A New Hierarchical Temporal Memory Algorithm Based on Activation Intensity

Algorithm 1

Temporal memory learning algorithm based on activation intensity.

	Input: active columns at time t, activation intensity vector at t and t − 1 ,
(1)	for each column colj in HTM //first phase: cell activation
(2)	if colj is active then //predictive cells at t − 1 on active column will become active cell at t
(3)	for each cell cei on colj
(4)	if cei has activeSegments(t − 1)
(5)	activeCells(t) ← cei
(6)	learningSegments(t) = activeSegments(t − 1)
(7)	else //if active column has no predict cells at t − 1, all cells become active cells at t
(8)	activeCells(t) ← cei ∀i ∈ (1,…, nce)
(9)	learningSegments(t) = bestMatchingSegment(colj)
(10)	for each segment d in learningSegments(t) //second phase: cell synapse updating
(11)	for each synapse s on d
(12)	if s.presynapticCell in activeCells (t − 1) and//reinforcement
(13)	s.permanence+ =   =
(14)	else //punishment
(15)	s.permanence+ =  =
(16)	for each segment d on all columns//third phase: predictive cell selection
(17)	for each synapse s on d
(18)	if s.presynapticCell in activeCells (t) then
(19)	if s.permanence > θc numActiveConnected++
(20)	if numActiveConnected > θa
(21)	activeSegments(t) ← d
(22)	predictiveCells(t) ← cell(d)