Metablake

Minimum Volume Embedding

Blake — Mon, 29 Jan 2007 18:03:00 +0000

Here is a pre-final-version of my first published paper in Machine Learning. The following paper is written with Prof. Jebara at Columbia, and will be published at the AI and Statistics 2007 conference. How exciting!

Abstract:
Minimum Volume Embedding (MVE) is an algorithm for non-linear dimensionality reduction that uses semidefinite programming (SDP) and matrix factorization to find a low-dimensional embedding that preserves local distances between points while representing the dataset in many fewer dimensions. MVE follows an approach similar to algorithms such as Semidefinite Embedding (SDE), in that it learns a kernel matrix using an SDP before applying Kernel Principal Component Analysis. However, the objective function for MVE directly optimizes the eigenspectrum of the data to preserve as much of its energy as possible within the few dimensions available to the embedding. Simultaneously, remaining eigenspectrum energy is minimized in directions orthogonal to the embedding thereby keeping data in a so-called minimum volume manifold. We show how MVE improves upon SDE in terms of the volume of the preserved embedding and the resulting eigenspectrum, producing better visualizations for a variety of synthetic and real-world datasets, including simple toy examples, face images, handwritten digits, phylogenetic trees, and social networks.

Minimum Volume Embedding (not final version)

Learning from a Visual Folksonomy: Automatically Annotating Images from Flickr

Blake — Mon, 08 Jan 2007 22:18:00 +0000

Recently, a large visual dataset has emerged from a web-based photo service called Flickr which utilizes the organizational power of folksonomy to label a tremendous amount of visual data. Flickr users upload snapshots from their digital cameras to the web, and if marked as public, the community annotates these images with descriptive tags. Can this large collective labeling effort be used to train a computer to annotate images? What concepts are we able to train a computer to visually identify?

This project uses a simple crawler to download photos from Flickr labeled with a certain tag, and then extracts color and texture features from these images so that they can be used to train a classifier, such as a Support Vector Machine (SVM). By automating this process of downloading images, extracting features, training, and testing, we are able to apply our system to many different tags and see which tags correspond to identifiable visual features. We have found that the system performs relatively well annotating images with one label, selected from a small vocabulary, for images belonging to concepts with distinct color and texture features. (Full paper found here)

Utilizing Folksonomy: Similarity Metadata from the Del.icio.us System

Blake — Sun, 01 Jan 2006 16:55:00 +0000

Traditionally, metadata is thought of simply as keywords that describe some content, and while the primary aim of folksonomic systems like the Del.icio.us bookmarking tool is to produce these keywords, a richer set of metadata is also produced. Because these keywords are now contributed from many different individuals and aggregated, useful information comes not only from the keyword itself but also from the information about who contributed to labeling the content with that keyword. This idea can be broadened to a general framework for producing a new layer of metadata: similarity between concepts. By analyzing the distributions of how users apply tags, how tags are applied to links, and how users pick content, we should be able to calculate the "distance" between tags, users, and content. This "distance" metric could then be used to construct a more powerful tool for browsing content, allowing the user to specify a query made up of keywords, content, or even other users. Furthermore, this metadata can be condensed into a lower dimensional space and visualized in order to gain better insight into the relationships between the concepts themselves. (Full paper found here)

Visualizing Folksonomies using Machine Learning Algorithms

Blake — Tue, 03 Jan 2006 19:45:00 +0000

This paper, written for my Adv. Machine Learning class, investigates using Semidefinite Embedding (SDE) to visualize data collected from a folksonomy. The del.icio.us social bookmarking service is a perfect example of a folksonomy; a community of users label websites with descriptive tags. Each tag exists in a high-dimensional space corresponding to the frequency of use of that tag among all the users of the system. We are motivated by the following question: can we find a simple low-dimensional structure for these tags that captures the significant relationships inherent in the data? In this paper we explore Semidefinite Embedding, an algorithm for non-linear dimensionality reduction, and its application to visualizing folksonomic systems, focusing on the effects of specifying different levels of connectivity for the data and the heuristics which can be used to find the best parameters for the algorithm. (Full paper found here)

Machine Learning Techniques for Visualizing CUtunes Data

Blake — Sat, 14 May 2005 02:15:00 +0000

Here are some cool images generated using CUtunes data. The first is a map of users, and the second is a map of artists. Distance on the map is a measure of similarity. The third points out some interesting features of the artist map.

