combining plurality of input control policies to provide a compositional output control policy
- 6,338,066 : Surfaid predictor: web-based system for predicting surfer behavior
- 6,278,966 : Method and system for emulating web site traffic to identify web site usage patterns
- 6,230,153 : Association rule ranker for web site emulation
3 Patents Pending