{"id":202,"date":"2025-04-19T19:29:38","date_gmt":"2025-04-19T19:29:38","guid":{"rendered":"http:\/\/course.ece.cmu.edu\/~ece500\/projects\/s25-teame7\/?p=202"},"modified":"2025-04-19T19:52:13","modified_gmt":"2025-04-19T19:52:13","slug":"olinas-status-report-for-4-19-2025","status":"publish","type":"post","link":"http:\/\/course.ece.cmu.edu\/~ece500\/projects\/s25-teame7\/2025\/04\/19\/olinas-status-report-for-4-19-2025\/","title":{"rendered":"Olina\u2019s Status Report for 4\/19\/2025"},"content":{"rendered":"

This week, I finalized the fix for the gesture recording issue we identified earlier.\u00a0 I re-<\/span>checked<\/span>\u00a0the dataset and re-identified the\u00a0<\/span>top\u00a0<\/span>six gestures with the\u00a0<\/span>top<\/span>\u00a0success rates.\u00a0<\/span>I<\/span>\u00a0<\/span>modified\u00a0<\/span>the CNN model architecture accordingly.<\/p>\n

<\/span>Besides<\/span>, I\u00a0<\/span>scaled<\/span>\u00a0<\/span>up\u00a0<\/span>the dataset to\u00a0<\/span>twice\u00a0<\/span>the<\/span>\u00a0<\/span>size<\/span>,\u00a0<\/span>making<\/span>\u00a0<\/span>the model<\/span>\u00a0<\/span>even<\/span>\u00a0<\/span>more<\/span>\u00a0<\/span>robust.<\/span>\u00a0<\/span>Consequently<\/span>, the wand\u00a0<\/span>can<\/span>\u00a0<\/span>correctly<\/span>\u00a0<\/span>identify<\/span>\u00a0<\/span>all<\/span>\u00a0<\/span>the<\/span>\u00a0<\/span>six target gestures with\u00a0<\/span>much-enhanced<\/span>\u00a0<\/span>performance.
\n<\/span>
\n<\/span>I<\/span>\u00a0<\/span>remain\u00a0<\/span>on\u00a0<\/span>schedule<\/span>\u00a0<\/span>as<\/span>\u00a0<\/span>per\u00a0<\/span>the project timeline.<\/p>\n

<\/span>When<\/span>\u00a0<\/span>developing<\/span>, I\u00a0<\/span>tried<\/span>\u00a0<\/span>out<\/span>\u00a0<\/span>quite<\/span>\u00a0<\/span>a number of\u00a0<\/span>model architectures\u00a0<\/span>with<\/span>\u00a0<\/span>the<\/span>\u00a0<\/span>aim of determining\u00a0<\/span>the\u00a0<\/span>one that\u00a0<\/span>best\u00a0<\/span>fits<\/span>\u00a0<\/span>the<\/span>\u00a0<\/span>task of\u00a0<\/span>classifying\u00a0<\/span>gestures\u00a0<\/span>based\u00a0<\/span>on the IMU<\/span>. I started with a simple RNN, which could capture basic motion patterns but showed poor generalization.<\/span>\u00a0I then\u00a0<\/span>experimented<\/span>\u00a0<\/span>with\u00a0<\/span>an LSTM\u00a0<\/span>architecture<\/span>, which improved\u00a0<\/span>on\u00a0<\/span>training accuracy but\u00a0<\/span>overfitted\u00a0<\/span>quickly\u00a0<\/span>with<\/span>\u00a0<\/span>our small dataset and required long training times. To balance spatial and temporal modeling<\/span>, I\u00a0<\/span>chose<\/span>\u00a0Conv1D + LSTM<\/span>, but it\u00a0<\/span>did<\/span>\u00a0<\/span>not<\/span>\u00a0<\/span>produce better\u00a0<\/span>accuracy<\/span>.\u00a0<\/span>Lastly<\/span>, I tested Conv1D-based CNN with two small-kernel convolutional layers,\u00a0<\/span>flatten<\/span>,<\/span>\u00a0and dropout layers. This model\u00a0<\/span>had<\/span>\u00a0<\/span>the\u00a0<\/span>highest<\/span>\u00a0validation accuracy\u00a0<\/span>throughout\u00a0<\/span>and was within the size\u00a0<\/span>limit<\/span>.<\/p>\n

<\/span><\/p>\n

I learned new skills in time-series modeling, model optimization, and data augmentation. I watched YouTube tutorials to learn key modeling techniques. I referred to TensorFlow documentation for implementing and tuning. Additionally, I read blog posts and Stack Overflow discussions to troubleshoot overfitting and understand best practices for dropout and kernel sizing. These helped me to quickly iterate and effectively tailor the model to meet our system constraints.<\/p>\n

\u00a0<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"

This week, I finalized the fix for the gesture recording issue we identified earlier.\u00a0 I re-checked\u00a0the dataset and re-identified the\u00a0top\u00a0six gestures with the\u00a0top\u00a0success rates.\u00a0I\u00a0modified\u00a0the CNN model architecture accordingly. Besides, I\u00a0scaled\u00a0up\u00a0the dataset to\u00a0twice\u00a0the\u00a0size,\u00a0making\u00a0the model\u00a0even\u00a0more\u00a0robust.\u00a0Consequently, the wand\u00a0can\u00a0correctly\u00a0identify\u00a0all\u00a0the\u00a0six …<\/p>\n","protected":false},"author":1035,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[2],"tags":[3],"_links":{"self":[{"href":"http:\/\/course.ece.cmu.edu\/~ece500\/projects\/s25-teame7\/wp-json\/wp\/v2\/posts\/202"}],"collection":[{"href":"http:\/\/course.ece.cmu.edu\/~ece500\/projects\/s25-teame7\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/course.ece.cmu.edu\/~ece500\/projects\/s25-teame7\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/course.ece.cmu.edu\/~ece500\/projects\/s25-teame7\/wp-json\/wp\/v2\/users\/1035"}],"replies":[{"embeddable":true,"href":"http:\/\/course.ece.cmu.edu\/~ece500\/projects\/s25-teame7\/wp-json\/wp\/v2\/comments?post=202"}],"version-history":[{"count":3,"href":"http:\/\/course.ece.cmu.edu\/~ece500\/projects\/s25-teame7\/wp-json\/wp\/v2\/posts\/202\/revisions"}],"predecessor-version":[{"id":207,"href":"http:\/\/course.ece.cmu.edu\/~ece500\/projects\/s25-teame7\/wp-json\/wp\/v2\/posts\/202\/revisions\/207"}],"wp:attachment":[{"href":"http:\/\/course.ece.cmu.edu\/~ece500\/projects\/s25-teame7\/wp-json\/wp\/v2\/media?parent=202"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/course.ece.cmu.edu\/~ece500\/projects\/s25-teame7\/wp-json\/wp\/v2\/categories?post=202"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/course.ece.cmu.edu\/~ece500\/projects\/s25-teame7\/wp-json\/wp\/v2\/tags?post=202"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}