my name is tone each of are i mean assistant professor computer science i been
at columbia for a little over five years now and might area of research is
machine learning i direct the company machine learning lab and you have a large group
of students doing several really exciting projects the machine learning lab is really about this
velocity of combining complication computer science
and marrying into statistics because there is so much data out there where we have
not just an information age of an information overload age and the real hopes to
use computers to help us make sense of the data automatically
we want them to learn much with people learn and also work with the types
of data that we care about so the text we read the images we see
we want computers to be able understand that you to that we're generating everyday at
a faster and faster rate
in biology for example there's millions of variables extends of thousand genes and it's almost
impossible to have someone look at this data and come up with a theory about
how the biology works
so increasingly side to succumbing to computer scientists and machine order saying
we've made all these measurements there's just too many variables and machine learning is one
of the few tools that really can work with this type of data machine learning
can provide us with a network description of visualisation clustering production and so on which
sciences finding very valuable these days
another thing we've been working with its social network analysis and
biological networks is another natural counterpart
looking at networks of proteins figure out how they interact with the proteins functions are
how expression levels vary over time i think machine learning is one of these
unusually lucky feels in that
the foundations it's working from are useful to many other disciplines
are particular research is machine learning applied to really complicated problems and datasets
where there is some additional structure that space so images transform in various ways if
you see picture someone it rotates the latter the right or you move around somebody's
face in an image you still recognizing and so we're trying to incorporate that same
type of
structure into all are machine learning algorithms we also design algorithms and machine learning models
at work on sequences so then we can handle things like a string of text
we've been able to do that very successful using machine learning by modeling the sequence
structure of the text
give me two documents are not tell you they were uttered by the same person
or there are some subtle stylistic things between those two documents that say that this
is in the same person
in my mind that's the really exciting future direction from machine learning one of the
areas we concentrate on my group which is how to incorporate some of this invariance
we know exist special you picture a short you tell to you still recognise that
actually see later on
and that's kind of the key i think too many real-world problems that
invariance for that particular problem