Ganesh Swami

As I had previously written, I’m doing a fair bit of machine learning these days for automated disease classification. I found a whitepaper by Tom Mitchell that gives a broad overview of the field.

How can we build computer systems that automatically improve with experience, and what are the fundamental laws that govern all learning processes?

Application Successes: Speech recognition, Computer vision, Bio-surveillance, Robot control and Accelerating empirical sciences.

Current research questions:

• Using unlabeled data for supervised learning
• Transfer knowledge from learning one task to another
• Relationship between different learning algorithms
• Strategy for learners that collect their own data
• Data privacy in data mining

The top conferences and journals in Machine Learning as stated in the paper:

• International Conference on Machine Learning (ICML).
• Conference on Neural Information Processing Systems (NIPS).
• Annual Conference on Learning Theory (COLT).
• Journal of Machine Learning Research (JMLR). Freely available.
• Machine Learning.

Still Alive. It’s been relatively quite here because I’ve been swamped with end-of-term activities. The problem with working on five year long projects is that deadlines are self-imposed. My strategy has been to make sticky notes with tasks and not getting up until I strike off the task and crumple the note. This has been super effective for me. It’s nice to feel a sense of accomplishment at the end of the day with the big pile of crumpled notes. Hey, what ever works!

This isn’t so good for learning stuff. I’ve had to do a fair bit of machine learning for some of the tasks I’m doing right now, and all of the language is foreign to me. It took me about a week to learn how to check for group mean differences between samples, but only about an hour to implement it in code. I guess this is “holding a program in one’s head.“

To meet accreditation requirements, Engineering students are required to do a major work term report atleast once. This is a big undertaking. Fortunately, I’m doing mine on my latest infatuation: fluid match. I’ve written a couple of sections, but it’s still lacking cohesiveness.

I’m also happy to say that I’ve decided to do my honors thesis at the Medical Image Analysis Lab, where I’ve been working for the past eight months. The exact details of my project are to worked out, so I don’t have much to say at this point.

I didn’t know what some of the settings on my camera were and this bothered me. I whipped out the manual (I thought I had trashed it, as I usually do) and checked it out. I can now take pictures like the one above with little difficulty, and it looks like I know what I’m doing. It’s still long ways before I exhaust the capabilities of my current camera.

Life’s Good.

From Wikipedia, “The Bloomberg Terminal is a computer system that enables financial professionals to access the Bloomberg Professional® service through which users can monitor and analyse real-time financial market data movements and place trades.”

Design firms thehappycorp, IDEO and Ziba design were asked to redesign the terminal, and the above is IDEO’s proof of concept. I like this a lot.

Named after Anton van Leeuwenhoek, the father of microbiology, this is the latest from D. E. Shaw Research. This paper describes the massively-parallel hardware part of their super software-hardware technology. I’ve previously described their software here.

Scheduled for completion at the end of 2008, Anton is built with the aim of doing millisecond simulations. In contrast with approaches like Folding@Home, they are interested in a single trajectory. Folding@Home patches together many small trajectories that are more feasible on a regular desktop PC. Their initial configuration will have 512 nodes.

Many of the ideas implemented in their software simulation engine Desmond were developed with Anton in mind. For long range electrostatic interactions, they use a method called k-space Gaussian split Ewald, reducing the computational burden from O(n^2) to O(n log(n)). Another advantage is that the same kernel is used for force interpolation and spreading saving real estate on the chip.

As any molecular dynamics aficionado knows, about 90% of the computational time is spent in computing non-bonded forces. With this is mind, they’ve done everything possible to speed up the process devoting much of the ASIC space. One of their ideas that struck me was the explicit computation between all pairs and then to go back subtract correction terms. This leads to a clear separation in the accounting between what particles are required for what forces – a major win for scalability.

They get major speedups by building a specialized hardware datapath and control logic tuned for common molecular dynamics communication patterns. They even have dedicated support to accumulate forces for reducing latency. I was previously excited by the addition of the dot product instruction haddps to recent generations of processors. Anton does dot products in hardware. Dot products can be used everytime you’d want to “squish” a vector.

Anton does all computation in fixed point arithmetic, which is faster and more accurate for the same silicon area (their claim.) Molecular dynamics is expensive in computation and communication, but not so much in memory. All of their data for a 25,000 particle system fits in L1 cache.

One of the courses in Engineering at SFU (ENSC 250) deals with designing a RISC processor from the ground up. I vaguely remember building branch prediction logic in VHDL, not really knowing if I’d ever use that knowledge again. If only I’d known…

Anyways, I’m missing out on some of the finer details of their technology. Go read the paper for the real deal.

Not for grocery, clothes, shoes, electronics or a car, but for university courses!

This made me laugh so much that it made me cry. Or maybe the other way.