Machine Learning Final Project: Locally Linear Embedding for Visualizing High-Dimensional Data

Building A Better Folksonomy

Blake — Sun, 01 Jan 2006 16:45:00 +0000

We live in an age flooded with information. New technologies are making available many large unstructured sets of information. As this information becomes more available, it becomes more difficult to navigate without a guide. Now that a typical user can carry around 10,000 songs in his pocket, the choice of picking which song to listen to becomes increasingly more difficult. Now that a typical user can access 13 billion websites, how does a person know which sites are relevant to him?

The solution to this problem resides in building new web-based technologies that aid in the formation of folksonomies. Folksonomy is commonly defined as a large group of people spontaneously cooperating to organize information into categories [24]. Many websites today are taking advantage of the organizational powers of folksonomies, such as Wikipedia, Flickr, Technorati, Del.icio.us, Yahoo!, and others. All of these sites employ a simple tagging mechanism, where users attribute words or phrases to content. When these tags are aggregated, new metadata for that content is created.

Tagging offers amazing possibilities for information retrieval by using collective social intelligence to organize information instead of relying on one person’s description or categorization. However, tagging only begins to approximate an ideal folksonomy. By simplifying the ways in which we collect metadata from the user, coupling this information collection more strongly with a social framework, and providing more powerful tools for categorization, we should be able to greatly improve systems for retrieving relevant information. (Full paper found here)

CUtunes Update

Blake — Mon, 02 Jan 2006 20:57:00 +0000

CUtunes is looking better after another semester of work. Here is the updated documentation, and some screenshots. Notable new features are user profile pages, flash-based visualizations of your musical neighborhood, and inteligent playlist creation in itunes, allowing the user to say make me a playlist that is like a specified list of musical artists and CUtunes users.

CUtunes

Blake — Sat, 14 May 2005 01:45:00 +0000

CUtunes is a tool for exploring the music listening habits of the Columbia University community. We provide basic services, including lists of top 40 songs, albums, and artists being played, as well as compatibility ratings between users, and recommendations for new music.

I started this project with a few other students this semester. Here is our end of the semester report.

CUtunes Project Report

LLE/NMF for Visualzing Clusters in Genetic Data

Blake — Sat, 14 May 2005 01:33:00 +0000

For my computational genomics class this semesters we created a tool to visualize clusters in gene expression data collected from microarrays. Here is a link to the paper, and an image from our analysis of a leukemia data set.

CS4761 Project: Visualizing Clusters in Microarray Data

Visualizing Music

Blake — Fri, 01 Apr 2005 01:29:00 +0000

This is a screen shot from some of the work i have been doing for cutunes, and for my databases class. There will be more extensive posts about each soon. This pic shows a portion of the bar chart for my most played songs. The size of the bars represent play counts.

Computer Vision Project

Blake — Sat, 20 Nov 2004 21:39:00 +0000

I have started working on the team final project for my computer vision class. I think we are going to try to do gesture recongition using an apple iSight, and either a wand or a Minority Report style glove with leds on it. Here is the initial test where we simply track a green highligher.

Parcel

Blake — Wed, 12 May 2004 21:31:00 +0000

Parcel is a two-dimensional graphical processing language designed to simplify the task of producing meaningful representations of data models and simulations for statistical, scientific, and artistic applications. The fundamental object in Parcel is the cel. A cel is defined proportionally to other cels; it can be anything from a pixel to the whole screen depending on context. Furthermore, cels can be nested and tiled within each other. This fundamental data structure allows for unique highlevel, abstract drawing and animation, independent of the underlying graphical implementation.

Here are some sample images:

and documentation:
Reference Manual

Hough Transform

Blake — Tue, 09 Nov 2004 20:19:00 +0000

Here is some cool stuff that I have been working on for my Computer Vision class. We have been playing around with using the Hough Transform for images.

Here is the original Image:

Here is the corresponding hough space, which is a representation of how strong lines are in the image.

Then by finding local maximas in the hough space we can plot the strong lines of the image back onto the image.

I really think that the hough space is beautiful. It's so neat that it is simply another representation of the image. I really want to do some large prints of the hough space.

