Presentation is loading. Please wait.

Presentation is loading. Please wait.

A New Theory of Neocortex and Its Implications for Machine Intelligence TTI/Vanguard, All that Data February 9, 2005 Jeff Hawkins Director The Redwood.

Similar presentations


Presentation on theme: "A New Theory of Neocortex and Its Implications for Machine Intelligence TTI/Vanguard, All that Data February 9, 2005 Jeff Hawkins Director The Redwood."— Presentation transcript:

1 A New Theory of Neocortex and Its Implications for Machine Intelligence TTI/Vanguard, All that Data February 9, 2005 Jeff Hawkins Director The Redwood Neuroscience Institute

2 Intelligence Paradigms Artificial Intelligence (AI) 1940s - 1980s - ignores biology - computer programs - emulate human behavior Neural Networks 1970s - 1990s - mostly ignores biology - networks of “neurons” - classify spatial patterns

3 Intelligence Paradigms Artificial Intelligence (AI) 1940s - 1980s - ignores biology - computer programs - emulate human behavior Neural Networks 1970s - 1990s - mostly ignores biology - networks of “neurons” - classify spatial patterns “Real Intelligence”2005 – - biologically derived - hierarchical temporal memory - pattern prediction

4 Hierarchical Temporal Memories (HTMs) A Fundamental technology Automatically discover causes in complex systems Predict future behavior of complex systems Can build super-human intelligence (not C3PO) - faster - more memory - novel senses

5 Agenda Introduction to neocortex What does the neocortex do? How does it do it? Can we express this mathematically? How do we build it? What problems can be solved?

6 Agenda Introduction to neocortex What does the neocortex do? How does it do it? Can we express this mathematically? How do we build it? What problems can be solved?

7

8 Agenda Introduction to neocortex What does the neocortex do? How does it do it? Can we express this mathematically? How do we build it? What problems can be solved?

9 1) The neocortex is a memory system. 2) Through exposure, it builds a model the world. 3) The neocortical memory model predicts future events by analogy to past events.

10

11 Reptilian brain Sophisticated senses Behavior

12 Mammalian brain Reptilian brain Sophisticated senses Behavior Neocortex

13 Human brain Reptilian brain Sophisticated senses Complex behavior Neocortex

14 Agenda Introduction to neocortex What does the neocortex do? How does it do it? Can we express this mathematically? How do we build it? What problems can be solved?

15

16 Hierarchical connectivity

17 touch motor auditionvision spatially specific spatially invariant fast changing slow changing “features” “details” “objects”

18 touch motor auditionvision Prediction

19 touch motor auditionvision Prediction across senses

20 touch motor auditionvision Sensory/motor integration

21 touch motor auditionvision

22 touch motor auditionvision

23 touch motor auditionvision What does each region do? ?

24 touch motor auditionvision What does each region do? Every region: 1) Stores sequences 2) Passes sequence “name” up 3) Predicts next element 4) Converts invariant prediction into specific prediction 5) Passes specific prediction “down” Hierarchical cortex captures hierarchical structure of world - sequences of sequences - structure within structure

25 Unanticipated events rise up the hierarchy until some region can interpret it.

26 Hippocampus is at the top. Novel inputs that cannot be explained as part of known structure automatically rise to the top. HC Unanticipated events rise up the hierarchy until some region can interpret it.

27 Hierarchical Temporal Memories Can Explain Many Psychological Phenomena - Creativity, Intuition, Prejudice - Thought - Consciousness - Learning

28 How does a region work - biology Every region: 1) Stores sequences 2) Passes sequence “name” up 3) Predicts next element 4) Converts invariant prediction into specific prediction 5) Passes specific prediction “down”

29 Agenda Introduction to neocortex What does the neocortex do? How does it do it? Can we express this mathematically? How do we build it? What problems can be solved?

30 All inputs and outputs from a memory region are probability distributions Lower regions Higher regions

31 Learning S A (x t,x t+1,...) S B (x t,x t+1,...) Lower regions C Higher regions C = causes or context S = sequences X = input X P(S|C)

32 Recognition without context S A (x t,x t+1,...) S B (x t,x t+1,...) Lower regions P(C) Higher regions X P(S|C)

33 Recognition with context can lead to new interpretation S A (x t,x t+1,...) S B (x t,x t+1,...) Lower regions C1C1 Higher regions X P(S|C) C1C1

34 Passing a belief down the hierarchy S A (x t,x t+1,...) S B (x t,x t+1,...) Lower regions Higher regions XtXt P(S|C) C f ( X t, P(S|C) ) C

35 Predicting the future S A (x t,x t+1,...) S B (x t,x t+1,...) Lower regions C Higher regions XtXt P(S|C) C f ( X t+1, P(S|C) )

36 Belief Propagation can determine most likely causes of input in a hierarchy of conditional probabilities P(Z1|Y1)P(Z2|Y1)P(Z3|Y1)P(Z4|Y1) P(Y1|X)P(Y2|X) P(X)

37

38 System Architecture 4 pixels Level 1 Level 2 Level 3

39 Recognition : Examples Correctly Recognized“Incorrectly” recognized

40 Correctly Recognized Test Cases

41 Prediction/Filling-in : Example1

42 Prediction/Filling-in : Example2

43 What’s new? Hierarchical Neocognitron HMax Seemore, Visnet Sequence memory auto-associative memories synfire chains Prediction/feedback HMMs ART Sensory/motor integration Biologically derived/constrained/testable

44 Agenda Introduction to neocortex What does the neocortex do? How does it do it? Can we express this mathematically? How do we build it? What problems can be solved?

45 Hierarchical Temporal Memories (HTMs) A Fundamental technology Automatically discover causes in complex systems Predict future behavior of complex systems Can build super-human intelligence (not C3PO) - faster - more memory - novel senses

46 What problems can be solved with HTMs? Traditional AI applications - Vision - Language - Robotics Novel modeling applications - markets - weather - demographics - protein folding - gene interaction - mathematics - physics

47 www.stanford.edu/~dil/invariance/ www.OnIntelligence.org

48 Thank ---

49

50 Learning sequences L5/matrix thalamus/L1 auto-associative loop

51 Creating a sequence “name”

52 Turning an invariant prediction into a specific prediction


Download ppt "A New Theory of Neocortex and Its Implications for Machine Intelligence TTI/Vanguard, All that Data February 9, 2005 Jeff Hawkins Director The Redwood."

Similar presentations


Ads by Google