<html>
<head>
<meta http-equiv="Content-Type" content="text/html; charset=iso-8859-1">
<style type="text/css" style="display:none;"> P {margin-top:0;margin-bottom:0;} </style>
</head>
<body dir="ltr">
<div style="font-family: Calibri, Arial, Helvetica, sans-serif; font-size: 12pt; color: rgb(0, 0, 0);">
PhD dissertation defense.<br>
</div>
<div>
<div>
<div dir="ltr">
<div dir="ltr" class="gmail_signature">
<div dir="ltr">
<div><br>
</div>
<div>Sho Nakagome</div>
<div>Advisor: Dr.Jose L. Contreras-Vidal</div>
<div>Department: Electrical and Computer Engineering (ECE)</div>
<div>Title: A NON-INVASIVE BRAIN COMPUTER INTERFACE DECODER FOR GAIT</div>
<div><br>
</div>
<div>Abstract:</div>
<div>Brain Computer Interface (BCI) systems enable control of machines and computers using signals extracted from the brain, such as data recorded using electroencephalography (EEG). Naturally, this technology is expected to help people with disabilities, such
as lost speech or motor impairment, by providing an alternative approach to interact with the world. Being able to walk is one of the most fundamental human functions, and BCIs could help those with walking impairment by providing direct control of an exoskeleton
directly from brain signals. The most crucial part of building such a system is the neural decoding--i.e., the specific algorithm that translates neural signals into movement signals. Developing an effective neural decoding model does not only provide accurate
control of the device, but could also open a new path towards understanding the neural representation of gait. A wide variety of algorithms have been proposed for neural decoding, such as linear regression, kalman filters, and artificial neural networks. However,
there is a lack of rigorous comparisons of different decoding models and parameter choices. Furthermore, it is unclear how well each of these models will generalize to new data from either new environments or different subjects. This dissertation thesis aims
to investigate those issues by: 1) Benchmarking the proposed models and understanding the representation of the brain during gait and 2) Study ways to generalize the model. In the first specific aim, we showed that neural networks not only performed better
than conventional methods when trained within a specific walking environment, but resulted in models that were robust to external disturbances such as channel distortion. In the second aim, we showed intra-subject decoding works in all the combinations (e.g.,
inter-subject decoding of different terrains, level ground walking only, treadmill walking, etc.), but inter-subject decoding only works for electromyography (EMG) to kinematics decoding. To deal with this problem, several methods were used to improve inter-subject
decoding and the transfer learning revealed the most promising results. The work in this dissertation contributes to a greater understanding of the decoding models and their performance/generalizability on non-invasive gait decoding.<br>
</div>
<div><br>
</div>
<div>Zoom info:</div>
<div>Topic: Sho Nakagome dissertation defense's Zoom Meeting<br>
Time: Apr 28, 2020 01:00 PM Central Time (US and Canada)<br>
<br>
Join Zoom Meeting<br>
<a href="https://uofh.zoom.us/j/92816634327?pwd=NklHeUR3eXpobmJabjlrYWNlNUxMZz09">https://uofh.zoom.us/j/92816634327?pwd=NklHeUR3eXpobmJabjlrYWNlNUxMZz09</a><br>
<br>
Meeting ID: 928 1663 4327<br>
Password: 298529<br>
</div>
</div>
</div>
</div>
</div>
</div>
</body>
</html>