Jude

Blake — Wed, 01 Sep 2004 01:47:00 +0000

In order to thoroughly procrastinate studying for the GRE, I have been working on a cocoa application to generate fractals. I love the screensaver Fracture however was disappointed that the developer never built a standalone application. So here is the first release of Jude, a simple cocoa application which makes julia fractals. I am very happy with how it has turned out so far. It can import julia .fracture files (minus the color scheme) as well as save and write its own filetype. Jude also has a wide range of options for controlling the color scheme and zooming in on the fractal. Check it out

Forest Fire Image Segmentation

Blake — Thu, 29 Jul 2004 20:12:00 +0000

Here is yet another post about work, although probably it will be one of the last. I have finished the Forest Fire Image Segmentation Package that I have been working on. It builds on any *nix machines, check out the documentation for more information on building and dependencies. I don't know how useful this will be for most people not interested in forest fire simulation, however, here it is.

Presentation...

Blake — Mon, 26 Jul 2004 04:44:00 +0000

I am giving a short presentation at work tomorrow about my aspect of the project. Here.

Haiku Checker 0.2

Blake — Tue, 27 Jul 2004 06:00:00 +0000

Just finished up the new version of the Haiku Checker that I am working on... This version even has a Mac OS X application Feel free to download and try it out. Complaints can be sent to haiku@metablake.com.

Haiku Checker

Blake — Tue, 13 Jul 2004 17:25:00 +0000

A bunch of my friends and I have started writing haikus at work to pass the time, and posting them to my friend Lesley's Live Journal. I was curious to see if I could write a Haiku Checker. A simple program that checks all of the syllables against dictionary.com and tells you if in fact what you wrote is a haiku (17 syllables). Here is my first beta. It currently is just a command line tool, and does not accept any punctuation and is wrong sometimes. However, it is getting there, and I figured I should share it. Haiku Checker Beta Source

Summer Job

Blake — Thu, 08 Jul 2004 04:46:00 +0000

I realized that I have yet to blog about by current employment. Now that I am actually starting to get some results I figured now would be a good time. I am working with David Keyes, a professor at columbia in the applied math department on a project with the goal of finding a better way to simulate forest fires. My current aspect of the project is working on automatically detecting the boundary of fires in past forest fire images. Here are some shots of the GUI I have written to do this. Currently I am using Java for the GUI and initial image analysis and a C++ Level sets package to do the segmentation. Check out the pic.

A Graphical Programming Environment?

Blake — Thu, 20 May 2004 22:09:00 +0000

Here is an interesting proof of concept for a graphical programming language. I'm actually not sure if 'language' is the correct word. It would really be more like a graphical programming environment.

I found out that omnigraffle pro is incredibly scriptable. Through applescript one can have complete knowledge of every object in the document and how each object is connected. So here is a simple hello world 'program' drawn in omnigraffle, and an interpreter written in applescript.

This is clearly a very simple implementation, but it is interesting to think about the possibilities of this kind of graphical language. One of the first features I would like to implement would be an "Abstract" button, which would take a selection of objects and replace them with one small object with the same properties. This abstraction would allow the language to conquer large problems while maintaining its visual simplicity.

Download program and interpreter

A New Metablake.com

Blake — Tue, 18 May 2004 07:33:00 +0000

Well, it's official. Here is the new metablake.com. The old flash site can still be used to browse old projects. I think this blog style will be much better at organizing my projects. And on that note, here is a pretty fractal from parcel.

MetaBlogger 0.2 is done. Here is a list of new fe...

Blake — Tue, 18 May 2004 23:55:00 +0000

MetaBlogger 0.2 is done. Here is a list of new features:

Remembers username and blogid

GUI rearranged, actions moved to menus with keyboard shortcuts

Uploading and Downloading templates works

Better status reports

An actual installer

Here it is for download: MetaBlogger 0.2

Here is the first beta of MetaBlogger. Enjoy! ...

Blake — Sun, 16 May 2004 07:05:00 +0000

Here is the first beta of MetaBlogger. Enjoy!

MetaBlogger 0.1

Note: This xml-rpc code needs to be added to your java classpath. Drag to ~/Library/Java/Extensions

I was so excited when I found out that you could a...

Blake — Sun, 16 May 2004 03:36:00 +0000

I was so excited when I found out that you could access blogger using xml-rpc commands, that I decided to start working on a little cocoa app that will interface with blogger. It isn't much to look at for the time being, but I definitely plan on developing it further in the near future. Check out this screenshot.