Brain decoders use neural recordings to infer the activity or intent of a user. To train a decoder, one generally needs to infer the measured variables of interest (covariates) from simultaneously measured neural activity. However, there are cases for which obtaining supervised data is difficult or impossible. Here, we describe an approach for movement decoding that does not require access to simultaneously measured neural activity and motor outputs. We use the statistics of movement—much like cryptographers use the statistics of language—to find a mapping between neural activity and motor variables, and then align the distribution of decoder outputs with the typical distribution of motor outputs by minimizing their Kullback–Leibler divergence. By using datasets collected from the motor cortex of three non-human primates performing either a reaching task or an isometric force-production task, we show that the performance of such a distribution-alignment decoding algorithm is comparable to the performance of supervised approaches. Distribution-alignment decoding promises to broaden the set of potential applications of brain decoding.

大脑解码器使用神经记录来推断用户的活动或意图。为了训练解码器，人们通常需要从同时测量的神经活动中推断出感兴趣的测量变量（协变量）。然而，有些情况下很难或不可能获得监督数据。在这里，我们描述了一种不需要访问同时测量的神经活动和运动输出的运动解码方法。我们使用运动的统计数据——就像密码学家使用语言的统计数据一样——找到神经活动和运动变量之间的映射，然后通过最小化它们的 Kullback-Leibler 散度来将解码器输出的分布与运动输出的典型分布对齐。通过使用从执行到达任务或等距力产生任务的三个非人类灵长类动物的运动皮层收集的数据集，我们表明这种分布对齐解码算法的性能与监督方法的性能相当。分布对齐解码有望扩大大脑解码的潜在应用范围。