Building on simple unsupervised matrix factorization techniques, the seqNMF algorithm successfully recovers neural sequences in a wide range of simulated and real datasets.
A new automated and unsupervised algorithm, Risk Assessment Population IDentification, identifies risk-stratifying cells in single cell datasets with robust statistical and biological validation.
A biologically plausible learning rule enables recurrent neural networks to model the way in which neural circuits use supervised learning to perform time-dependent computations.
Model-based analyses of human behaviour and neural activity show that representations of concurrent task-sets emerge by merging together representations of individual stimulus-response associations that occur in temporal proximity.
Emerging evidence suggests a broad role for cerebellar circuits in generating and testing predictions about movement, reward, and diverse cognitive processes.
A general machine learning scheme for integrating time-series data from single-molecule experiments and molecular dynamics simulations is proposed and successfully demonstrated for the folding dynamics of the WW domain.
A new analysis algorithm (DISC) enables accurate analysis of data from high-throughput single-molecule paradigms and reveals a non-cooperative binding mechanism of cyclic nucleotide-binding domains from HCN ion channels.
DeepFly3D, a deep learning-based software, measures limb and appendage movements in tethered, behaving Drosophila and enables precise behavioral measurements during neural recordings, stimulation, and other biological experiments.