Google Granted Patent on Mobile Machine Learning

That phone in your pocket is filled with applications, with sensors to measure movement and the world around us, with communications tools that put us in touch with work, home, family, friends, service providers and strangers.

That phone in your pocket is poised to teach itself how to work better, based upon how you use it, which applications you run, and how you use it to communicate with others.

A patent granted to Google last week explores different ways that parts and pieces of your phone can communicate with each other to remember settings in different contexts, to re-rank information based upon location and time and place, under a mobile machine learning system.

Imagine, for instance, landing at San Francisco International Airport to visit your brother. As you step off the plane, your phone resets its location and displays time and weather information on its home page for San Francisco. You open your phone, and the number for your limo appears at the top, with your hotel next, and then your brother’s home number (it would show his work number if it were earlier in the day).

You had booked a room and limo online, added the trip to your calendar, sent a few emails and made a phone call to your brother while planning the trip. Your phone recognized that you had arrived after you turned your phone back on, and moved (or re-ranked) phone numbers to the top of your contacts based upon your need for them.

The patent is:

Native machine learning service for user adaptation on a mobile platform
Inventors: Hrishikesh Aradhye, Wei Hua, and Ruei-sung Lin
Assignee: Google
United States Patent 8,429,103
Granted April 23, 2013
Filed: August 2, 2012

Abstract

Disclosed are apparatus and methods for providing machine-learning services. A machine-learning service executing on a mobile platform can receive data related to a plurality of features.

In some cases, the received data can include data related to features received from an application and data related to features received from the mobile platform. The machine-learning service can determine at least one feature based on the received data. The machine-learning service can generate an output by performing a machine-learning operation on the at least one feature.

The machine-learning operation can be selected from among an operation of ranking the at least one feature, an operation of classifying the at least one feature, an operation of predicting the at least one feature, and an operation of clustering the at least one feature. The machine-learning service can send the output.

The patent provides a fairly deep look at features that might make up a machine learning system on a phone, from an Application Program Interface (API) that allows communication with software applications on your phone, to the changes to settings that you might make (changes in volume, selection of applications, how you listen to media, how images might be tagged, and more.)

For instance, you decide to take some pictures on your San Francisco trip, and snap a few of your brother. The phone suggests some labels from the picture that include your brother’s name, and your location, using facial recognition to understand the picture is of your brother and GPS to know where the picture was taken.

You turn on a song application, and your music player suggests a playlist based upon the estimated travel time to your hotel, and your previous listening habits.
Some other example uses of the machine-learning service might include:

  • Predicting a duration of a mobile session before the mobile session starts, based on location, time, calendar entries, prior behavior, etc.
  • Predicting a phone number to be dialed at the onset of utilizing a phone dialing application, based on location, time, calendar entries, prior behavior, etc.
  • Predicting speaker and/or mute settings for the mobile platform, based on location, time, calendar entries, prior behavior, etc.
  • Classifying locations based on their ability to use Wi-Fi and/or other communication networks.
  • Generating example photo names and photo album names for a camera application utilizing the mobile platform.
  • Many other examples are possible as well

The point behind such a learning system is to make phones (and their many apps) easier to use, more efficient for users, and enable people to save time and get the most of the applications that they have.

Context Signals in a Mobile Machine Learning System

The patent tells us that a machine learning system from a smart phone might look at a lot of different types of signals when working to make it easier to use a phone. One area that plays a significant role is the context of situations. The patent included a list of examples of such signals:

