Beruflich Dokumente
Kultur Dokumente
Technologies for
Next-Generation
Cognitive Computing
Research Almaden
Geoffrey W. Burr
October 11, 2016
1) Quantum computing
much more sensitive to noise (need low temperatures)
but much more functionality PER device (qubit)
G. W. Burr
IBM Research Almaden
human
brain
PRO:
can follow gradient descent thru backpropagation
convergence to good solutions
Machine Intelligence
flexible systems that continuously learn from
unlabeled data, and that perform (motor)
actions, predict consequences of those
actions, and then plan ahead to reach goals
PRO:
algorithm is scalable:
more resources better performance
CON:
were sure the brain doesnt do backpropagation
Brain-inspired computing
CON:
G. W. Burr
IBM Research Almaden
(images,
raw speech data, etc.)
2) classification results
compared to labels
3) corrections
backpropagated
& all weights updated
Combine
100-1000
input vectors
into an
input matrix
(mini-batch)
multiply by current
weight matrix,
excitation
into next
hidden
neurons
G. W. Burr
IBM Research Almaden
x2A
wij
S xiA wij
B
x528
G. W. Burr
IBM Research Almaden
Selector device
x1A
Conductance
pairs
x2A
N1
wij
wij
x528
N2
I=G+ V(t)
S xiA wij
A
xi
I= S
Neuromorphic Technologies for
Next-Generation Cognitive Computing
Nn
G+ V
I=G- V(t)
G. W. Burr
IBM Research Almaden
+M
I= S G- V
October 11, 2016
NVM-for-on-chip-learning
4096 cores
Backpropagation:
local learning rule within a global architecture
correct
answer
x1A
x1B
d1 = y1
g1
d2 = y2
g2
x2A
wij
xj B
NN results
B
x250
A
x528
dk = yk
gk
Dwij = h * xi * dj
learning rate
G. W. Burr
IBM Research Almaden
x1A
x1B
d1 = y1
g1
d2 = y2
g2
x2A
wij
xj B
??NN results
B
x250
A
x528
dk = yk
gk
Change in
Synaptic
Conductance
DG
G. W. Burr
IBM Research Almaden
OUTPUTS:
Predictions
Context
Stable Concepts (SDR)
Motor commands
INPUT:
Spatial-temporal data
streams of any kind
Context-Aware Learning
winfriedwilcke@us.ibm.com
10
G. W. Burr
IBM Research Almaden
CPU
Memory
BUS
Sell it to
LOTS of people
for vastly
different
purposes
Design 1 piece
of hardware
3) Modularity of design
11
Specifications
Specifications
Requirements
Requirements
G. W. Burr
IBM Research Almaden
Big data
sets
www.datasciencecentral.com/profiles/blogs/concise-visual-summary-of-deep-learning-architectures
12
Specifications
Specifications
Requirements
Requirements
G. W. Burr
IBM Research Almaden
On-chip learning
13
Machine Intelligence
flexible systems that continuously learn from
unlabeled data, and that perform (motor)
actions, predict consequences of those
actions, and then plan ahead to reach goals
Brain-inspired computing
For helping to understand the brain
For computation
STDP (e.g., spikes for learning)
need a global architecture
HTM/CAL (e.g., based on
Sparse Distributed Representations)
Need to show
generalization
scalability
(at least one) killer app
G. W. Burr
IBM Research Almaden