2 of 12

HTM model proposed by Jeff Hawkins (inventor of the Palm Pilot)

Numenta: founded by Hawkins, Donna Dubinsky, and Dileep George.

2004 book by Hawkins, On Intelligence. Refers to HTM as CLA (Cortical Learning Algorithm).

https://numenta.com/resources/on-intelligence/

2008 dissertation by George (Stanford), How The Brain Might Work: a hierarchical and temporal model for learning and recognition.

http://alpha.tmit.bme.hu/speech/docs/education/02_DileepThesis.pdf

3 of 12

COURTESY: https://biology.stackexchange.com/questions/29759/is-the-six-layer-cortex-model-of-the-

mammalian-cortex-still-the-most-accepted-mo

4 of 12

Unsupervised method

similar to clustering, self-organized maps.

learn continuously from new input patterns.

Specialized for time-dependent sequences

temporal context, also spatial context.

natural statistics (scenes, object properties).

Robust with respect to inputs and outputs

high-capacity, high noise tolerance.

suited for prediction, anomaly detection, sensorimotor control.

COURTESY: https://www.analyticsvidhya.com/blog/2018/05/alternative-deep-learning-hierarchical-temporal-memory-htm-unsupervised-learning/

Attributes of HTM technique

5 of 12

Pyramidal Neuron (top):

HTM Neuron (bottom):

feedforward branches to many sources, feedback branches from many sources.

context is nonlinear, occurs in contact with feedforward branches (dendrites).

feedback and context are parallel registers.

feedforward is a convolution of feedback and context.

6 of 12

An example of encoding proprioceptive stimuli using a cortical hierarchy.

Two locational reference frames:

inputs (change relative to sensor position).

outputs (remain stable relative to sensor position).

7 of 12

“If you solve a problem no one has solved before, people will take notice” Jeff Hawkins (paraphrased)

Can we solve (address) morphogenesis with the HTM model?

Turing’s R-D Model

Positional Information

Kondo and Miura (2010). Science, 329(5999), 1616-1620.

Petkova et.al, (2019). Cell, 176(4), 844–855.

8 of 12

Other Alternative Models

9 of 12

Neuromorphic Models

Ray (2019). Neuromorphic computing finds new life in machine learning.

ZD Net, July 1.

https://www.zdnet.com/article/neuromorphic-computing-finds-new-life-in-machine-learning/

10 of 12

Neuromorphic Models

Neftci et.al (2013). PNAS, 110(37), E3468-E3476.

Rachmuth et.al (2011). PNAS, 108(49), E1266-E1274.

11 of 12

Dendritic Trees

Poirazi and Mel (2001). Impact of Active Dendrites and Structural Plasticity on the Memory Capacity of Neural Tissue. Neuron, 29, 779–796.

12 of 12

Neuroevolution and NEAT

Stanley and Miikkulainen (2002). Evolving Neural Networks Through Augmenting Topologies. Evolutionary Computation, 10(2), 99-127.

Stanley et.al (2019). Designing neural networks through neuroevolution. Nature Machine Intelligence, 1, 24–35.