Increased regularization does not necessarily decrease the number of support vectors

Although neural networks seem to be getting all the attention in machine learning these days, support vector machines (SVMs) are still quite useful in some situations. For example, if you want a classifier to work on a wearable/implantable device that consumes very little energy, SVMs-on-a-chip are currently probably your best option. If you are making … More Increased regularization does not necessarily decrease the number of support vectors

Why the Kaggle 2016 Seizure Prediction Competition was unrealistically difficult

One of the most difficult parts of living with epilepsy is that seizures are often unpredictable and occur without warning. For decades, the epilepsy research community has dreamed of developing technology that could warn individuals that a seizure was likely so that they could take action to prevent the seizure or mitigate its effects. A … More Why the Kaggle 2016 Seizure Prediction Competition was unrealistically difficult

How to get MNI coordinates for Epilepsiae electrodes

One of the best databases for focal epilepsy research is the Epilepsiae database of over 100 annotated datasets of inpatient intracranial and scalp electrode recordings from individuals suffering from epilepsy. It is a fantastic resource. However, it is not clear how to interpret the intracranial electrode coordinates given the current state of the Epilepsiae documentation. … More How to get MNI coordinates for Epilepsiae electrodes

Linear spatial filters (e.g., ICA, PCA) cannot perfectly extract EEG/MEG sources

Linear spatial filters (LSFs) derived from techniques like independent component analysis (ICA) or principal components analysis (PCA) are extremely popular techniques for analyzing electroencephalographic (EEG) and magnetoencephalographic data. They work by exploiting the fact that different EEG/MEG signal sources (e.g., a cortical patch or the eyes) project across the scalp with different topographies. An LSF … More Linear spatial filters (e.g., ICA, PCA) cannot perfectly extract EEG/MEG sources

A decline in neurosurgeon-scientists means a decline in neuro-innovation

A recent Nature editorial summarizes disturbing evidence that surgeons are becoming less involved in research. For example, from 2006-2014 the proportion of NIH funding to surgical departments at the top 25 academic medical centers fell from 3% to 2.3% and a 2016 survey of 1000 academic surgeons found that most surgeons thought that they did not … More A decline in neurosurgeon-scientists means a decline in neuro-innovation

Einstein’s EEG

I’ve been reading through neurological atlases of the intracranial electroencephalogram to research the frequencies of oscillations that are characteristic of different brain areas in health and disease.  Today I was able to exhume from a library Epilepsy and the Functional Anatomy of the Human Brain, a 1954 book by the pioneering Wilder Penfield and Herbert … More Einstein’s EEG

Public clinical data science data sets

Looking for a clinical data science project? Here are some free data sets that might be of use (in no particular order): Human Physiological and Neuroimaging Data Cam-CAN (Cambridge Centre for Ageing Neuroscience) dataset: fMRI, MRI, MEG, and behavioral human neurodata on people of various ages (i.e., cross-sectional data) Physionet: A wide array of clinically relevant … More Public clinical data science data sets