Brian McFee brm2132@columbia.edu

Luke / Gert / Brian

I develop machine learning tools to analyze multimedia data. This includes recommender systems, image and audio analysis, similarity learning, cross-modal feature integration, and automatic annotation.

As of Fall of 2012, I am a postdoctoral research scholar in the Center for Jazz Studies and LabROSA at Columbia University.

Previously, I was advised by Prof. Gert Lanckriet in the Computer Audition Lab and Artificial Intelligence Group at the University of California, San Diego.

In May, 2012, I defended my dissertation, titled More like this: machine learning approaches to music similarity [bib|pdf].

For a full history, here's my curriculum vitæ.

Teaching

Publications

2014
code | bib | pdf
Learning to segment songs with ordinal linear discriminant analysis
McFee, B. and Ellis, D.P.W.
International conference on acoustics, speech and signal processing (ICASSP)
2014
code | bib | pdf
Better beat tracking through robust onset aggregation
McFee, B. and Ellis, D.P.W.
International conference on acoustics, speech and signal processing (ICASSP)
2013
bib | pdf
Iterative category discovery via multiple kernel metric learning
International Journal of Computer Vision
2013
code | bib | pdf
Robust structural metric learning
Lim, D.K.H., McFee, B., and Lanckriet, G.R.G.
30th International Conference on Machine Learning (ICML)
2012
slides | code | data | bib | pdf
Hypergraph models of playlist dialects
McFee, B. and Lanckriet, G.R.G.
13th International Society for Music Information Retrieval (ISMIR) conference
2012
data | bib | pdf
How significant is statistically significant? The case of audio music similarity and retrieval
Urbano, J.Downie, J.S., McFee, B., and Schedl, M.
13th International Society for Music Information Retrieval (ISMIR) conference
2012
data | bib | pdf
The Million Song Dataset Challenge
4th International Workshop on Advances in Music Information Research (AdMIRe)
2012
bib | pdf
Learning content similarity for music recommendation
IEEE Transactions on Audio, Speech and Language Processing
2011
poster | data | bib | pdf
Best poster presentation award
The natural language of playlists
McFee, B. and Lanckriet, G.R.G.
12th International Society for Music Information Retrieval (ISMIR) conference
2011
code | bib | pdf
Large-scale music similarity search with spatial trees
McFee, B. and Lanckriet, G.R.G.
12th International Society for Music Information Retrieval (ISMIR) conference
June, 2011
bib | pdf
From region similarity to category discovery
IEEE conference on Computer Vision and Pattern Recognition (CVPR)
February, 2011
bib | pdf
Learning multi-modal similarity
McFee, B. and Lanckriet, G.R.G.
Journal of Machine Learning Research (JMLR)
February, 2011
bib | pdf
Contextual object localization with multiple kernel nearest neighbor
IEEE Transactions on Image Processing (TIP)
2010
bib | pdf
Learning similarity from collaborative filters
11th International Society for Music Information Retrieval (ISMIR) conference
2010
bib | pdf
Collaborative filtering based on P2P networks
11th International Society for Music Information Retrieval (ISMIR) conference
2010
poster | code | data | bib | pdf
Metric learning to rank
McFee, B. and Lanckriet, G.R.G.
Twenty-seventh International Conference on Machine Learning (ICML)
2010
bib | pdf
Multi-class object localization by combining local contextual interactions
IEEE conference on Computer Vision and Pattern Recognition (CVPR)
2009
poster | data | bib | pdf
Best presentation award
Heterogeneous embedding for subjective artist similarity
McFee, B. and Lanckriet, G.R.G.
10th International Society for Music Information Retrieval (ISMIR) conference
2009
Partial order embedding with multiple kernels
McFee, B. and Lanckriet, G.R.G.
Twenty-sixth International Conference on Machine Learning (ICML)

Software

Ordinal LDA
Python (sklearn) implementation of ordinal linear discriminant analysis.
LibROSA
A python package for music and audio signal analysis.
MLR
MATLAB implementation of metric learning to rank.
Hypergraph playlists
Python implementation of the model from this paper.
Spatial trees
Python implementation of spatial trees for approximate nearest neighbor search, as used in this paper.

More projects can be found on my GitHub profile.

Data

MSD Challenge
Large-scale music recommendation on the Million Song Dataset. See also the year 1 test set.
AotM-2011
Annotated playlists from Art of the Mix, indexed to the Million Song Dataset.
AotM-2003
An earlier collection of playlists from Art of the Mix, also indexed to the Million Song Dataset.
aset400 kernels
Kernel matrices for aset400 artist similarity experiments
eHarmony
Matchings and anonymized features for several hundred thousand eHarmony users.

Hack-day projects

Yankomatic
Automatically generate song title parodies using rhyming food words.
Handmadedness
Calculate how hand-made a song sounds by measuring repetition patterns and tempo stability.
Frankenmasher 2000
Isolate and recombine vocals and instruments from different songs to create mash-ups of hypothetical super-groups.
Mend-a-break
Resynthesize any audio clip using sparse convolutions of the Amen Break.
Auto-chip-tune
Generate Nintendo chip-tune programs from audio analysis.

See my hack-day blog for more projects.