text
stringlengths
0
8.13M
Moreover,inordertoimprovetheperformancewealso dx
perform the temporal multiplexing. The temporal mul- =a(y x), (85)
dt −
tiplexing has been found to be useful to extract com-
dy
plex dynamics on the exponentially large hidden nodes =x(b z) y, (86)
dt − −
through the restricted number of the true nodes [38]. In
dz
temporal multiplexing, not only the true nodes just af- =xy cz, (87)
dt −
ter the time evolution U , also at each of the subdivided
τ
V time intervals during the unitary evolution U to con-
τ with (a,b,c) = (10,28,8/3), (ii) the chaotic attractor of
structV virtualnodes,asshowninFig.3(b). Aftereach
Mackey-Glass equation,
input by S , the signals from the hidden nodes (via the
xk
true nodes) are measured for each subdevided intervals d x(t τ)
after the time evolution by U (v =1,2,...V), i.e., x(t)=β − γx(t) (88)
vτ/V dt 1+x(t τ)n −
r(kτ +(v/V)τ) U S r(kτ). (82)
≡ (v/V)τ xk with (β,γ,n) = (0.2,0.1,10) and τ = 17, (iii) R¨ossler
attoractor,
In total, now we have N V nodes, and the output is
×
defined as their linear combination:
dx
= y z, (89)
N V dt − −
(cid:88)(cid:88)
y k = W jo ,u vtr¯ l(kτ +(v/V)τ). (83) dy
=x+ay, (90)
l=1v=1 dt
dz
By using the teacher data y¯ L, the linear readout =b+z(x c), (91)
weights Wout can be determ{ ink e} dk by using the pseudo dt −
j,v
inverse. In Ref. [38], the performance of QRC has been
with (0.2,0.2,5.7), and (iv) H´enon map,
investigated extensively for both binary and continuous
inputs. The result shows that even if the number of the x =1 1.4x +0.3x . (92)
t+1 t t−1
qubits are small like 5-7 qubits the performance as pow- −
erfulastheechostatenetworkofthe100-500nodeshave Regarding (i)-(iii), the time series is obtained by us-
been reported both in short term memory and parity ing the fourth-order Runge-Kutta method with step size
check capacities. Note that, although we do not go into 0.02, and only x(t) is employed as a target. For the time
detail in this chapter, the technique called spatial multi- evolution of quantum reservoir, we employ a fully con-
plexing [40], which exploits multiple quantum reservoirs nected transverse-field Ising model
with common input sequence injected, is also introduced
(cid:88)
to harness quantum dynamics as a computational re- H = J X X +hZ , (93)
ij i j i
source. Recently, QRC has been further investigated in
ij
Refs. [41, 71, 74]. Specifically, in Ref. [71], the authors
use quantum reserovir computing to detect many-body where the coupling strengths are randomly chosen such
entanglement by estimating nonlinear functions of dein- that J is distributed randomly from [ 0.5,0.5] and
ij
sity operators like entropy. h=1.0. Thetimeintervalandthenumberofthevirtual
nodes are chosen to be τ = 4.0 and v = 10 so as to ob-
tain the best performance. The first 104 steps are used
D. Emulating chaotic attractors using quantum for training. After the linear readout weights are deter-
dynamics mined, several 103 steps are predicted by autonomously
evolving the quantum reservoir. The results are shown
To see a performance of QRC, here we demonstrate in Fig. 4 for each of (a) Lorenz attractor, (b) the chaotic
an emulation of chaotic attractors. Suppose x L is attractor of Mackey-Glass system, (c) R¨ossler attractor,
a discretized time sequence being subject to a{ cok m} pk lex and (d) H´enon map. All these results show that training
nonlinear equation, which might has a chaotic behavior. is done well and the prediction is successful for several
In this task, the target, which the network is to output, hundredssteps. Moreover,theoutputfromthequantum
is defined to be reservoir also successfully reconstruct the structures of
these chaotic attractors as you can see from the delayed
y¯ =x =f( x k ). (84) phase diagram.
k k+1 { j }j=1
11
(a) Lorenz attractor
1
0.9 0 .1 9
0.8 00 .. 78
000 ... 567 00000 ..... 23456
xk-15
0.4 0 . 01
000 ... 0123 9600t Qe Rac Cher 9800 10000 tim 10 e20 s0 tep 10400 10600 10800 11000 0 0.1 0.2 0.3 0.4 x 0. k5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0. x7 0. k8 0 -. 69 1
(b) Mackey-Glass time series
1 0.9
0.9 0.8
0.8 0.7
0.7 0.6
0.6
00 .. 45 xk-15 00 .. 45
0.3 0.3
0.2 teacher 0.2
0.1 QRC 0.1
0 9600 9800 10000 10 e20 s0 10400 10600 10800 11000 0 0 0.1 0.2 0.3 0.4xk 0.5 0.6 0.7 0.8 0.9
tim tep
(c) Rössler attractor
1