From robot to food - 3 AI projects here

Prof Thalmann with Nadine, a humanoid robot that acts and speaks like a human.
Prof Thalmann with Nadine, a humanoid robot that acts and speaks like a human.PHOTO: ALICIA CHAN FOR THE STRAITS TIMES

NADINE

A companion robot designed by Professor Nadia Thalmann, director of the Nanyang Technological University's Institute for Media Innovation.

Nadine is a humanoid robot that acts and speaks like a human. She can answer questions posed to her, remember past conversations and even gauge how a person is feeling by reading their facial gestures.

These are enabled through machine learning, which lets her learn speech patterns, recognise words and correct her database of facial gestures.

Her creator, Prof Thalmann, said: "Learning is just the first step. Our future research is linking learning with memory to build a stronger relationship between robot and human."

One of the ambitious future goals for Nadine is to act as a robotic caregiver for the sick or elderly. For instance, she can stand in as a virtual human presence to help people with Alzheimer's by talking to them and keeping them company.

FOODAI

A machine learning-powered software developed by a team at the Singapore Management University (SMU), led by Dr Steven Hoi, an associate professor of information systems.

FoodAI allows users to snap pictures of various types of food which the software will recognise immediately. It currently has a database of more than 100 types of cuisine, ranging from chilli crab to duck rice. The team is expanding its database to more than 1,000 different food types and to better differentiate similar-looking dishes, such as pasta from mee goreng.

The food-recognition code will soon be released as an API which allows app developers to integrate it into their apps, such as those which help users track calories.

MALICIOUS URL FILTER

A machine-learning algorithm that SMU's Dr Hoi is working on with the Defence Ministry to identify malicious links in e-mails in real time.

Traditional methods of catching such links rely on software cross-referencing to an established blacklist of URLs known to be malicious. This makes systems potentially vulnerable to new links not in the blacklist.

Machine-learning software can identify the patterns of such links, such as their IP address or s source, and prevent such links from appearing in a user's inbox. Such software can reduce the chances of users clicking on malicious links, and falling prey to cybercrime, by blocking the links with up to 97 per cent accuracy.

Lester Hio

A version of this article appeared in the print edition of The Straits Times on April 20, 2016, with the headline 'From robot to food - 3 AI projects here'. Print Edition | Subscribe