Objectives

Promote open science through shared data and collaboration

We share our own code and data openly, run and support public data archives, and develop the data standards that let labs share and combine their studies.

Much research data never leaves the lab that collected it. Results are then hard to check, and questions that need larger or more diverse samples go unanswered.

11 publications (4 papers, 7 talks)

Search and filter them on the publications page

Connect tools, standards, and communities

We work across data standards, such as BIDS, NWB, and DICOM, and across research initiatives, so that data, tools, and conventions from one community can be used in another.

Standards and initiatives often grow up separately. The same data then has to be converted again for each one, and solutions found in one community rarely reach the next.

Read more:

34 publications (14 papers, 6 posters, 14 talks)

Search and filter them on the publications page

Preserve research artifacts

We are building tools and practices that keep data, code, recordings, and other research materials findable and recoverable, with copies in more than one place.

A paper is only part of the work. The data, code, and notes behind it are easily lost when a service shuts down, a link breaks, or a project ends.

21 publications (4 papers, 4 posters, 13 talks)

Search and filter them on the publications page

Self-hosted research tools and services

We are making research software and services that groups can run on their own computers, such as NeuroDebian, Debian Med, DANDI, and ORINOCO Lite; con/serve collects how to run them.

Many research groups depend on tools and services they do not control. When a provider changes its terms or shuts down, a group can lose the services its work depends on, and sometimes its data.

20 publications (6 papers, 4 posters, 10 talks)

Search and filter them on the publications page

Turn research ideas into working tools

We turn published methods into turnkey software that researchers can install, inspect, and run on their own data. PyMVPA, our framework for machine learning analyses of brain imaging data, was one of the first to do this for neuroscience.

Many new methods are described in papers but never released as usable software, so other researchers cannot try them.

41 publications (11 papers, 8 posters, 22 talks)

Search and filter them on the publications page

Understand brain function during everyday experiences

We are developing ways to analyze brain activity and behavior recorded during rich experiences, such as watching a film. Hyperalignment, which aligns such responses across people, was first implemented in PyMVPA; this work builds on Psychoinformatics' work on naturalistic studies.

Films combine images, sound, and events that unfold over time. Most analysis methods were made for simple, controlled tasks, so much of what such recordings contain goes unused.

42 publications (11 papers, 8 posters, 23 talks)

Search and filter them on the publications page

Build tools for reproducible, shareable science

We are building tools that record which data and software produced each result, so an analysis can be rerun, checked, and shared.

Analyses often depend on manual steps, unrecorded software versions, and files that change without notice. Others then cannot repeat the results, and even the original authors may struggle to.