Apple Reveals Machine Learning Systems Applied to Apple Music and Apple Watch Heartbeat Activity Patterns
Today the US Patent & Trademark Office published a patent application from Apple that relates to machine learning. More specifically, Apple's invention covers systems and methods for a server generating a plurality of proxy prediction models for each of a plurality of applications for use by client devices.
Providing a prediction model based on a data distribution that more closely matches the user private data of the actual user of the client device, as opposed to a generic model, can improve the accuracy of predictions and improve the benefits to the user of the client device. The plurality of prediction models are generated by a server and provided to a plurality of clients. For a particular application, a client device can use previously collected private user data to select one of the plurality of prediction models for the application as being a closest matching prediction model to use for a particular user.
In an overview of machine learning, Apple notes that conventional machine learning relies on gathering large data sets and training prediction models from a large sample of users. The prediction model is trained to achieve the highest possible prediction accuracy on the training dataset, but will often need to be personalized for individual users by taking into account a particular user's personal data.
An example would be a handwriting recognition algorithm that is adapted to a particular user's style of writing or a health sensor algorithm that is adapted, or calibrated, to a particular user's heartbeat or activity pattern.
This is focus that Apple places in their patent claims with machine learning used in conjunction with a sensor. Moe specifically Claim #4 states: "wherein the one or more sensors comprise a health sensor that includes a heart-rate monitor."
Apple further notes that "A speech recognition subsystem may need to be tailored to a particular user's speech patterns or inflections." This is also a focus in Apple's patent claims.
Apple's Claim 6 states: "The method of claim 1, wherein the user predictive model calibrates a speech recognition module to the user, wherein the user predictive model incorporates feature vectors including components comprising a language in which the user is speaking, mel-frequency cepstral coefficients, and activations of deep neural networks."
In Apple's patent claim 7 they continue: "wherein the user predictive model predicts one of: a subject matter of text that a user is composing or editing in the application, and the user predictive model suggests words or phrases to the user in the application; or a genre or artist of media type to present to the user, and the model suggests or plays media items in the application."
Apple further noted that "a music suggestion subsystem is preferably tailored to the particular user's tastes, preferences, and listening patterns." This is likely used in Apple Music and likely to be used with Apple's HomePod.
Apple goes on to explain that in prior art, machine learning that is used to enhance a particular user's experience can be trained on the client device using the particular user's data only. Training using only on-device data preserves privacy at the cost of training accuracy and time. A machine learning algorithm typically needs to collect a substantial amount of data before machine learning can generate reasonably accurate prediction models. In some cases, such as health sensors that measure one or more attributes of a user's physical body, the sensor may need to be calibrated to that person before the data collected can be used by a machine learning algorithm to generate predictors on the client device for that user. Calibration of a sensor to a particular human can be intrusive, very personal, and time consuming and thus an undesirable step that a user may not wish to take.
Another way to generate a machine learning prediction model that can be used to enhance a user's on-device experience can include a server system collecting a substantial amount of crowd-sourced data that can be used to generate one or more prediction models. This method of training predictors collects a user's personal data and may violate privacy, even if the information collected is de-identified. If privacy is maintained, then another problem with training predictors using crowd-sourced data is that a server does not know which prediction model, among a plurality of predictive models for, e.g. music preferences, a particular client should use. Maintaining privacy means ensuring that the server cannot identify a particular client with any specificity. If the server cannot identify a particular client with specificity, e.g. due to privacy constraints, then the server cannot provide a personalized prediction model to the client device to enhance the user's on-device experience.
Apple's patent covers systems and methods for enhancing a user's experience when using a client device by providing a plurality of prediction models for each of a plurality of applications on the client device. Providing a prediction model that more closely matches the actual user of the client device can improve the speed and accuracy of machine learning of a prediction model that represents the user of the client device.
The invention will be used on Apple's full range of devices from the Mac to iDevices including Apple TV and "entertainment systems" that could be referring to Apple's future entertainment content service.
Apple's patent FIG. 2 below illustrates, in block form, a detailed view of a system for learning proxy models used by clients while preserving client privacy.
Our cover graphic covers patent FIG. 1 which illustrates, in block form, an overview of a system for learning a proxy models used by clients while preserving client privacy.
Apple's patent application was filed back in Q3 2016. Apple's inventors include Jared Zerbe, Apple's Silicon Systems Architect and two others that have since moved on to create a company called Kraenion which is an applied machine learning company developing a visual cortex (a "brain in a box") for machines that interact with the physical world.
The other inventors include: Julia Ng who is now the Co-Founder and Chief Scientist at Kraenion. At Apple she was a Research Scientist; and Binu Mathew who is now Chief Technology Officer at Kraenion. At Apple Matthew was a Senior Scientist up until July 2017
Patently Apple presents a detailed summary of patent applications with associated graphics for journalistic news purposes as each such patent application is revealed by the U.S. Patent & Trade Office. Readers are cautioned that the full text of any patent application should be read in its entirety for full and accurate details. About Making Comments on our Site: Patently Apple reserves the right to post, dismiss or edit any comments. Those using abusive language or negative behavior will result in being blacklisted on Disqus.