Liang Tian

DeSide: A unified deep learning approach for cellular decomposition of bulk tumors based on limited scRNA-seq data

DeSide uses limited scRNA-seq data to estimate cell type proportions in bulk tumor samples, outperforming existing methods and enabling patient stratification based on survival rates.

Top-down generation of low-resolution representations improves visual perception and imagination

The paper shows that using low-resolution top-down signals improves visual perception and imagination, and enhances sketch generation in AI by training models on blurred images.

Beyond the Delay Neural Dynamics: a Decoding Strategy for Working Memory Error Reduction

This paper trains RNNs on a color delayed-response task to show that decoding phase strategies reduce memory error by attributing broader neural states to high-probability colors.

Evolutionary learning in the brain by heterosynaptic plasticity

This paper proposes an evolutionary algorithm model of brain learning using non-local rules and demonstrates its effectiveness in training neural networks with biologically plausible weights.