Russell Greiner

Positions for PostDoctoral Fellows

The Alberta Innovates Centre for Machine Learning is looking to hire strong researchers as postdoctoral fellows (PDFs), for various projects, both application pull and technology push; see also the 1page ad (PNG PDF); this page is http://webdocs.cs.ualberta.ca/~rgreiner/index1.php?section=PostDoc15

If you are interested in any of the positions mentioned below (note the Other category), please email ...

a cover letter, specifying which positions most interest you and indicating why you feel that you qualify;
(summarize what projects you would like to work on, and how you would work on each)
your CV, including a description of any previous research or industrial jobs you have held; and
names and email addresses of at least 3 references.

to Leslie Acker (acker@ualberta.ca).
Use 'PDF Medical Informatics' as the subject of your email.

While we have funding for some of these positions, it is always helpful if you have funding to bring in (part of) your salary from some external sources (eg, a PostDoctoral Fellowship).

For a quick overview, see Slides

Patient-specific Treatment

Patient-specific Survival Prediction

Brain Tumor Analysis Project
Proteome Analyst
Budgeted Learning
Learning Belief Nets
Other

See also our recent Publications; note that essentially all involve collaborations with students, colleagues, and postdocs!

Patient-specific Treatment

Learn which treatment should be most effective for each specific patient, based on

Genetic data -- SNP (single nucleotide polyomorphism) profiles
Proteomic data -- microarrays (showing gene expression levels)
Metabolomic data -- eg, NMR urinalysis
Histological data, ...

Many benefit from analysis of Patient-Specific Survival Prediction.
See also Medical Tasks.

This involves several different projects, with different teams of collaborators:

Cancer: within PolyomX project -- a team of Medical Researchers (cancer genomics) at the Cross Cancer Institute; see also ad.pdf).
Inflamatory Bowel Disease -- with Alberta IBD Consortium; see also Metabolomic Technologies Inc
Diabetes -- see Intelligent Diabetes Management
Psychiatric disorders -- see fMRI Projects
Organ Transplants -- with Alberta Transplant Applied Genomics Centre [ATAGC]

Brain Tumor Analysis Project

Learn to predict clinically useful patient characteristics:

Survival time (see also Patient-Specific Survival Prediction)
Cancer progression -- transformation, pseudo-progression, ...

... in collaboration with Radiation Oncologists at the Cross Cancer Institute.
Project Webpage

(Details)

Proteome Analyst

The Proteome Analyst system can analyse a set of peptide sequences (proteins) in a given proteome and return the general function, and subcellular location, of each protein, as well as a functional summary for the entire proteome. The current version first maps each novel protein to a set of attributes -- namely the tokens that appear in certain fields of the (known proteins) homologs found by Blast -- then finds the general function (resp., subcellular location) most associated with this token-set, based on a learned classifier. We are looking for a researcher (summer student, grad student, postdoctoral fellow) to help us extend Protein Analysis in several ways:

to take a genome as input (rather than a proteome),
to use other information in the classification, including the secondary structure of possible homologs, as well as information about the other sequences given in the proteome
to use more sophisticated learning algorithms, (including "mixture" methods to combine multiple classifiers, learned based on different features of a protein) to produce more accurate classifications
to build a generic configurable learning tool, that can be used to learn classifiers for other related tasks.

(Note this project is currently "on hold", but we are planning to ressurrect it.)
Project Webpage

Budgeted Learning

Learning tasks typically begin with a data sample --- eg, symptoms and test results for a set of patients, together with their clinical outcomes. By contrast, many real-world studies begin with no actual data, but instead with a budget --- funds that can be used to collect the relevant information. For example, one study has allocated $30 thousand to develop a system to diagnose cancer, based on a battery of patient tests, each with its own (known) costs and (unknown) discriminative powers. Given our goal of identifying the most accurate classifier, what is the best way to spend the $30 thousand? Should we indiscriminately run every test on every patient, until exhausting the budget? Or, should we selectively, and dynamically, determine which tests to run on which patients? We call this task budgeted learning.

There are many open questions, both theoretic and empirical.

Formal analysis: is the task NP-hard given certain standard assumptions?
Are there any algorithms that are PTAS (approximation algorithms)?
Better heuristic algorithms and other efficiency tricks
Further empirical studies, over other datasets

Project webpage

Learning and Validating Belief Nets

Bayesian belief nets (BN) are becoming the preferred tool for a wide variety of tasks, ranging from sensor fusion to information retrieval. We are currently developing and experimenting with various tools for learning these BNs from training data. We are looking for a student to help us here, both in developing and implementing these learning system, and also in running careful experiments to help us compare these different approaches. We also plan to investigate ways to learn, and use, Probabilistic Relational Models --- extension to belief nets that allow the representation of relationships.

We will also explore ways to compute and use "variance" around the belief net response; see webpage: extending the work on "mixture using variance" and perhaps a variance-based model of value-of-information.

Other

We are always open to ideas, especially ones that are motivated by medical tasks... perhaps related to Patient-Specific Survival Prediction ... or covariate shift? ... or "large p, small n"? ...)

Here, you need to motivate the task, by explaining why it is relevant to some real (medical?) problem, and why it involves good science to solve!