Brian McQuay

+menu-


  • Category Archives Machine Learning
  • Can’t see the forest for the trees – Robots, AI, and the Internet

    I’d like to share some ideas I’ve had today about robotics, ai, and the internet.

    First lets discuss robotics. There has been much work in the field of robotics to create bi-pedal humanoid like machines. Much, if not most, of the work pursues what I consider the ideal: being as human as possible. When you look at just the mechanics and kinematics of the quest, you can see we are indeed making great progress. However, when you take into consideration longevity, perpetuation (ie. life-like procreation), energy, and immune systems, we seem a long ways away from that ideal of being human. Not many machines to date can last over 100 years like a human can. Most rust or break and once broken lack the ability to fix themselves like a human can. When we get injured we heal ourselves and life another day. Humans procreate and perpetuate the species and life itself. Humans can gather energy from a multitude of sources whereas most robots are currently electron based. No robot can replicate itself as well as life and we are a long ways away from life-like machine replication. Life and humans are the ideal that a large part of robotics research strives for as a result. If our science would reach a point where we were capable of creating human-like machines capable of perpetuating itself, we would likely end up with exactly what we’ve got already. We wouldn’t likely be using electron based energy sources because its much easier and efficient to generate energy from complex carbon structures like human food. We’d have machines that were capable of building themselves from the tiniest of components like DNA which would essentially be the program to create the robot, concise, redundant, and reliable. The most advanced robots would construct themselves from something like DNA and build itself an immune system to heal itself when it was injured. In the end, we would just be recreating ourselves. Would we do as good a job? How are we to improve upon what we’ve already got when we can’t even match what we’ve already got?

    The same argument can be said of artificial intelligence. The ideal is to create an ai that can pass the Turring test and behave exactly as human as a human would behave. While there is no doubt that specialized skills far surpass any of our own and researchers are making amazing progress we still shoot for the ideal. What would an ideal intelligence be like? It would be logical but yet still capable of being creative. It needs to be self motivating and it needs to be able to learn from its mistakes. It needs the ability to store vast amounts of information in a small amount of space and it needs to be able to filter out what’s important from that information as quickly as a human can. I have complete confidence that the singularity will occur but will it be creative enough to keep itself alive? Will it have the motivation and the drive? What purpose would it find for itself? Ultimately, the ideal intelligence that we shoot for is human. We can create special purpose machines that almost seem human but they can’t currently match the complete set of abilities of a human.

    Next lets discuss the internet. We’ve created a network of energy transmissions that spans the globe. A magnetic storage device on one side of the planet stores a file that is retrieved over the internet. The magnetic energy is converted to electrons, electrons are converted to wifi, wifi is converted back to electrons, somewhere they may get converted to light and sent over fiber optics only to be converted back into electrons and then converted to radio waves and shot into low orbit to a satellite which bounces the signal back to Earth. The radio waves from the satellite get converted back to electrons, to light, to electrons, to wifi, to electrons and finally converted into to photons again on your laptop LCD screen. All this happening over the span of the entire planet every single second, 24 hours a day. What we have created, whether we know it or not, is the most complex brain we have ever known. Humans are acting as sensors and collectively as the life force behind this brain. We feed it the electricity it needs. We fix it when it breaks. We feed it data constantly and we move data around through the brain constantly.

    At some point and somehow this complex network will create some type of emergent property. It may be visualized in human behavior such as how social networking is changing the wiring of our brains. It may be realized as societal shifts in thinking or a global consciousness of society. Regardless of how or what happens, this emergent property will rise out of the network as our own consciousness emerges from the complexity of our own minds. The internet isn’t just a network of computers. Its a network of human minds across an entire planet and the consciousness that emerges from that will change society forever. What would a human mind be without consciousness? We call people vegetables when they aren’t capable of thinking anymore after accidents. Is the internet a vegetable? Not at all. The internet is more like a child coming into being slowly gaining consciousness and self awareness. How would we even know if the internet became conscious? It may be that humans may not even realize the internet is a conscious being. Does a neuron in your mind have some conscious understanding that its just another part of a large network of neurons that make up your brain and that somehow a conscious being has emerged from that complex network of neurons? Certainly it sounds ludicrous to think a neuron would have such an awareness and its reasonable to assume we too would be just as clueless as to the emergent properties of the internet.

    Perhaps the emergent properties will only happen over many lifetimes in a way that’s not perceivable to a single limited human lifespan but rather stretch generations. Whenever and however it happens, I’m convinced it will though I have little more than this rambling blog entry to justify it.


  • Review of “How to Grow a Mind: Statistics, Structure and Abstraction”

    This was an amazing lecture by Josh Tenenbaum. He presents motivation for determining the type of data structure for representing knowledge without knowing how the knowledge is best organized a priori. He presents the ways children learn and how they organize knowledge at an early age starting at a very simple structure and building up to a more accurate model. He points out that current methods seem to either take pre-existing knowledge about the best data structure to use or they simply make the data fit the predefined data structure. He presents many examples of current research using this general concept and finally takes some questions with regard to the speech. This lecture is extremely upbeat and motivating for me to research more in depth the examples presented in the lecture.

