Promoters are key DNA regions that control gene transcription, but their activity varies greatly across different cell types. This heterogeneity makes it difficult for existing computational methods to identify promoters reliably. A team led by Professors Zhangyu Mei and Hao Wu from Shandong University, China, has developed MuSE‑Promoter, a deep ensemble learning framework that combines multi‑scale feature fusion, transformer attention, and a learnable weighted ensemble of neural network and random forest. The system outperforms state‑of‑the‑art methods on human cell lines from different tissues and on Arabidopsis thaliana datasets, and shows excellent generalization in cross‑cell‑line and promoter–enhancer discrimination tasks.

Harvard’s Visual Computing Group developed BRIDGE, a simulation system that converts standard standing-basketball footage into realistic wheelchair-basketball videos.

The new “MIGHTY” system rapidly generates travel routes for autonomous robots navigating in uncertain situations, allowing them to react to obstacles in milliseconds while staying on a smooth flight path that minimizes travel time. 

Researchers have introduced a novel canonical correlation guided deep neural network (CCDNN) to improve multi-source data acquisition and fusion. The model is designed to integrate heterogeneous data more effectively, addressing limitations of existing fusion methods in handling diverse and high-dimensional inputs. By leveraging advanced neural network structures, CCDNN enhances feature representation and decision-making accuracy. Experimental results demonstrate superior performance across benchmark tasks, highlighting its potential for applications in intelligent control, automation, and data-driven engineering systems.