Hi Ryan! I have submitted a draft proposal titled "Nonlinear dimensionality reduction techniques for mlpack". Please do have a look at it when possible :)
Regards, Sahil C. ----- Original Message ----- From: "ryan" <[email protected]> To: "Sahil Chelaramani" <[email protected]> Cc: "mlpack" <[email protected]> Sent: Friday, March 24, 2017 2:47:24 AM Subject: Re: Ideas for improving Maximum Variance unfolding in mlpack On Thu, Mar 23, 2017 at 11:14:53PM +0530, Sahil Chelaramani wrote: > Hi Ryan, > > I think that both LLE and IsoMap would be useful to add, even though > they are a little bit older. The other techniques that I had in mind > were, the following:- > > a. Deep Canonical Correlation Analysis - This works specifically when > considering two different modalities and the transformation space is > maximally correlated.( > http://ttic.uchicago.edu/~klivescu/papers/andrew_icml2013.pdf) -- This > is not really a dimensionality reduction technique but is a very > useful transformation to have. > > > b.Maximally Correlated Principal Component > Analysis(https://arxiv.org/pdf/1702.05471.pdf) > > c. I was looking at other papers which use autoencoders to perform > dimensionality reduction like https://arxiv.org/pdf/1511.05644.pdf > (This paper claims that we can use an adversarial autoencoder for > dimensionality reduction, the results of which seem promising) > > I think we could try getting LLE and IsoMap or one of the above > mentioned? Also, if this is fine with everyone, could I start drafting > a proposal for these projects? Hi Sahil, I agree, I think these could make up an interesting project if you chose not to go the MVU reformulation route. The MCPCA paper doesn't consider runtime at all, so it would be interesting to see if it would be fast enough for general-purpose use. You are of course welcome to draft a proposal---submissions are open now on Google's Summer of Code website. Thanks, Ryan -- Ryan Curtin | "Why is it that the landscape is moving... but the boat is [email protected] | still?" - Train Driver _______________________________________________ mlpack mailing list [email protected] http://knife.lugatgt.org/cgi-bin/mailman/listinfo/mlpack