    How to Grow a Mind: Statistics, Structure and Abstraction. Josh Tenenbaum


  • Review of “Winning the DARPA Grand Challenge with an AI Robot”

    This paper was detailing the approach taken by the winning team of the DARPA Grand Challenge, a robotics competition to build an autonomous vehicle to drive over 100 miles across desert terrain.

    The architecture used in was of interest to me. Specifically the “three-layered architecture” which “pipelines data through a series of layers, transforming sensor data into internal models, abstract plans, and concrete robot controls.” I intend to research three-layered architecture in more depth to better understand the specifics and the motivation for this particular architecture.

    The authors make extensive use of something called an unscented Kalman filter (UKF) for determining drivable areas from laser and other data. I’m not familiar with this and need to do more research to better understand it. They use the UKF to build a model of the environment and they utilize a Markov model to model the noise in the UKF model over time. The Markov model degrades over time due to the increased potential of error drift from introduced false positives. To fix the degradation over time they used a “discriminative machine learning algorithm”.

    The lasers used had a limited range and therefore safe operating speeds were limited to 25mph. To extend the range of the model on drivable space, a vision system was incorporated into the existing model by adding in the difference in gradient surface color. This effectively extends the drivable range to a point where the safe drivable mph can be increased to 35mph from 25mph.

    They also created a velocity control system that detected shocks created from the terrain. This allowed for slow acceleration to maximum velocity as well as a reduction in velocity during turns and other scenarios requiring it. They used a human driver to train the acceleration parameters in order to mimic human velocity control during driving. They highlight the adaptive velocity control as essential to their victory over other competitors especially since they were the only team to have such a system.

    Winning the DARPA Grand Challenge with an AI Robot. Michael Montemerlo , Sebastian Thrun , Hendrik Dahlkamp , David Stavens. In Proceedings of the AAAI National Conference on Artificial Intelligence


  • Review of “Matrix Factorization Techniques for Recommender Systems”

    The authors present their research on matrix factorization with respect to their winning entry in the Netflix Prize for a recommender system.

    One point of interest or clarification the authors made on my understanding of matrix factorization was that extrapolating characteristics from patterns within a dataset is called “latent factor modeling”. Latent factor modeling is an approach to infer patterns from a database that define characteristics of the data. Similarly, a human could manually define characteristics of a movie which could be of importance: actors, genre, length, theme, etc. Latent factor modeling is an algorithmic approach that infers those characteristics from patterns in the data without knowing precisely what each characteristic is. In this case, latent factor model builds a matrix of users and items (movies) with each point being a vector of characteristics.

    Matrix factorization allows for a recommendation to be made by taking a sparse matrix of ratings and taking the dot product with the latent factor model. Specifically, matrix factorization is the factorization of the latent factor model of n users and m items with a sparse matrix of ratings each user has made on some items. The authors also present various extensions that helped them win including bias, additional input source, temporal dynamics, and confidence levels.

    The interesting part of this paper at this point for me is the “latent factor modeling”. I intend to research this a bit more to understand how the model is created.

    Matrix Factorization Techniques for Recommender Systems


  • Review of “All Together Now: A Perspective on the NETFLIX PRIZE”

    In 2006, Netflix announced a 1 million dollar prize for an algorithm to recommend movies to users that they would be interested in that could beat their current algorithm by at least 10%. This paper reviews the methodology taken by the winning team to achieve victory after 3 years of work.

    The winning solution was a combination of methods incorporating weighted nearest neighbors, matrix factorization, gradient decent, temporal drift, and various extensions to the standard methods employed with each technique. Numerous models of the data were created using these methods and the winning solution was an ensemble, or averaging, of the models. The winning team was a collaborative effort among many researchers and the authors highlight the benefit of croudsourcing to finding a winning algorithm.

    The part of the algorithm that I liked most was matrix factorization. Its a technique that can extract meaningful relationships from an unknown dataset. In this case it was able to separate funny movies from serious movies, dramas from comedies, and so on. The utility of such a method can be easy extrapolated to a vast number of other problems. I plan to investigate matrix factorization in more depth as a result.

    All Together Now: A Perspective on the NETFLIX PRIZE. Robert M. Bell, Yehuda Koren, and Chris Volinsky


  • Review of “Introduction to the special issue on empirical evaluations in reinforcement learning”

    This is a brief review the introduction article for the Machine Learning Journal “Special Issue on Empirical Evaluations in Reinforcement Learning”, Volume 84, Numbers 1-2 / July 2011. The article does a good job at tying together seemingly disparate empirical research papers on machine learning. It presents a few past empirical studies and provides brief summaries for each of the papers presented in the current issue of the journal. Topics include:

    • How should software for empirical research be designed?
    • What are the right evaluation methodologies?
    • What are suitable benchmark tasks?
    • How do current methods fare in empirical comparisons?

    While no definitive answers to any of the questions is presented, the papers summarized by the author’s of this introduction highlight current progress towards answering them from an empirical research perspective. Overall, this was a well organized paper that brought with it the depth of the author’s understanding of the field of machine learning. It puts the unfamiliar researcher on firm footing for further exploration of the state of the art in machine learning.

    Introduction to the special issue on empirical evaluations in reinforcement learning