News

NRF Fellowship - Dr Zhang Mengmi

Congratulations to Dr Zhang Mengmi, Scientist from A*STAR’s Centre for Frontier AI Research (CFAR) for receiving the National Research Foundation (NRF) Fellowship. The NRF Fellowship aims to develop a strong science research base in Singapore and provides opportunities for early career researchers to carry out independent research in Singapore, over a five-year period.

Dr Zhang’s work titled “What AI cannot do but humans can: Closing gaps in visual search efficiency between AIs and humans with neuroscience-inspired approaches” aims to enable AIs to do what humans can do in cognitive tasks by incorporating memory, visual attention, and reasoning abilities to existing AI frameworks with neuroscience-inspired approaches. 

Visual search is a common daily visual activity that we engage in and is exemplified by looking for a friend in a crowd. During the search, we examine the entire scene efficiently with a sequence of eye movements. While it may seem effortless to humans, a lack of prior contextual knowledge and memory guidance makes artificial intelligence (AI) inferior to humans in visual search. Dr Zhang’s research would thus address the following question: What are the memory computations in the human brain that modulate eye movements to ensure a time-efficient and transformation-invariant visual search in complex real-world scenes? 

Conducted by Dr Zhang and her team at the Deep NeuroCognition lab, this research is at the intersection of AI and neuroscience. Apart from conducting human behavioural experiments, the team also develops biologically inspired computational models for cognitive functions, such as visual attention, memory, knowledge, learning, and reasoning.

The successes of the team’s approaches would add the following synergies to the studies of biological and artificial intelligence:


  1. Modern AI will become better at performing visual search tasks with human-level search efficiency and robustness, closing the gap between biological and artificial intelligence. These AI models can be deployed for real-world applications, such as assistive robots, medical imaging, and surveillance.
  2. The scientific deliverables would be beneficial in advancing our understanding of high-level cognitive functions in biological brains, including memory and learning.
  3. The curated datasets and evaluation techniques will add to de facto standards for the community to benchmark biological and artificial intelligence.