Visual attention neuroprosthesis in marmoset to elucidate midbrain mechanisms of bottom-up attention

今年的主要成果是分析来自对照组和人类婴儿的自由观看注视数据（自上而下输入较少的模型），实现更好的预测模型以取代缺失的视觉皮层活动，以及设计和规划上丘硅探针植入。有趣的是，与我们的预测相反，12个月大的婴儿与狨猴的长相模式似乎没有什么不同（因此，使用12个月大的婴儿作为与狨猴进行比较的零模型是站住脚的）。在建模方面，我实现了目标预测（即显著性地图）模型的注视点和以凝视为中心的版本。由于低估了实时执行的难度，这花费了比预期更长的时间。这些都可以在github上免费获得。最后，为了实现更好的吞吐量和数据管理，我实现了一个管道来分析凝视数据和显著性（包括扫视和其他眼动检测），它使用关系数据库而不是以前的分布式文件格式，以节省空间和计算。我提交了一篇大型综述论文，以及其他几篇关于婴儿数据的原创论文，以及狨猴的免费观察数据。

This year, primary achievements were analysis of free viewing gaze data from control subjects and from human infants (a model with less top-down input), implementation of better predictive models for replacing missing visual cortex activity, and design and planning of the superior colliculus silicon probe implantations. Interestingly, infants past 12 months do not appear to have different looking patterns than marmosets, in contrast to our predictions (and thus using 12 month old infants as a null model against marmosets for comparison is tenable).On the modeling side, I implemented foveation and gaze-centered versions of the target prediction (i.e. saliency map) models. This took longer than expected due to underestimation of their difficulty for real-time implementation. These are available freely on github.Finally, to enable better throughput and administration of data, I implemented a pipeline to analyze gaze data and saliency (including saccade and other eye movement detections) which uses relational databases instead of the previous distributed file format, to save space and computation. I submitted a large review paper as well as several other original papers regarding the infant data as well as free viewing data from marmoset.