Google Groups
3 views
Skip to first unread message
jo.wor...@gmail.com
Oct 3, 2011, 5:43:08 AM10/3/11
to oxford-mach...@googlegroups.com
Well done, Roko...how much is MATLAB setting you
back? And where are you intending to do your
PhD?
rmijic
Oct 3, 2011, 1:21:17 PM10/3/11
to oxford-machinelearning
MATLAB is 50 quid for me, which seems like a reasonable price if it
helps me to get a PhD. I really want to study something called
hierarchical Bayesian models, which is very much related to
psychology,particularly cognitive development. See
http://www.youtube.com/watch?v=k9MnT5fOLA0 for a video introducing the
ideas from Prof. Tenenbaum at MIT.
helps me to get a PhD. I really want to study something called
hierarchical Bayesian models, which is very much related to
psychology,particularly cognitive development. See
http://www.youtube.com/watch?v=k9MnT5fOLA0 for a video introducing the
ideas from Prof. Tenenbaum at MIT.
rmijic
Oct 3, 2011, 1:24:24 PM10/3/11
to oxford-machinelearning
see also http://www.brianesty.com/pdfs/Science-2011-Tenenbaum-1279-85.pdf
for more detail.. not sure where yet but probably london
for more detail.. not sure where yet but probably london
Jo Jordan
Oct 3, 2011, 2:05:22 PM10/3/11
to oxford-machinelearning
Great... I'll look at those later.. homework calls... I saw UCL has a
masters in machine learning... or are you trying to jump straight into
a PhD?
On Oct 3, 6:24 pm, rmijic <rmi...@gmail.com> wrote:
> see alsohttp://www.brianesty.com/pdfs/Science-2011-Tenenbaum-1279-85.pdf
masters in machine learning... or are you trying to jump straight into
a PhD?
On Oct 3, 6:24 pm, rmijic <rmi...@gmail.com> wrote:
> see alsohttp://www.brianesty.com/pdfs/Science-2011-Tenenbaum-1279-85.pdf
Jo Jordan
Oct 3, 2011, 4:12:57 PM10/3/11
to oxford-machinelearning
OK watched the Video and get the general direction.. makes sense... I
had precisely that problem on my current project.. gave up data mining
for sitting and reading academic papers in another discipline.. one by
one.. Very tedious.
Will read the paper from Science tomorrow.
> see alsohttp://www.brianesty.com/pdfs/Science-2011-Tenenbaum-1279-85.pdf
had precisely that problem on my current project.. gave up data mining
for sitting and reading academic papers in another discipline.. one by
one.. Very tedious.
Will read the paper from Science tomorrow.
> see alsohttp://www.brianesty.com/pdfs/Science-2011-Tenenbaum-1279-85.pdf
rmijic
Oct 3, 2011, 5:14:14 PM10/3/11
to oxford-machinelearning
The techniques in the paper are quite advanced but the idea is
simple:
In the past, there has always been a dichotomy between logic-based
theories of knowledge which are highly structured but very brittle and
nearly impossible to connect to sensory data i.e. learn, and
statistical/connectionist theories of knowledge which operate over the
simplest, unstructured forms of knowledge.
It appeared necessary to accept either that people’s abstract
knowledge is not learned or induced in a nontrivial sense from
experience (hence essentially innate) or that human knowledge is not
nearly as abstract or structured (as “knowledge-like”) as it seems
(hence simply associations).
Now, with hierarchical Bayesian models, Tenenbaum & co seem to have
found a way to have it both ways: a way of expressing knowledge which
is both highly structured and based on statistics with all the
advantages in robustness and learnability which statistics brings.
Anyway this is just one direction which you can take machine learning,
it happens to be the one which interests me
simple:
In the past, there has always been a dichotomy between logic-based
theories of knowledge which are highly structured but very brittle and
nearly impossible to connect to sensory data i.e. learn, and
statistical/connectionist theories of knowledge which operate over the
simplest, unstructured forms of knowledge.
It appeared necessary to accept either that people’s abstract
knowledge is not learned or induced in a nontrivial sense from
experience (hence essentially innate) or that human knowledge is not
nearly as abstract or structured (as “knowledge-like”) as it seems
(hence simply associations).
Now, with hierarchical Bayesian models, Tenenbaum & co seem to have
found a way to have it both ways: a way of expressing knowledge which
is both highly structured and based on statistics with all the
advantages in robustness and learnability which statistics brings.
Anyway this is just one direction which you can take machine learning,
it happens to be the one which interests me
Reply all
Reply to author
Forward
0 new messages