ABCDEFGHIJKLMNOPQRSTUVWXYZ
1
Hopfield Networks
2
Plus and Minus Sign Recognition
3
M. Ali Yousuf
4
First version: circa 1998. Last updated: June 17, 2025
5
https://www.nobelprize.org/prizes/physics/2024/hopfield/facts/
6
7
Let's look at two simple patterns (a plus sign and a minus sign)
8
9
-11-1
10
V+ =111V+ =-11-1111-11-1
11
-11-1
12
13
-1-1-1
14
V- =111V- =-1-1-1111-1-1-1
15
-1-1-1
16
17
So
18
=-1
19
and
20
21
=1
22
23
Claim: The following matrix can recall these two patterns, even if they are noisy
24
25
26
202-2-2-2202
27
020000020
28
202-2-2-2202
29
M =-20-2222-20-2
30
-20-2222-20-2
31
-20-2222-20-2
32
202-2-2-2202
33
020000020
34
202-2-2-2202
35
36
37
Now lets try a new (noisy) vector and see if we can recover the true pattern.
38
39
1-1-1
40
111T =1-1-1111-1-1-1
41
-1-1-1
42
43
44
We are going to calculate the matrix product: MT
45
(The Threshold is 0, anything greater than 0 is one)
46
47
202-2-2-22021-10-1
48
020000020-1-4-1
49
202-2-2-2202-1-10-1
50
-20-2222-20-2X1101-1-1-1
51
-20-2222-20-21 =10 -->1-->111
52
-20-2222-20-21101-1-1-1
53
202-2-2-2202-1-10-1
54
020000020-1-4-1
55
202-2-2-2202-1-10-1
56
57
i.e., a minus sign
58
59
It is easy to check that we can recover the original vectors also:
60
V+
61
202-2-2-2202-1-14-1
62
020000020141
63
202-2-2-2202-1-14-1
64
-20-2222-20-21141-11-1
65
-20-2222-20-21 =14 -->1111
66
-20-2222-20-21141-11-1
67
202-2-2-2202-1-14-1
68
020000020141
69
202-2-2-2202-1-14-1
70
71
and V-
72
202-2-2-2202-1-14-1
73
020000020-1-4-1
74
202-2-2-2202-1-14-1
75
-20-2222-20-21141-1-1-1
76
-20-2222-20-21 =14 -->1111
77
-20-2222-20-21141-1-1-1
78
202-2-2-2202-1-14-1
79
020000020-1-4-1
80
202-2-2-2202-1-14-1
81
82
Try a distorted minus sign
83
84
-1-1-1
85
11-1T =-1-1-111-1-1-1-1
86
-1-1-1
87
88
202-2-2-2202-1-10-1
89
020000020-1-4-1
90
202-2-2-2202-1-10-1
91
-20-2222-20-21101-1-1-1
92
-20-2222-20-21 =10 -->1111
93
-20-2222-20-2-1101-1-1-1
94
202-2-2-2202-1-10-1
95
020000020-1-4-1
96
202-2-2-2202-1-10-1
97
98
99
Lets try the inverse of a minus sign
100