Computer Vision News - September 2016

Every month, Computer Vision News reviews a challenge related to our field. If you can’t find time to read challenges, but are interested in the new methods proposed by the scientific community to solve them, this section is for you. This month we have chosen to review the Decoding Brain Signals competition held by Microsoft . The website of the challenge, with all its related resources, is here . Background Microsoft Corporation saw an opportunity to further the cause of neuroscience research. To that purpose, it partnered with a Stanford neurosurgeon, Dr. Kai J. Miller , to create a competition based on his original research. On one hand, millions of people are affected every year by brain injuries, related disorders and strokes. On the other hand, science does not possess yet a good understanding of how our brain interprets electric signals , e.g. those originated by looking at an image. Hence, the call for experts in machine learning and data science to help decode these signals and play a key role in advancing neuroscience research to bring the next generation of care to patients. Challenge The challenge asked participants to build a learning model capable of accurately predicting the image shown to a human subject based on electric signals in the brain. The model needs to predict whether the person is seeing the image of a house or of a face, following the Electrocorticographic (ECoG) signals collected from a sample of four epileptic patients. That makes it an image recognition neuroscience and data science project. The Grand Prize winner of this competition is Alexandre Barachant , whose model reached an accuracy just below 94%. Actually, his final solution is a blend of 5 different models: 2 were dedicated to detection of evoked potential and 3 to induced activity. Among the general ideas which he followed to build them: • train the models independently on each individual patient; • avoid any preprocessing of the data and directly feed the raw signal to the models. These and other principles are carefully explained by the author of the winning entry and presented with his code here . Those interested can also use the solutions proposed by the runner-ups here and here . We are Challenge Brain signal data gathering: from Microsoft's Channel 9 video introducing the Decoding Brain Signals competition. 28 Computer Vision News Challenge Decoding Brain Signals How does our brain interpret electric signals?