SLIDE 1
Internet of Things Heterogeneous Systems Resource Constraints Higher-Order Intelligence Distributed Intelligence Deep Neural Networks Reinforcement Learning Machine Reasoning
for the Internet of Things Shaowei Lin GTC San Jose May 2017 - - PowerPoint PPT Presentation
for the Internet of Things Shaowei Lin GTC San Jose May 2017 - - PowerPoint PPT Presentation
Artificial General Intelligence for the Internet of Things Shaowei Lin GTC San Jose May 2017 Internet of Things Heterogeneous Systems Resource Constraints Higher-Order Intelligence Distributed Intelligence Deep Neural Networks
SLIDE 2
SLIDE 3
Deep Learning
SLIDE 4
Biologically-inspired multi-layer neural networks Deeper layers learn higher-order features
WHAT IS DEEP LEARNING?
SLIDE 5
Sensor networks form the nervous system of smart cities.
Deep visual cortex Deep learning Sensor networks
DEEP LEARNING FOR SENSOR NETWORKS
SLIDE 6
Original Data Estimates Masked Inputs Features
Exp 1 Exp 2
Joint work with Liangze Wong, Daniel Chen, Huiling Chen (A*STAR)
STRUCTURED MISSING DATA
SLIDE 7
Temperature Illuminance Humidity Fused Representation Measured Illuminance Measured Humidity Measured Temperature Joint work with Wenyu Zhang, Zuozhu Liu, Tony Quek Sensor Node Sensor Node Sensor Node Gateway Server
MULTIMODAL SENSOR FUSION
SLIDE 8
REINFORCEMENT LEARNING
SLIDE 9
How can the network learn to accomplish given tasks and distribute required steps while managing resources efficiently?
SLIDE 10
How can the network learn to accomplish given tasks and distribute required steps while managing resources efficiently?
what is the action space? self-programming machines? machine reasoning? what objective function should we use?
SLIDE 11
Machine Reasoning
SLIDE 12
12
SLIDE 13
RUSSELL’S PARADOX
TYPE THEORY SET THEORY CATEGORIES HOMOTOPY TYPES 1901 2009
SLIDE 14
14
SLIDE 15
CURRY-HOWARD CORRESPONDENCE
TYPE TERM THEOREM PROOF INTENT IMPLEMENTATION SPACE POINT
SLIDE 16
CURRY-HOWARD CORRESPONDENCE
TYPE TERM THEOREM PROOF INTENT IMPLEMENTATION SPACE POINT
SLIDE 17
Traditional Programming Intentional Programming
<<print elements of list>> (<<list with numbers 1 to 10>>)
SLIDE 18
INTENT AS A TYPE
Poor Type System sort: list nat → list nat Rich Type System sort: ∀ ℓ: list nat , ℓ′: list nat sorted ℓ′ ⋀ same_elements ℓ ℓ′}
SLIDE 19
FINDING IMPLEMENTATIONS FOR INTENTS
Context Intent
SLIDE 20
EQUIVALENT PROBLEMS THROUGH PATHS
Context Intent
SLIDE 21
INTUITION FOR PROBLEM-SOLVING
Problems Equivalences (Paths)
SLIDE 22
INTERNET OF THINGS
Functional Plane (Intents) Physical Plane (Implementations)
Publish- Subscribe Protocols Named- Function Networking Linked Data
SLIDE 23
CRITICAL SERVICES
Convert intent into implementation. Compilers. Check that implementation matches intent. Blockchain?
SLIDE 24