Important Dates

What is NAILS?

NAILS (Neurally Augmented Image Labelling Strategies)
is an NTCIR-13 workshop data challenge task. Participants in the challenge will build machine learning models to classify neural signals that have been captured from human volunteers performing high-speed image search tasks. The aim here is to label/detect those images presented in an RSVP (Rapid Serial Visual Presentation) sequence that are target images for the search task by using neural responses i.e. search-relevant or non-relevant. Participant organisations in the competition will be provided access to a training dataset comprising of EEG (Electroencephalography) data in a number of pre-processed formats along with the respective ground truth labels. Organisations are expected to use this data to build machine learning models that they will then benchmark on a test set (where the ground truth labels are withheld by the NAILS organisers). Evaluation runs are submitted via a REST API system where participating organisations will be ranked against other competing organisations. Prior to the NTCIR-13 NAILS workshop participating organisations will be expected to submit a paper outlining their approach which they will then present at the NAILS workshop in Tokyo, Japan this coming December 2017.

Important: Although this is a collaborative evaluation where participating organisation's machine-learning strategies will be ranked in terms of balanced accuracy, it is expected that many good machine-learning solutions that may perform sub optimally to others in terms of accuracy alone may offer other advantages in terms of speed, model complexity, neurophysiological interpretability and/or cross-task applicability. Papers exploring such aspects of the data and tasks set are highly encouraged.

Explanation for Neuroscience/BCI researchers : NAILS is an EEG single-trial (machine-learning) prediction task using trials generated across a variety of visual oddball tasks. The tasks used have been verified to elicit the commonly known P300 oddball signal. The aim is to build prediction models from the available training data to maximise the BA (Balanced Accuracy) score when predicting trials on a test set (with withheld ground truth).

For further details on the dataset and tasks please see
DATA and the NAILS dataset overview paper.

How to Participate?
In order to gain access to the dataset and participate in the competition you should consult the NTCIR-13 page detailing registration information for task participation:

*User Agreement Forms:

Who is organising this?

Graham Healy
Graham Healy (* Lead Contact / Lead Organiser)Dublin City University, Ireland

Tomás Ward
Tomás WardMaynooth University, Ireland

Cathal Gurrin
Cathal GurrinDublin City University, Ireland
Alan Smeaton
Alan SmeatonDublin City University, Ireland