New ?°better Siri?± algorithm can calculate risk
Source: Jiaying Shen
Researchers at the MIT Computer Science and Artificial Intelligence Laboratory are working on what they call “a better Siri” �C a virtual personal assistant that can actually calculate the odds of success when it comes to making plans and offer alternative suggestions.
For example, they note that you could tell your smartphone that you would like to drive from your Boston-area home to a hotel in upstate New York, that you would like to have lunch at around 12:30 at an Applebee’s, and that you want the entire trip to take under four hours.
Using new algorithms, the phone could calculate your odds of success, telling you that you only have a 66 percent chance of meeting those criteria �C but that those odds increase to 99 percent if you’re willing to push lunch back by one-half hour or eat at a different restaurant instead.
It will be possible someday, if MIT professor Brian Williams and his colleagues have anything to say about it. Utilizing the same basic framework used by NASA and Woods Hole Oceanographic Institution to plan missions, they are looking to take virtual personal assistants to the next level.
Williams, along with Peng Yu and Cheng Fang, who are graduate students in MIT’s Department of Aeronautics and Astronautics, have developed new software that can allow smartphone users to specify different constraints and reliability thresholds, then have their mobile devices come up with probability models that offer the best possible solution for any given problem.
But what if no solution exists using the specific criteria laid out by the individual? The program then suggests ways in which the planner might alter the problem constraints, suggesting a minor change that could make things work out more effectively. In essence, their software could take your basic plan and adapt it so that it works as well as possible in real-world conditions.
The researchers, who will present their findings this month at a meeting of the Association for the Advancement of Artificial Intelligence (AAAI), report that their software differs from other planning systems in that it can perform risk assessment. As Fang explained, it can be difficult working directly with probabilities, so they added the idea of risk allocation for the mission.
“The time it takes to traverse any mile of a bus route, for instance, can be represented by a probability distribution �C a bell curve, plotting time against probability,” the MIT News Office explained. “Keeping track of all those probabilities and compounding them for every mile of the route would yield a huge computation.”
“But if the system knows in advance that the planner can tolerate a certain amount of failure, it can, in effect, assign that failure to the lowest-probability outcomes in the distributions, lopping off their tails,” they added. “That makes them much easier to deal with mathematically.”
The researchers will be presenting two different papers at the AAAI conference. Williams and student Andrew Wang describe how to evaluate assignments efficiently so that quick solutions to soluble planning problems can be found. Yu and Fang, on the other hand, will focus on finding constraints that prevent a solution to a problem from being found.
Jiaying Shen, a research scientist at Nuance Communications (the firm that developed the voice-recognition technology used by Apple’s Siri), called the research “quite interesting… They had a flurry of papers on chance constraints, but in the recent papers, they added uncertainty in there.”
Doing so “makes the problems that it can model more complicated and unpredictable and more realistic,” Shen added, noting that she and her colleagues at Nuance “are very interested in constraint relaxation, including the chance constraints. If you expose what you need to consider in the planning stage, then you have a much higher success rate in carrying out the plan.”
| }
|