  • Current time,
  • Current date,
  • Current day of the week,
  • Current month,
  • Current season,
  • A time of a future event or future user-context,
  • A date of a future event or future user-based context,
  • A day of the week of a future event or future context,
  • A month of a future event or future user-context,
  • A season of a future event or future user-based context,
  • A time of a past event or past user-based context,
  • A date of a past event or past user-based context,
  • A day of the week of a past event or past user-based context,
  • A month of a past event or past user-based context,
  • A season of a past event or past user-based context, ambient temperature near the user (or near a monitoring device associated with a user),
  • A current, future, and/or past weather forecast at or near a current location, possibly based on the location of the mobile platform,
  • A current, future, and/or past weather forecast at or near a location of a planned event in which a user and/or a user’s friends plan to participate,
  • A current, future, and/or past weather forecast at or near a location of a previous event in which a user and/or a user’s friends participated,
  • Information on user’s calendar, such as information regarding events or statuses of a user or a user’s friends,
  • Information accessible via a user’s social networking account, such as information relating a user’s status, statuses of a user’s friends in a social network group, a user’s relationship with the user’s friends, and/or communications between the user and the user’s friends,
  • Noise level or any recognizable sounds detected by the mobile platform and/or a monitoring device,
  • Items that are currently detected by the mobile platform and/or a monitoring device,
  • Items that have been detected in the past by the monitoring device,
  • Items that other devices associated with a monitoring device (e.g., a “trusted” monitoring device) are currently monitoring or have monitored in the past,
  • Information derived from cross-referencing any two or more of: information on a user’s calendar, information available via a user’s social networking account, and/or other context signals or sources of context information,
  • Health statistics or characterizations of a user’s current health (e.g., whether a user has a fever or whether a user just woke up from being asleep),
  • A user’s recent context as determined from sensors on or near the user and/or other sources of context information (e.g., whether the user is walking, running, and/or jogging, among other possibilities),
  • A current location of the user and/or the mobile platform,
  • A past location of the user and/or the mobile platform, and
  • A future location of the user and/or the mobile platform.

Where from Here?

This is a long and very detailed patent (more than 70 pages when pasted into Word) from Google. I was tempted to try to summarize, and this is a summary of my summary. :) I’ve had a few posts planned from before I saw this patent, and thought that it would be a good introduction to what those posts are going to be about.

Machine learning for a phone isn’t search or SEO, but given that most people will be accessing search engines and online advertisements on their phones, and given the role of Google and Apple in both local search and mobile devices, it’s an area that people in SEO can’t ignore and expect to be able to provide SEO services in the future. And it can be argued (fairly easily) that search itself is turning into a machine learning system, recommending pages for you based upon your context (language, time, day, season, location), your use of social networks, and personalization.

Google’s predictive search application, Google Now, will be bringing us information that we need before we even search for it. Predictive algorithms in use in areas like machine learning for mobile are going to be a part of that.

I broke down aspects of how Google Now works in terms of self learning and predictive algorithms in the post, Why Google’s Predictive Personal Assistant is better than Siri.

That post didn’t cover aspects of how “machine learning” might take place in detail the way that this patent does, but they are different pieces of the same puzzle.

Google has made some recent acquisitions lately that may also play a big role in the evolution of how we gather information, and how search engines gather information as well. I’ll be covering those in some posts in the very near future.

Machine learning for mobile devices and the kinds of predictive algorithms they produce will have a profound impact upon local and organic search, upon sponsored advertisements, and upon how we use both phones and other wearable computing devices like Google Glass and smart watches.

Share

16 thoughts on “Google Granted Patent on Mobile Machine Learning”

  1. Nice post Bill,

    What is find interesting is how Google was able to capture this patent. Since there are numerous smart phone manufacturers promising this type of functionality it seems odd that this patent would be granted. It makes me wonder how Apple viewed this patent and whether it was considered a threat to product development.

    Is it me or does this seem like a big duh?

    Thanks for sharing.

    Kent

  2. Very interesting (and very creepy) stuff! With Google Now, Google Glass, and the other patents issued it will be interesting to see what the future holds and specifically what these devices can learn from a user. It also is slightly concerning from a privacy standpoint and what may be “too much” learning.

  3. Hi Robert,

    As I mentioned in the post, this is a pretty long patent that covers a lot of the issues involved in machine learning on a mobile device. There’s definitely a serious concern for privacy, but there’s also some discussion about what kinds of information might not be such a big deal from a privacy stance, such as noting that people going into movie theatres often put their phones into silent mode, and a system like this using that information to automatically change that setting on their behalf and not be too concerned about any privacy implications.

