"crossing the event horizon" by patrick ehlen, phd and chief scientist at loop ai labs...
TRANSCRIPT
Patrick Ehlen, Ph.D.
Chief Scientist
Loop AI Labs
CROSSING THE EVENT HORIZON
Digital Transformation Strategy
Digital Transformation Strategy
Big Data Moore’s Law
Digital Transformation Strategy
Big Data Moore’s Law
“Dark Data”
Digital Transformation Strategy
Big Data Moore’s Law
“Dark Data”
Event Horizon ?
DIGITAL SPRAWL
DIGITAL SPRAWL• 2001: 5 Exabtyes
DIGITAL SPRAWL• 2001: 5 Exabtyes
• 2011: 5 Exabytesevery 2 days
• 60% CAGR
• IoT / Internet 3.0will expand even more
CORPORATE SOLUTION• Add capacity, leading to unwieldy data centers
• Cost per watt to store & analyze data has become a key issue
• Cost for data analytics staff
• Already coming to the point where we need to do something else
LEARN FROM ANOTHER “BIG DATA” PROBLEM…
LEARN FROM ANOTHER “BIG DATA” PROBLEM…• 100 billion neurons
LEARN FROM ANOTHER “BIG DATA” PROBLEM…• 100 billion neurons
• 16.3 bn cortex
• 69 bn cerebellum
• 100x1012 synapses
THE BRAIN’S BIG DATA PROBLEM• Sensory neurons
• Receptor neurons
• touch, pressure, vibration, limb position,heat, cold, and pain mechanoreceptors, photoreceptors,thermoreceptors, nociceptors
• 31 spinal cord “nerves”
• Firing every ~10ms
THE BRAIN’S BIG DATA PROBLEM• Sensory neurons
• Receptor neurons
• touch, pressure, vibration, limb position,heat, cold, and pain mechanoreceptors, photoreceptors,thermoreceptors, nociceptors
• 31 spinal cord “nerves”
• Firing every ~10ms
• A lot of data!
THE BRAIN’S BIG DATA PROBLEM• Sensory neurons
• Receptor neurons
• touch, pressure, vibration, limb position,heat, cold, and pain mechanoreceptors, photoreceptors,thermoreceptors, nociceptors
• 31 spinal cord “nerves”
• Firing every ~10ms
• A lot of data!
THE BRAIN’S BIG DATA PROBLEM
HOW DOES THE BRAIN DO IT?• Doesn’t “store” information
• Immediate processing
• Plot different types of data into acommon representational space
• Encodes by compressing different types of information into meaningful chunks
• Stores those meaningful chunks, uses them to make predictions
WHAT DO I MEAN BY “MEANINGFUL”?• Compressed chunks of information
• Not raw data
• Pavlov’s dog
RADICAL PROPOSITION• What is the purpose of the brain?
RADICAL PROPOSITION• What is the purpose of the brain?
• …to compress information
• A “compression machine”
• Learns structure
BRAIN “BIG DATA” SOLUTION RECAP• Compression machine
• Fuses information as it comes in
• Stores fused informationin meaningful chunks
• Learns structure
• Associative learning
• …now we have a good basisfrom which to build a “cognitive” platform to handleour own data problems
• A platform based on “human capacity”
COMMON REPRESENTATIONAL SPACE• Prefrontal cortex
• Different types of sensory data
• Some pre-processing
“grandma”
COMMON REPRESENTATIONAL SPACE• Prefrontal cortex
• Different types of sensory data
• Some pre-processing
• Plot into common space
“grandma”
HUMAN CAPACITY COGNITIVE COMPUTING• Take unstructured, raw input from many different types of sensors
• IoT: door, windows, lights, mobile phones
• Plot into a common representational space
• Compress that information, fusing it into meaningful chunks
• Store only those meaningful chunks
• With the correct compression process, we learn patterns and are able to predict
• Because it handles unstructured data this way, we don’t need IoT protocols to tell devices how to communicate with each other
DISCUSSION