Thumbnails:
List:
Year:
Category:
Session:
Poster:
Getting poster data...
Kimberly Meechan (EMBL, Heidelberg, Germany)
A deep neural network imagined as an old, industrial machine. Neural networks are vital computational tools used for many steps of biological image analysis e.g. segmenting cells and organelles. Starting from the left side, training data (in this case, electron microscopy images) are loaded into the boiler to power the network. Copper pipes then connect this through various layers representing the architecture of a deep neural network – first through an ‘INPUT’ layer, then layers using a ‘RELU’ activation function. As the network trains, progress is monitored at the control panel in the bottom right. Important values like the training loss, and accuracy are measured, and different parameters adjusted to give the best performance e.g. the learning rate and batch size.