  4. Hi Kent,

    The patent isn’t too much like any of the other patents that I’ve read regarding mobile devices and how different aspects to them might communicate and learn. It’s an extremely detailed patent, and I couldn’t imagine most smartphone manufacturers coming out with a patent on the same level as this one.

    Google’s been following up with further development in areas like this – I’m guessing that if anyone at Apple is following along, they might see it as an issue. But Google’s provided something that’s so detailed that it probably wouldn’t stop Apple from doing something along similar lines with a different approach. That’s probably a good thing – the patent protects Google, and provides enough details so that Apple can follow a different path.

  5. Looks like the patent war is growing more and more intense. The lawsuit between Apple and Samsung last year made it pretty clear that things are going to get even uglier, though contextually, the patent awarded to google will definitely accelerate things a bit.

  6. Thanks Bill. A very interesting and thought provoking patent and blog indeed :-)

    It seems to me that since early 2011, Google have been focussing greatly on the areas of localisation and predictive personalisation. Google+ is also clearly an important project – did you see Rand Fishkin’s post on
    increasing Google+ referrals are their shopping centre promotions ? I was particularly interested in the Query-based circles patent you wrote about. Seemly like there will be convergence between social, local and personalisation via Google+.

    So although, in the last couple of years, we have also seen a lot of immediate spam crackdowns and effort to make immediate quality improvements, I have to agree that SEO’s need to have at least one eye on Google’s overall innovation focus. They seem to be getting ahead of the curve again by investing in and persisting with tech areas that that will significantly influence and possibly reinvent search altogether.

    Not quite sure what this will look or feel like – but it definitely merits attention – and I look forward to the coming posts you mention!

  7. This does seem like the evolution of Google NOW, i do think it’s good and that it should be further developed as technology is integrated more and more into our lives. specially with Glasses coming out soon and even further into the future with other types of wearable computing and different types of integration.

    I believe that this type of advancement should be integrated into the android system as most of those features seem good for the dialer and other native apps for it. Maybe with some extensions available via NOW.

    But the true question is, Will google gather and use this information?
    google is already customizing our search results based on our mails, previous searches and such. they display ads that fit our user profiles and with remarketing they can target us by pretty much everything. will this be another way for google to get data on us? will they now know our behavior and further customize our ads by our basic functions and routines?

    Also, will the data be accumulated from many users to define personality types? or maybe even to further understand user intent? they will be able to see patterns in human behavior like who you call when you finish work and further customize your basic machine learning algorithm by giving it “basic aggregated predictions” from other users that google found to match your profile via social, mails, searches and others.

  8. Hi David,

    Google appears to have been working on localization and personalization far longer than just a couple of years ago. Not sure that the Moz article really points to anything that earth shaking, but some of Google’s more recent moves look way past what Google Plus might bring us. :)

    Thanks.

  9. Hi Or,

    Yes, Google is collecting a lot of data, and will be collecting an increasing array of information in the future, and using it in more ways that will make the personalization of the past look like child’s play. Looking at how this might fit into predictive algorithms such as those used Google Now makes a lot of sense.

    It’s likely that much more will be integrated into Android systems, but given the fact that Google Now is going to become available on other platforms, such as IOS, I don’t think it’s going to be limited to Android.

  10. Hi again Bill – Do you think management changes with Larry back as CEO made any noticeable difference in terms of Google’s direction / areas of focus for innovation? I got engaged in SEO as my main area of focus a couple of years back (rather than one of many) so perhaps greater personal focus just leading to skewed conclusions :-)

  11. Hi David,

    It’s hard to say how much of an influence replacing Eric Schmidt with Larry Page has had – I think the things that the founders of Google wanted to pursue would still have been pursued, but Eric Schmidt would have probably followed a somewhat different path to get there. If you haven’t, I’d recommend reading both “In the Plex,” by Steven Levy, and “I’m Feeling Lucky, the Confessions of Google Employee # 59″ which both provide a lot of information about the culture of Google and its history. I still want to learn some more of the gaps that aren’t covered by the two books, but both were pretty interesting.

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>