Autonomous Discovery
    the future interface?


                              predictive computing
                            anticipatory computing
                            engineered serendipity
                                                …
Dinesh Vadhia
Founder/engineer at autonomous discovery startup
www.xyggy.com. Previously, senior product management roles
at Oracle and Sun in Silicon Valley and UK, with many years of
software engineering first.

Closely work with Professor Zoubin Ghahramani at Cambridge
University. One of the founding fathers of Bayesian machine
learning methods.

          dinesh@xyggy.com, @DineshVadhia
The Autonomous Age
Probabilistic robots

Commercial unmanned aerial vehicles

Self-driving cars
Autonomous Discovery
Discover new things without doing
anything

Convenience does matter

> Keyword search volume? Probably
Undefined Intent
Search intent defined by keywords

Autonomous discovery has undefined
intent

Window-shopping analogy
New Bayesian ML Algorithm
Deals with things only

Data finds data

Item(s) per query
Feature Engineering
Things defined by features

Can mix multiple feature classes,
including high- & low-level, unstructured
& structured text, ratings, relevance,
context and so on
Dynamic
No training

Dynamic predictions and ranking

Relevance feedback
                           0.85634
                           0.83678
                           0.68321
                           0.59142
Scalability
Sparsity

IR + RecSys

Scalability
Autonomous Discovery


I am the query
I Am the Query
Given data about me, find new things I will
like.

If I like a thing, find more like it, or remove
this thing and others like it.

Hit me with some serendipity.
Quick Look




            technology showcase.
   autonomously discovers games and apps.
          android only. free. ad free.
~300,000 apps, ~500,000 element feature vector.
                    Xyggy
Xyggy App Usage
Autonomously discovers new         Discover by Games or Apps
games and apps.                    categories, and Paid or Free.

Learns what you like.              Touch dice to take a chance
                                   with serendipity
No search box. No keyword
search.                            To install an app you like,
                                   touch its icon to go to the App
If you like a game or app, find    Store.
more (+) like it.
                                   Start-over with restart (o) at
If you don't like an app,          anytime.
remove (-) it and others like it
go away too.
Last But One
Data privacy

Digital doppelganger

Cold-Start Data: Help us spread the word to get more
Xyggy app installs so that collective intelligence kicks-in for even
better results. The app is free and ad-free. Directions to install
the app are at Xyggy. Thank-you!
A Must for ML Practitioners


"A few useful things to know
  about machine learning"
                                               Pedro Domingos
               Department of Computer Science and Engineering
                                      University of Washington

Autonomous Discovery: The New Interface?

  • 1.
    Autonomous Discovery the future interface? predictive computing anticipatory computing engineered serendipity …
  • 2.
    Dinesh Vadhia Founder/engineer atautonomous discovery startup www.xyggy.com. Previously, senior product management roles at Oracle and Sun in Silicon Valley and UK, with many years of software engineering first. Closely work with Professor Zoubin Ghahramani at Cambridge University. One of the founding fathers of Bayesian machine learning methods. dinesh@xyggy.com, @DineshVadhia
  • 3.
    The Autonomous Age Probabilisticrobots Commercial unmanned aerial vehicles Self-driving cars
  • 4.
    Autonomous Discovery Discover newthings without doing anything Convenience does matter > Keyword search volume? Probably
  • 5.
    Undefined Intent Search intentdefined by keywords Autonomous discovery has undefined intent Window-shopping analogy
  • 6.
    New Bayesian MLAlgorithm Deals with things only Data finds data Item(s) per query
  • 7.
    Feature Engineering Things definedby features Can mix multiple feature classes, including high- & low-level, unstructured & structured text, ratings, relevance, context and so on
  • 8.
    Dynamic No training Dynamic predictionsand ranking Relevance feedback 0.85634 0.83678 0.68321 0.59142
  • 9.
  • 10.
  • 11.
    I Am theQuery Given data about me, find new things I will like. If I like a thing, find more like it, or remove this thing and others like it. Hit me with some serendipity.
  • 12.
    Quick Look technology showcase. autonomously discovers games and apps. android only. free. ad free. ~300,000 apps, ~500,000 element feature vector. Xyggy
  • 13.
    Xyggy App Usage Autonomouslydiscovers new Discover by Games or Apps games and apps. categories, and Paid or Free. Learns what you like. Touch dice to take a chance with serendipity No search box. No keyword search. To install an app you like, touch its icon to go to the App If you like a game or app, find Store. more (+) like it. Start-over with restart (o) at If you don't like an app, anytime. remove (-) it and others like it go away too.
  • 18.
    Last But One Dataprivacy Digital doppelganger Cold-Start Data: Help us spread the word to get more Xyggy app installs so that collective intelligence kicks-in for even better results. The app is free and ad-free. Directions to install the app are at Xyggy. Thank-you!
  • 19.
    A Must forML Practitioners "A few useful things to know about machine learning" Pedro Domingos Department of Computer Science and Engineering University of Washington