August 29, 1997, 2:14am ET: Skynet gains consciousness August 29, - - PowerPoint PPT Presentation

▶
august 29 1997 2 14am et skynet gains consciousness
SMART_READER_LITE
LIVE PREVIEW

August 29, 1997, 2:14am ET: Skynet gains consciousness August 29, - - PowerPoint PPT Presentation

Oct 04, 2022 29 likes •664 views

August 29, 1997, 2:14am ET: Skynet gains consciousness August 29, 1997: Judgement Day What features do JARVIS and MK-IV have? What features do JARVIS and MK-IV have? Vision/Perception Speech Motion/Planning Object

slide-1
SLIDE 1
slide-2
SLIDE 2

August 29, 1997, 2:14am ET: Skynet gains ‘consciousness’

slide-3
SLIDE 3

August 29, 1997: Judgement Day

slide-4
SLIDE 4
slide-5
SLIDE 5
slide-6
SLIDE 6
slide-7
SLIDE 7
slide-8
SLIDE 8
slide-9
SLIDE 9
slide-10
SLIDE 10
slide-11
SLIDE 11
slide-12
SLIDE 12

What ‘features’ do JARVIS and MK-IV have?

slide-13
SLIDE 13

What ‘features’ do JARVIS and MK-IV have?

  • Vision/Perception
  • Speech
  • Motion/Planning
  • Object Manipulation
  • Goal-directed Behavior
  • Firepower / Thrusters
  • Shape Reconfiguration
  • Social Skills?
  • Ethics/Morality?
slide-14
SLIDE 14

Traveling back from MCU to our timeline…

slide-15
SLIDE 15

1960’s (1st Wave)

  • Single Layer networks
  • XOR problem killed research for two decades
slide-16
SLIDE 16

Mid-1980s (2nd Wave)

  • Multi-layer networks
  • Backpropagation algorithm
slide-17
SLIDE 17

2010s (3rd Wave)

  • Big Data

– O(M) labeled images

  • Big Compute
  • ‘Deep’ Learning
slide-18
SLIDE 18
slide-19
SLIDE 19

ACM Turing Award 2018

slide-20
SLIDE 20

ImageNet Architecture

slide-21
SLIDE 21

Deep Learning for Computer Vision

slide-22
SLIDE 22
slide-23
SLIDE 23

Slide Courtesy of Nervana Systems

slide-24
SLIDE 24

Deep Learning for Segmentation

slide-25
SLIDE 25

Deep Learning for Generating Faces

slide-26
SLIDE 26

Deep Learning for Speech

slide-27
SLIDE 27

Google Assistant

slide-28
SLIDE 28

IBM Watson

slide-29
SLIDE 29

IBM Project Debater

slide-30
SLIDE 30

IBM Deep Blue

slide-31
SLIDE 31
slide-32
SLIDE 32
slide-33
SLIDE 33
slide-34
SLIDE 34

DeepMind / Starcraft

slide-35
SLIDE 35

Libratus / Poker

slide-36
SLIDE 36

DARPA Robotics Challenge

slide-37
SLIDE 37

MIT Cheetah

slide-38
SLIDE 38

Boston Dynamics

slide-39
SLIDE 39

Object Manipulation

slide-40
SLIDE 40

Object Manipulation

slide-41
SLIDE 41

DARPA Autonomous Driving Challenge

slide-42
SLIDE 42

Deep Learning for Self-Driving Cars

slide-43
SLIDE 43

Drones

slide-44
SLIDE 44

Early accidents…

slide-45
SLIDE 45

Human Negligence?

slide-46
SLIDE 46

Backlash…

slide-47
SLIDE 47

Can Deep Learning work for Science? Similarities

Tasks

  • Pattern Classification
  • Regression
  • Clustering
  • Feature Learning
  • …

Differences

Unique attributes of Scientific Data

  • Multi-channel / Multi-variate
  • Double precision floating

point

  • Noise and Artifacts
  • Statistics are likely different
slide-48
SLIDE 48

Climate Simulations

  • 48 -
slide-49
SLIDE 49

Challenge: Multi-Variate Data

slide-50
SLIDE 50

Climate Science Tasks

  • 50 -

Liu, et al, ABDA’16 Racah, et al, NIPS’17 Kurth, et al, SC’17 Kurth, et al, SC’18 Prabhat, et al, GMD’20 Racah, et al, NIPS’17

slide-51
SLIDE 51
slide-52
SLIDE 52

Contributors: Thorsten Kurth, Sean Treichler, Joshua Romero, Mayur Mudigonda, Nathan Luehr, Everett Phillips, Ankur Mahesh, Michael Matheson, Jack Deslippe, Massimiliano Fatica, Michael Houston. NERSC, NVIDIA, UC Berkeley, OLCF

slide-53
SLIDE 53

Deep Learning for Analytics

Galaxy shape modeling Oxford Nanopore sequencing ECoG speech decoding Daya Bay event clustering https://www.oreilly.com/ideas/a-look-at-deep-learning-for-science IceCube Neutrino classification LHC Particle tracking

slide-54
SLIDE 54

Deep Learning for Simulations

Mesh-free space-time super-resolution Max Jiang et. al in review arXiv:2005.01463 Enforcing statistical constraints for GANs Jinlong Wu et. al, https://arxiv.org/pdf/1905.06841.pdf CosmoGAN Mustafa Mustafa et. al, https://arxiv.org/abs/1706.02390 CaloGAN Michela Paganini et. al, https://arxiv.org/abs/1706.02390 Highly scalable PI-GANs for learning SPDE solutions Liu Yang et. al, https://arxiv.org/pdf/1910.13444.pdf

slide-55
SLIDE 55

Deep Learning Hardware

  • 55 -
slide-56
SLIDE 56

Deep Learning Software

Technologies Deep Learning Frameworks Multi Node libraries Single Node libraries Hardware

GRPC MLSL Horovod Theano, Neon, CNTK, MXNet, … CuDNN MKL-DNN CPUs (KNL) GPUs FPGAs Accelerators MPI Cray Plugin

slide-57
SLIDE 57

Open Challenges

  • Short-Term

– Handling Complex Data – Performance and Scaling – Hyper-parameter optimization – Scarcity of Labeled data

  • Long-Term

– Lack of Theory – Interpretability – Formal Protocol

slide-58
SLIDE 58
slide-59
SLIDE 59

Assumptions…

  • Communities will self-organize and conduct labeling

campaigns

– Active Learning systems can determine optimal strategies for seeking labels

  • Incorporation of domain science principles into

learning algorithms

– Solution spaces that satisfy physical constraints

  • Pattern Classification, Clustering, Anomaly

Detection are solved problems

slide-60
SLIDE 60

What is the role of scientists?

HPSS

/project

Labels Patterns, Clusters, Anomalies Mechanisms, Hypothesis

slide-61
SLIDE 61

How close are we to creating JARVIS and MK-IV?

  • Vision/Perception ! " #
  • Speech " $ %
  • Motion/Planning ! $
  • Object Manipulation &
  • Goal-directed Behavior '
  • Firepower/Thrusters &
  • Shape Reconfiguration '
  • Social Skills (
  • Ethics/Morality )
slide-62
SLIDE 62

Conclusions

  • AI appears to be working

– Genuine breakthroughs in vision, speech, control – Wide range of commercial applications

  • Tremendous potential for scientific applications

– Low-hanging fruit, but hard questions are coming next

  • NERSC is at the forefront of Deep Learning for

Science

– Science Applications, Methods, Software, Hardware – Internships, staff positions – Collaboration opportunities!

slide-63
SLIDE 63

Questions? prabhat@lbl.gov