The visual cortex of the human brain is very powerful. With just a few glances, it can enable people to quickly process visual information and identify objects – no man-made machine can approach what it can achieve with the same level of accuracy and speed.
Amarjot Singh, founder and CEO of Skylark Labs, has been studying the visual cortex since he started his Ph.D. studies at the University of Cambridge. Prior to this, he studied at Simon Fraser University in Canada and the National Institute of Technology in India. His research is focused on developing a Human-like Learning Regime based on optimal deep learning models that are efficient and effective on data usage, memory utilization, and computational resources.
In 2018, he introduced a computationally efficient hybrid architecture named ‘ScatterNet Hybrid Deep Learning (SHDL)’ framework, which adopted the principles of the human visual cortex to optimize the conventional large deep networks. This hybrid system can learn meaningful information from limited labeled samples and work in a computationally efficient and low-memory way, making it the ideal choice for building practical AI systems.
“Growing up in India made me aware of the several safety and security challenges individuals, especially women, face every day,” says Singh. “This had a big impact on me and served as my motivation to use the artificial intelligence technology (hybrid networks) I developed to solve these security challenges.”
As the CEO of Skylark Labs, Singh led the company to develop a children rescue system called CENSER (ChildrEN SafEty Retrieval) in 2019. It is being used by Guria, a non-profit organization in India dedicated to fighting child prostitution and sex trafficking, to identify and rescue kidnapped children. The evidence collected and identified by CENSER has led to the rescue of several children.
This task is not as straightforward as it sounds. The facial features of children could have significantly changed since being kidnaped, and the images of the child may not be available at the age when he/she was kidnapped. Therefore, the CENSER system has to take these into consideration and implements techniques like age-invariant face recognition and kinship analysis to increase its effectiveness.
Singh also worked with the Red Cross refugee camps in Europe in 2017-2018 to develop a face recognition system that helps Syrian refugee family members reunite. The software managed to reunite several family members during its use over two months, according to Singh.
In both cases, the hybrid framework he developed played a critical role. Its low-memory design empowered the possibility of running such computation intense applications on mobile platforms.
His company, Skylark Labs, is actively involved in a series of projects that applied face recognition technology to address major real-world challenges, aiming to make the world “a safer and secure place." For instance, during the COVID-19 outbreak, the company developed and launched AI Enabled Drones to help the Indian government enforce social distancing and identify rule breakers. Its face recognition technology was widely piloted by banks, airports, schools, and border protection systems to identify suspicious individuals or detect violence.
Recently, Singh’s research has extended to the field of Artificial General Intelligence, developing AI networks that have the ability to learn from their own experiences and are capable of progressively acquiring, fine-tuning, and transferring new knowledge over long time spans. The goal is to build systems that can continuously self-identify and learn new information without human intervention.
“These properties are crucial for designing practical real-world systems that are expected to operate and continuously evolve when needed,” says Singh. He believes that these hybrid architectures may be “the first step” towards constructing true artificial general intelligence machines.