PyMVPA - NiPy week

The authors of the PyMVPA Multivariate Pattern Analysis toolkit will be visiting the UC Berkeley Brain Imaging Center from April 14 until April 16 2009, to collaborate on the integration between PyMVPA and NiPy.

As part of their visit, they will be presenting a two-part seminar on multivariate pattern analysis. The first part is a general discussion on the growing use of these techniques for data analysis in neuroscience, and the second part is a more hands-on presentation of the pymvpa project. We will take a short break between the two parts (at 11am) so that you can attend only one of them if you need.

See below for details on these presentations, and do not hesitate to contact me if you’d like to schedule a technical discussion with Michael and Yarick while they are visiting us.


Part 1: Reliable Decoding of Neural Data

  • When: April 14 2009, 10 am
  • Where: Tolman 5101

In the last five decades the number of techniques available for non-invasive functional brain imaging has increased dramatically. Researchers today can choose from a variety of neural data modalities such as fMRI, EEG, MEG, etc. The peculiarities of each data acquisition modality and the lack of strong interaction between the neuroscience communities employing them have produced distinct analysis pipelines specialized for the conventional analyses within a particular modality. Some analysis techniques have become, due to normative concerns, de facto standards despite their limitations and inappropriate assumptions for the given data type (e.g. GLM in fMRI, ERP in EEG).

However, outside the neuroscience community, machine learning research has spawned a set of analysis techniques that are typically generic, flexible (e.g. classification, regression, clustering), powerful (e.g. multivariate, linear and non-linear), and capable of efficient use of both spatial and temporal evolutions of the signal.

Recently, neuroimaging researchers have begun to explore various multivariate methods to address the shortcomings of the conventional analysis approaches. Drawing on the field of statistical learning theory, new analysis techniques possess explanatory power that could provide new insights into the functional properties of the brain. By constructing a decoder of neural data new analysis methods often reverse the direction of the analysis and target the description of the behavior or the environment in terms of the registered neural activity. Such reversed paradigm can account for the covariance/causality structure within the data and often allows for single-trial analysis of various neural data modalities by decoding them into variables of interest. Unbiased testing of the decoder on new samples of the data provides an assessment of decoding performance validity. Furthermore, consecutive analysis of the constructed decoder’s sensitivity allows to identify neural signal components relevant to the task of interest, hence providing desired localization. These features make decoding approach more powerful than conventional hypothesis-testing univariate methods which currently dominate the neuroscience field.

This part of the PyMVPA-series would urge the necessity in developing decoding methods of neural data analysis and will present already published approaches and results.

Part 2: PyMVPA: Fathom Brain Function with Multivariate Pattern Analysis

  • When: April 14 2009, 11 am
  • Where: Tolman 5101

Unlike the wealth of software packages for univariate analyses, there are few packages that facilitate multivariate pattern classification analyses of neural data. We offer a complete analysis framework to exploit techniques from statistical learning theory for the reliable analysis of neural data.

Here we introduce a Python-based, cross-platform, and open-source software toolbox, called PyMVPA, for the application of statistical learning methods toward the analysis of various neural data modalities. PyMVPA makes use of Python’s ability to access libraries written in a large variety of programming languages and computing environments to interface with the wealth of existing machine learning packages. In our strong belief it is important not to limit the researcher in a single analysis paradigm as it is often done by GUI-based toolkits formalizing a specific analysis approach. Having that in mind, PyMVPA can be viewed as a complete scripting environment which is capable not only of replicating existing analysis methods, but also versatile enough to implement a variety of custom analysis paradigms based on statistical learning methods.

We will describe the PyMVPA framework: its building blocks and convenience facilities. Furthermore we will provide illustrative examples on its usage, features, and programmability.