hippocampus-inspired adaptive neural algorithms · 2017-03-22 · kris carlson, william severa,...
TRANSCRIPT
![Page 1: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/1.jpg)
Photos placed in horizontal position with even amount of white space
between photos and header
Sandia National Laboratories is a multi-mission laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin Corporation, for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-AC04-94AL85000. SAND NO. 2011-XXXXP
Hippocampus-inspired Adaptive Neural Algorithms
Brad AimoneCenter for Computing Research
Sandia National Laboratories; Albuquerque, NM
![Page 2: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/2.jpg)
Can neural computing provide the next Moore’s Law?
![Page 3: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/3.jpg)
Moore’s Law was based on scientific discovery and successive innovations
Adapted from Wikipedia
![Page 4: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/4.jpg)
Each successive advance made more computing feasible
Adapted from Wikipedia
![Page 5: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/5.jpg)
Better devices made better computers, which allowed engineering new devices…
Images from Wikipedia
Circa 1980
![Page 6: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/6.jpg)
Better devices made better computers, which allowed engineering new devices…
Images from Wikipedia
Circa 2017
![Page 7: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/7.jpg)
If we extrapolate capabilities out, it is not obvious better devices is the answer…
Com
putin
g In
tellig
ence
per
sec
ond
per $
Conventional Computing
2020
ANNs
When Deep Nets became efficient
What Comes Next?Devices? or Neural Knowledge?
![Page 8: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/8.jpg)
Cycle of computing scaling already has begun to influence neuroscience
![Page 9: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/9.jpg)
Even if Moore’s Law ends, computing willcontinue to scale to be smarter
1950-2020
2010-???
![Page 10: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/10.jpg)
The reservoir of known neuroscience untapped for computing inspiration is enormous
James, et al., BICA 2017
![Page 11: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/11.jpg)
The brain has many mechanisms for adaptation; which are important for computing?
Current hardware focuses on synaptic plasticity, if anything
![Page 12: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/12.jpg)
There are different algorithmic approaches to neural learning
In situ adaptation Incorporate “new” forms of known neural
plasticity into existing algorithms
Ex situ adaptationDesign entirely new algorithms or algorithmic
modules to provide cognitive learning abilities
![Page 13: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/13.jpg)
NeurogenesisDeep
Learning
![Page 14: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/14.jpg)
Deep Networks are a function of training sets
0 5 10 15 20 25 30 35 40 45 500
10
20
30
40
50
60
70
1
7
![Page 15: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/15.jpg)
Deep Networks are a function of training sets
0 5 10 15 20 25 30 35 40 45 500
10
20
30
40
50
60
70
![Page 16: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/16.jpg)
Deep networks often struggle to generalize outside of training domain
0 5 10 15 20 25 30 35 40 45 500
10
20
30
40
50
60
70
![Page 17: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/17.jpg)
Neurogenesis can be used to capture new information without disrupting old information
Brain incorporates new neurons in a select number of regions Particularly critical for novelty detection and
encoding of new information “Young” hippocampal neurons exhibit increased
plasticity (learn more) and are dynamic in their representations
“Old” hippocampal neurons appear to have reduced learning and maintain their representations
Cortex does not have neurogenesis (or similar mechanisms) in adult-hood, but does during development
Aimone et al., Neuron 2011
![Page 18: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/18.jpg)
Neurogenesis can be used to capture new information without disrupting old information
Brain incorporates new neurons in a select number of regions
Hypothesis: Can new neurons be used to facilitate adapting deep learning?
?
![Page 19: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/19.jpg)
Neurogenesis can be used to capture new information without disrupting old information
Brain incorporates new neurons in a select number of regions
Hypothesis: Can new neurons be used to facilitate adapting deep learning?
Neurogenesis Deep Learning Algorithm Stage 1: Check autoencoder reconstruction to
ensure appropriate representations
0 10 20 30 40 500
10
20
30
40
50
60
70
0 10 20 30 40 500
10
20
30
40
50
60
70
![Page 20: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/20.jpg)
Neurogenesis can be used to capture new information without disrupting old information
Brain incorporates new neurons in a select number of regions
Hypothesis: Can new neurons be used to facilitate adapting deep learning?
Neurogenesis Deep Learning Algorithm Stage 1: Check autoencoder reconstruction to
ensure appropriate representations Stage 2: If mismatch, add and train new neurons
Train new nodes with novel inputs coming in (reduced learning for existing nodes)
0 10 20 30 40 500
10
20
30
40
50
60
70
0 10 20 30 40 500
10
20
30
40
50
60
70
![Page 21: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/21.jpg)
Neurogenesis can be used to capture new information without disrupting old information
Brain incorporates new neurons in a select number of regions
Hypothesis: Can new neurons be used to facilitate adapting deep learning?
