Digital assistants are getting increasingly powerful, but so is the infrastructure required to run them. For the most part, Alexa, Google Assistant, and Siri require an internet connection to function, which is no fun if your service is spotty.
However, researchers at Amazon appear to be making significant leaps in Alexa’s offline functionality. According to a research paper, the team was able to craft offline music playback, temperature control, and navigation algorithms.
The significance here is that the AI’s natural language processing usually utilizes huge rooms of servers to function. With them, it can return answers in milliseconds, perform complex computations, and reference location data.
Perfect Hashing and Smart Vehicles
Through the use of perfect hashing, the researchers were able to reduce offline memory usage by 14-times with barely any impact on speed or accuracy. According to Amazon’s Grant Stimmel, the technique improves on traditional hashing in a number of ways.
“Traditional hash functions sometimes produce collisions. We address this problem by using a more advanced hashing technique called perfect hashing, which maps a specific number of data items to the same number of memory slots but guarantees there will be no collisions,” he explains. “With perfect hashing, the system can simply hash a string of characters and pull up the corresponding weights — no metadata required.”
The breakthrough is particularly important when you consider that Alexa is beginning to be implemented in cars and other mobile devices. Though 4G and even 5G is slowly getting more coverage, there are still many areas that will be underserved for some time.
If you’re curious about the methods, you can read the paper here or attend the Interspeech 2018 conference where they will present the findings.