Deep Temporal Memory

Author: PeterOvermann

Mathematica/deeptemporalmemory.m

@@ -0,0 +1,93 @@
+
+(*
+
+deeptemporalmemory.m
+
+Mathematica implementation of a Deep Temporal Memory based on multiple triadic memory instances
+Copyright (c) 2022 Peter Overmann
+
+Permission is hereby granted, free of charge, to any person obtaining a copy of this software
+and associated documentation files (the “Software”), to deal in the Software without restriction,
+including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense,
+and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so,
+subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all copies or substantial
+portions of the Software.
+
+THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT
+LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
+WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE
+OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+
+Requires TriadicMemory.
+
+*)
+
+
+
+R[f_Symbol, {n_Integer, p_Integer} ] := Module[ {T, x, y, z, overlap},
+
+	(* instantiate a triadic memory unit *)
+
+	TriadicMemory[ T, {n, p}];
+
+	overlap[a_SparseArray, b_SparseArray] := Total[BitAnd[a, b]];
+
+	x = y = z = SparseArray[{0}, {n}];
+
+	(* reset x, y, z if input is zero *)
+
+	f[SparseArray[{0}, {n}]] := x = y = z = SparseArray[{0}, {n}];
+
+	f[input_SparseArray] := Module[ {},
+
+		x = BitOr[y, z]; (* binarize x and y using ranked-max algorithm *)
+		y = input;
+
+		If[Total[x] > 0 && overlap[T[_, y, z = T[x, y, _]], x] < p, T[x, y, z = T[]]];
+
+		z
+		]
+	];
+
+
+
+TemporalMemory[t_Symbol, {n_Integer, p_Integer}] :=
+
+	Module[{M, R1, R2, R3, R4, R5, R6, R7, x, y, z, t1, t2, t3, t4, t5, t6, t7, t8},
+
+	(* predictions / readout memory *)
+	TriadicMemory[M, {n, p}];
+
+	(* bigram encoder units *)
+	R[ #, {n, p}] & /@ {R1, R2, R3, R4, R5, R6, R7 };
+
+	(* initialize state variables with null vectors *)
+	x = y = z = t1 = t2 = t3 = t4 = t5 = t6 = t7 = M[0];
+
+	t[inp_] := Module[ {},
+
+		(* flush state if input is zero - needed when used as a sequence memory *)
+		If[Total[inp] == 0, x = y = z = M[0]];
+
+		(* store new prediction if necessary *)
+		If[z != inp, M[x, y, inp]];
+
+		(* encoding chain *)
+		t1 = R1[inp];
+		t2 = R2[t1]; 
+		t3 = R3[t2];
+		t4 = R4[t3];
+		t5 = R5[t4];
+		t6 = R6[t5];
+		t7 = R7[t6];
+
+		(* prediction readout from t1, t2, t4 and t7 *)
+		z = M[x = BitOr[t1, t4], y = BitOr[t2, t7], _] 
+		]
+
+	];
+