acm cikm october 31, 2013 jeff hawkins jhawkins@groksolutions
DESCRIPTION
On-line Learning From Streaming Data. ACM CIKM October 31, 2013 Jeff Hawkins [email protected]. Industrial Research Track. 1) Discover operating principles of neocortex. Anatomy, Physiology. Theoretical principles. Software. Build systems based on these principles. - PowerPoint PPT PresentationTRANSCRIPT
1) Discover operating principles of neocortex
2) Build systems based on these principles
Anatomy,Physiology
Theoreticalprinciples
Software
Industrial Research Track
Anomaly detection in high velocity data
Cortical algorithms
The neocortex is a memory system.
data streamretina
cochlea
somatic
The neocortex learns a model from sensory data
- predictions - anomalies - actions
The neocortex learns a sensory-motor model of the world
Principles of Neocortical Function
retina
cochlea
somatic
1) On-line learning from streaming data
data stream
Principles of Neocortical Function
2) Hierarchy of memory regions
retina
cochlea
somatic
1) On-line learning from streaming data
data stream
Principles of Neocortical Function
2) Hierarchy of memory regions
retina
cochlea
somatic
3) Sequence memory- inference- motordata stream
1) On-line learning from streaming data
Principles of Neocortical Function
4) Sparse Distributed Representations
2) Hierarchy of memory regions
retina
cochlea
somatic
3) Sequence memory
data stream
1) On-line learning from streaming data
Principles of Neocortical Function
retina
cochlea
somatic
data stream
2) Hierarchy of memory regions
3) Sequence memory
5) All regions are sensory and motor
4) Sparse Distributed Representations
Motor
1) On-line learning from streaming data
Principles of Neocortical Function
retina
cochlea
somatic
data stream
x xxx x x
x x x x x xx
2) Hierarchy of memory regions
3) Sequence memory
5) All regions are sensory and motor6) Attention
4) Sparse Distributed Representations
1) On-line learning from streaming data
Principles of Neocortical Function
4) Sparse Distributed Representations
2) Hierarchy of memory regions
retina
cochlea
somatic
3) Sequence memory
5) All regions are sensory and motor6) Attention
data stream
1) On-line learning from streaming data
These six principles are necessary and sufficientfor biological and machine intelligence.
- All mammals from mouse to human have them
Sparse Distributed Representations (SDRs) • Many bits (thousands)
• Few 1’s mostly 0’s• Example: 2,000 bits, 2% active
• Each bit has semantic meaning• Meaning of each bit is learned, not assigned
01000000000000000001000000000000000000000000000000000010000…………01000
Dense Representations• Few bits (8 to 128)• All combinations of 1’s and 0’s• Example: 8 bit ASCII
• Individual bits have no inherent meaning• Representation is assigned by programmer
01101101 = m
A Few SDR Properties
1) Similarity: shared bits = semantic similarity
subsampling is OK Indices12|10
2) Store and Compare: store indices of active bits
Indices12345|40
Coincidence detectors
How does a layer of neurons learn sequences?
Sequence Memory (for inference and motor)
Each cell is one bit in our Sparse Distributed Representation
SDRs are formed via a local competition between cells.
SDR (time =1)
SDR (time =2)
Cell forms connections to subsample of previously active cells.Predicts its own future activity.
Multiple Predictions Can Occur at Once
With one cell per column, 1st order memoryWe need a high order memory
High Order Sequence MemoryEnabled by Columns of Cells
Cortical Learning Algorithm (CLA)Distributed sequence memoryHigh orderHigh capacityMultiple simultaneous predictionsSemantic generalization
1) NuPIC Open Source Project
Three Current Directions
www.Numenta.org
Single source tree (used by GROK)GPLv3
Steady community growth– 67 contributors (+26 since July)– 245 mailing list subscribers– 1621 total messages
eBook from community member OS community joining Kaggle CompetitionsFall Hackathon: 70 attendees
NuPIC Open Source Project
1) NuPIC Open Source Project
2) Custom CLA Hardware- Needed for scaling research and commercial applications- DARPA “Cortical Processor”- IBM, Seagate, Sandia Labs
3) Commercialization
Three Current Directions
2. Look at data 3. Build models
Problem: - Doesn’t scale with velocity and #models
Solution: - Automated model creation - Continuous learning - Temporal inference
1. Store data
Stream data Automated model creationContinuous learningTemporal inference
PredictionsAnomaliesActions
Past
Future
Data: Past and Future
Anomaly Detection Using Predictive Cortical Models
Cortical Memory
Encoder
SDR
Prediction Point anomaly score Time average Distribution of averages Metric anomaly score
Metric 1
Cortical Memory
Encoder
SDR
Prediction Point anomaly score Time average Distribution of averages Metric anomaly score
SDRMetric N
.
.
.
SystemAnomalyScore
Largely predictable
Largely unpredictable
Met
ric
valu
eAn
omal
y scor
eM
etric
va
lue
Anom
aly sc
ore
Smartphone-centric Ranks anomalous
instances Rapid drill down Continuously updated User-controlled
notifications
Breakthrough Science for Anomaly Detection
Reinventing UX for IT Monitoring
Grok for IT Monitoring
Detects problems thresholds miss
Continuous learning Automated model building State-of-the art neocortical
model
In private beta for Amazon AWS cloud [email protected]
Custom metrics for any application/server
Web interface and mobile client source code available under no-cost license
Engine API to be published NuPIC open source community
Extensible Architecture