New Median

Today I was at a presentation of Partha Niyogi's work. Professor Niyogi is working on machine learning problems and is currently employed by the university of Chicago. Today's presentation at UCL's Gatsby computational neuroscience unit, entitled "A geometric perspective on machine learning", presented his current research.

Machine learning problems are fundamentally ill-posed. The main reason for this being the fact that given a finite set of vectors, there is an infinitely large set of functions that can interpolate the data. The usual approach would usually deployed in today's methods and literature is based on regularization on the ambient space. The framework presented today is based on the assumption that although the data may be living in extremely high dimensional space, we can somehow relate the data to a low dimensional manifold, embedded in this space.

The approach presented is very data-centric, trying to derive a manifold through associating a graph with the dataset and moving through graph and manifold representation. A number of interesting findings, including the fact that the bounds do not not depend on vc dimension, but only by the norm.

However just before you decide you're sold on this method keep in mind that there are a number of problems associated with this approach. More specifically this method depends on a number of model parameters which right now cannot be ab initio determined, and more importantly the fundamental assumption of the manifold being well defined and embedded.
However in my view, it is a really interesting approach, and it has a wide spectre of applications on various machine learning methods from disparate fields, including semi-supervised learning, clustering, and online compression learning.

This project is starting out as an attempt to illustrate how new media can make use of various different machine learning methods. In trying to do so the focus is shifted on a number of simple solutions that can be combined in order to achieve workflows that can produce unique and innovative results. In the last few years numerous technologies have emerged that allow for easy creation, and publishing of content. We live in a world where millions of people have a voice on a global network and countless bits of information are created and posted every second. Big media is now in a transitive state to something vastly different, communication mediums are being democratized. The web is evolving into a 2-way communication platform, in which information can be transmitted, processed, manipulated and seeded in unique ways. However in a world where everyone has a voice trying to keep track of what can be considered as useful information can be exceedingly hard. The possibilities of this new medium are bounded by its very strength, the fact that it is a place where the possibilities and the resources are countless.

What we want to do is to provide methods, tools ideas and an overall paradigm that will help in achieving the full potential of this process. In our view this mission entails the exploration of the mediums internals and dynamics, the findings of which should be shared with the public. In order to do just that it becomes natural to support open standards, that have helped the medium develop so far, in order to spark further development through experimentation and mutual contributions.

Finally, we believe that contributions to a project like this can have various forms, but the what we wish to give back is what we need. And that, in its most basic form is ideas. We wish to explore new ideas and see how they work out. They don't have to be novel and definitive approaches to hard problems in order to be viable. What the web is missing right now, is just some simple ideas to make our stay more useful and more fun.

This is the first spot here and it is actually posted quite rushfully. I'm going to keep it simple and just state a draft of what may possibly become this blog's mission statement. New Median is about how artificial intelligence and machine learning methods can be used in the blurry context of what is known new media publishing. We have a wealth of algorithms that can be used in this framework but the intention is to create a systematic framework for the creation, management, enhancement and accessibility of online publications, while trying to study the engineering, social and philosophical effects of such paradigms