Neurogenesis Deep Learning Algorithm Stage 1: Check autoencoder reconstruction to
ensure appropriate representations Stage 2: If mismatch, add and train new neurons
Train new nodes with novel inputs coming in (reduced learning for existing nodes)
Intrinsically replay “imagined” training samples from top-level statistics to fine tune representations
Stage 3: Repeat neurogenesis until reconstructions drop below error thresholds
0 10 20 30 40 500
10
20
30
40
50
60
70
0 10 20 30 40 500
10
20
30
40
50
60
70
![Page 22: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/22.jpg)
Neurogenesis algorithm effectively balances stability and plasticity
![Page 23: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/23.jpg)
Neurogenesis algorithm effectively balances stability and plasticity
![Page 24: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/24.jpg)
NDL applied to NIST data set
![Page 25: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/25.jpg)
A New View of the
Hippocampus
![Page 26: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/26.jpg)
Deep learning ≈ Cortex What ≈ Hippocampus?
![Page 27: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/27.jpg)
Can a new framework for studying the hippocampus help inspire computing?
Desired functions Learn associations between cortical modalities Encoding of temporal, contextual, and spatial
information into associations Ability for “one-shot” learning Cue-based retrieval of information
Desired properties Compatible with spiking representations Network must be stable with adaptation Capacity should scale nicely Biologically plausible in context of extensive
hippocampus literature Ability to formally quantify costs and performance
This requires a new model of CA3
Entorhinal Cortex
Dentate Gyrus
CA3
CA1
![Page 28: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/28.jpg)
Formal model of DG provides lossless encoding of cortical inputs Constraining EC
inputs to have “grid cell” structure sets DG size to biological level of expansion (~10:1)
Mixed code of broad-tuned (immature) neurons and narrow tuned (mature) neurons confirms predicted ability to encode novel information
28
William Severa, NICE 2016Severa et al., Neural Computation, 2017
![Page 29: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/29.jpg)
Classic model of CA3 uses Hopfield-like recurrent network attractors
Problems “Auto-associative” attractors make more
sense in frequency coding regime than in spiking networks
Capacity of classic Hopfield networks is generally low
Quite difficult to perform stable one-shot updates to recurrent networks
29
Deng, Aimone, Gage, Nat Rev Neuro 2010
![Page 30: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/30.jpg)
Moving away from the Hopfield “learned auto-association” function for CA3
Hopfield dynamics are discrete state transitions
time
Hillar and Tran, 2014
![Page 31: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/31.jpg)
Spiking dynamics are inconsistent with fixed point attractors in associative models
Biology uses sequence of spiking neurons?Hopfield dynamics are discrete state transitions
time time
![Page 32: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/32.jpg)
Spiking dynamics are inconsistent with fixed point attractors in associative models
time time
One can see how sequences can replace fixed populations
![Page 33: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/33.jpg)
Path attractors, such as orbits, are consistent with spiking dynamics
![Page 34: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/34.jpg)
A new dynamical model of CA3
Problems “Auto-associative” attractors make more
sense in frequency coding regime than in spiking networks
Capacity of classic Hopfield networks is generally low
Quite difficult to perform stable one-shot updates to recurrent networks
34
Orbits of Spiking Neurons
![Page 35: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/35.jpg)
Neuromodulation can shift dynamics of recurrent networks
35
Carlson, Warrender, Severa and Aimone; in preparation
![Page 36: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/36.jpg)
Cortex and subcortical inputs can modulate CA3 attractor access
Modulation can be provided mechanistically by several sources
Spatial distribution of CA3 synaptic inputs suggests EC inputs could be considered modulatory
Metabotrophic modulators (e.g., serotonin, acetylcholine) can bias neuronal timings and thresholds
Attractor network can thus have many “memories”, but only fraction are accessible within each context
![Page 37: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/37.jpg)
A new modulated, dynamical model of CA3
Problems “Auto-associative” attractors make more
sense in frequency coding regime than in spiking networks
Capacity of classic Hopfield networks is generally low
Quite difficult to perform stable one-shot updates to recurrent networks
37
Orbits of Spiking Neurons
Context modulation
![Page 38: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/38.jpg)
CA1 encoding can integrate cortical input with transformed DG/CA3 input CA1 plasticity is dramatic
Synapses appear to be structurally volatile Representations are temporally volatile Consistent with one-shot learning
Can consider EC-CA1-EC loosely as an autoencoder, with DG / CA3 “guiding” what representation is used within CA1
38
Current State
Average State of the CA3
OrbitCombined
Representation across Time
![Page 39: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/39.jpg)
A new modulated, dynamical model of CA3
Problems “Auto-associative” attractors make more
sense in frequency coding regime than in spiking networks
Capacity of classic Hopfield networks is generally low
Quite difficult to perform stable one-shot updates to recurrent networks
39
Orbits of Spiking Neurons
Context modulation
Schaffer Collateral (CA3-CA1) Learning
![Page 40: Hippocampus-inspired Adaptive Neural Algorithms · 2017-03-22 · Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard. Title: PowerPoint Presentation Author:](https://reader033.vdocuments.site/reader033/viewer/2022050109/5f46ce71d2abcb1b705164ec/html5/thumbnails/40.jpg)
Thanks!
HAANA Grand Challenge LDRDDOE NNSA Advanced Simulation and Computing Program
Neurogenesis Deep Learning:
Tim Draelos, Nadine Miner, Chris Lamb, Jonathan Cox, Craig Vineyard, Kris Carlson, William Severa, and Conrad James
Hippocampus Algorithm:
Kris Carlson, William Severa, Ojas Parekh, Frances Chance, and Craig Vineyard