15. 15
put all that descriptive information together
to know what’s in a video
on a second-by-second basis
6
so, we can …
Proprietary and Confidential
16. 16
add also behavioral information
to know how people watch & talk about videos
6
and, we can …
Proprietary and Confidential
Viewership
Channels profiles
Social Media Buzz
Audience Demographics
Audience Contexts
Platform profiles
19. 19
SEARCH vs. EXPLORATION
/ helps only when you know what to search for
/ assumes you understand a topic
/ accuracy driven
/ click-through driven
/ helps when you don’t know what to search for /
/ helps you understand & deepen in a topic /
/ serendipity driven /
/ focused on engagement /
26. Proprietary and Confidential
HUMAN-IN-THE-LOOP AI FOR VIDEO EXPLORATION
/ Machines help to break-down video into granular moments, i.e. shots & scenes
/ Machines generate multitude of paths within and across videos
/ Humans perform simple actions, e.g. watching, following and rating a path
/ Machines generalise from these actions using explicit semantics
/ Machines learn to evolve & improve exploration path
/ Orchestrate a continuous human and machine symbiosis
/ The ultimate aim is to reach a tipping point for video exploration,
e.g. web search, speech recognition
41. MOMENTS REDEFINE CURRENT VIDEO SEARCH
MOBILE/SOCIAL
OPTIMIZED
RELEVANT SEARCH RESULTS
PREVIEW SPECIFIC MOMENTS
OR WATCH FULL VIDEOS
DISCOVER ACTIONABLE
MOMENTS WITHIN VIDEO
42. MOMENTS: PERSONAL VIDEO CHANNEL
SEARCH FOR
ANYTHING, WITH
ANYTHING
HYPERMEDIA PLAYER
W/ LINKED MOMENTS &
FULL VIDEOS
AI ACTIVELY
LEARNS USER
PREFERENCES
DISCOVERS
MORE & MORE
CONTENT
43. AI, DEEP LEARNING & NETWORK EFFECTS
Ensemble algorithmic processing
Video, audio & text
Extract & understand entities
Does what computers are good at
AI & DEEP LEARNING:
//
//
// Human assisted computing
Collective intelligence (incl. fans)
Waze effect
Does what humans are good at
NETWORK EFFECTS:
//
//
//
OBSERVE & LEARN
VERIFY & EXTEND
*U.S. Patent Application #13/863,751
Web-scale layer of structured, linked data.
////
DOMAIN-SPECIFIC DATA
MOMENTS
48. Proprietary and Confidential
HUMAN-IN-THE-LOOP AI FOR VIDEO EXPLORATION
/ Machines help to break-down video into granular moments, i.e. shots & scenes
/ Machines generate multitude of paths within and across videos
/ Humans perform simple actions, e.g. watching, following and rating a path
/ Machines generalise from these actions using explicit semantics
/ Machines learn to evolve & improve exploration path
/ Orchestrate a continuous human and machine symbiosis
/ The ultimate aim is to reach a tipping point for video exploration,
e.g. web search, speech recognition