Getting your Trinity Audio player ready...
|
Statisticians at the National University of Singapore (NUS) have unveiled a groundbreaking method that promises to revolutionise data analysis, particularly in fields like single-cell RNA sequencing, by effectively capturing and analysing intricate data patterns with unprecedented accuracy and efficiency.
Led by Associate Professor Yao Zhigang and Research Fellow Su Jiaji from NUS’s Department of Statistics and Data Science, the team introduced a cutting-edge approach for estimating low-dimensional manifolds within high-dimensional data, leveraging deep Generative Adversarial Networks (GANs) to achieve unparalleled accuracy and computational efficiency.
Collaborating with Professor Shing-Tung Yau from the Yau Mathematical Sciences Centre (YMSC) at Tsinghua University, this work builds upon Prof. Yao’s collaboration with Prof. Yau during his sabbatical visit to the Centre of Mathematical Sciences and Applications (CMSA) at Harvard University.
Conventional data analysis techniques often rely on linear dependencies among features, failing to adequately capture the intricate patterns present in high-dimensional data that are often situated near low-dimensional manifolds.
To address this limitation, manifold-learning methods have emerged as a promising alternative, yet existing approaches like manifold embedding and denoising have been hindered by a lack of detailed geometric understanding and robust theoretical foundations.
Professor Yao and Su forged an innovative method for accurately estimating low-dimensional manifolds concealed within high-dimensional data. This method not only achieves state-of-the-art accuracy in estimation and convergence rates but also improves computational efficiency by leveraging deep Generative Adversarial Networks (GANs).
Overcoming the limitations of conventional data analysis methods, the new technique focuses on accurately fitting manifolds to reduce data dimensionality while preserving crucial information, thus enabling a more comprehensive understanding of complex data structures across diverse fields such as genomics, social media analysis, and IoT sensor networks.
Assoc Prof Yao emphasised the significance of this advancement, stating, “By accurately fitting manifolds, we can reduce data dimensionality while preserving crucial information, including the underlying geometric structure. This represents a major leap in data analysis, enhancing both accuracy and efficiency.”
The manifold fitting method holds promise for applications in RNA sequencing and biodata analysis, particularly in refining clustering accuracy and enhancing data visualisation in single-cell RNA sequencing research. Additionally, its integration with deep learning could revolutionise disease prediction models, offering insights into complex neurological conditions and advancing the field of personalised medicine.
Moving forward, Assoc Prof Yao’s team is actively refining the framework to process even more complex data, collaborating closely with researchers at Tsinghua University to push the boundaries of data analysis and unlock new possibilities for scientific discovery.
As NUS statisticians unveil this new method for revolutionising data analysis, at NTU’s Lee Kong Chian School of Medicine, Assistant Professor Hiroshi Makino’s research sheds light on the brain’s ability to integrate learned skills, forging a significant connection between advancements in artificial intelligence and our understanding of cognitive processes.
By training mice in behavioural experiments and observing neural activity, Makino uncovers a mechanism where the brain combines representations of pre-learned action values from constituent subtasks to accomplish composite tasks.
This insight not only enhances understanding of cognitive functions but also holds implications for improving AI models, as it highlights parallels between deep reinforcement learning algorithms and biological learning mechanisms.
The study’s incorporation of theoretical predictions from deep reinforcement learning provides a framework for understanding the brain’s learning process observed in mice. Assistant Professor Makino’s research showcases a convergence between artificial and biological systems, as theoretical predictions are validated through empirical testing on mice.
This work not only unravels the mysteries of how the brain combines learned skills but also establishes a crucial connection between artificial intelligence models and biological systems, fostering further exploration into the mechanisms that promote exploration for future learning in both neuroscience and AI.