Machine Learning and Children with Special needs

Have you ever noticed how kids learn while playing with toys…

By making association of unknown objects with known objects and classifying them into various categories. But when we come across a child with learning disability, he fails to make the same association due to cognitive impairment. These children mostly have one type of learning disability, either they have impairments in social skills like language and communication or they have trouble in reading, writing and math. Although AI based educational tools have been developed to provide an adaptive learning platform, much needs to be done to integrate these children into mainstream and help them lead life independently. 113 million euros (about $132 million) in funding has been granted for the Autism Innovative Medicine Studies-2-Trials, or AIMS-2-Trials, which brings together 48 partners from 14 nations, including academic institutions, pharmaceutical giants and charities.

As per market research report published by Dolcera, project of such large proportions can be efficiently managed over cloud.

Pharma companies exploring cloud computing have reported positive experiences through

• Easier implementation
• More computational transparency
• A clear-cut IP policy

However, AI enabled assistive gadgets for children with special needs can be another key area for technology giants.

This is where Machine Learning based AI chips will play a key role in future.

Machine learning happens in two phases- training and inferencing. In the training phase, models are developed based on known data. In the inferencing phase, the developed models are used on real-world data to find solutions. For a special child to decide whether it is safe to cross the street or not, it would really help if he has an assistive AI gadget that can make the right decision based on the traffic. To make such a real-time decision, fast inferencing with reasonable accuracy is the key.

Training of neural network is data intensive and takes both forward and backward propagation. GPUs have architecture that are more suitable for this high volume data flow. However, inference and action needs more specific processing, so customized AI Chips will be more efficient for something as challenging as assisting a child with Autism. It is difficult to diagnose autistic patterns which evolve over time – something which can be done better on AI chip.

Usually, CPUs solve problems by collecting blocks of data, then running algorithms or logic operations on that information in sequence. With AI systems, computers need to pull huge amounts of data in parallel from various locations and process it quickly.  By using cloud computing and vast datasets, some neural networks function sufficiently well. The more powerful AI systems in development, however, struggle to process complex rapid-fire calculations at speed if using computer processing units (CPUs) which work sequentially.

An alternative technology implemented in Graphcore’s new chip, an intelligence processing unit (IPU), emphasises graph computing with massively parallel, low-precision floating-point computing. It has more than 1,000 processors which communicate with each other to share the complex workload required for machine learning. Graph computing focuses on nodes and networks rather than if-then instructions. Random graphs that can be used to represent random processes like autistic behavior can be possibly implemented on these IPUs.

We need to take this promising development in machine learning and apply it to the issue at hand – assisting children with special needs. It takes tremendous effort for the family members to find something meaningful in the randomness of autistic behavior and trying to help their kid lead a normal life. Fifteen percent of the U.S. population, or one in seven Americans, has some type of learning disability, according to the National Institutes of Health. While significant development has been made in AI enabled Speech processing, assistive gadgets for people with cognitive impairment and learning disabilities are yet to be developed. Its very likely that we may get interesting results from the inference models that will help our understanding of human brain – and pave the path for a more inclusive AI.