
This paper deals with a class of fractional-order octonion-valued neural networks (FOOVNNs) with impulsive effects. Firstly, although the multiplication of octonion numbers does not satisfy the commutativity and associativity, we don't need to separate an octonion-valued system into eight real-valued systems. Secondly, by applying the appropriate Lyapunov function, and inequality techniques, we obtain the global asymptotical synchronization of FOOVNNs. Finally, we give two illustrative examples to illustrate the feasibility of the proposed method.
Citation: Jin Gao, Lihua Dai. Asymptotic synchronization analysis of fractional-order octonion-valued neural networks with impulsive effects[J]. AIMS Mathematics, 2023, 8(1): 1975-1994. doi: 10.3934/math.2023102
[1] | Fu-Tao Hu, Xing Wei Wang, Ning Li . Characterization of trees with Roman bondage number 1. AIMS Mathematics, 2020, 5(6): 6183-6188. doi: 10.3934/math.2020397 |
[2] | Abel Cabrera Martínez, Iztok Peterin, Ismael G. Yero . Roman domination in direct product graphs and rooted product graphs. AIMS Mathematics, 2021, 6(10): 11084-11096. doi: 10.3934/math.2021643 |
[3] | Jianwei Du, Xiaoling Sun . On symmetric division deg index of trees with given parameters. AIMS Mathematics, 2021, 6(6): 6528-6541. doi: 10.3934/math.2021384 |
[4] | Zepeng Li . A note on the bounds of Roman domination numbers. AIMS Mathematics, 2021, 6(4): 3940-3946. doi: 10.3934/math.2021234 |
[5] | Bana Al Subaiei, Ahlam AlMulhim, Abolape Deborah Akwu . Vertex-edge perfect Roman domination number. AIMS Mathematics, 2023, 8(9): 21472-21483. doi: 10.3934/math.20231094 |
[6] | Jian Yang, Yuefen Chen, Zhiqiang Li . Some sufficient conditions for a tree to have its weak Roman domination number be equal to its domination number plus 1. AIMS Mathematics, 2023, 8(8): 17702-17718. doi: 10.3934/math.2023904 |
[7] | Rangel Hernández-Ortiz, Luis Pedro Montejano, Juan Alberto Rodríguez-Velázquez . Weak Roman domination in rooted product graphs. AIMS Mathematics, 2021, 6(4): 3641-3653. doi: 10.3934/math.2021217 |
[8] | Mingyu Zhang, Junxia Zhang . On Roman balanced domination of graphs. AIMS Mathematics, 2024, 9(12): 36001-36011. doi: 10.3934/math.20241707 |
[9] | Saeed Kosari, Yongsheng Rao, Zehui Shao, Jafar Amjadi, Rana Khoeilar . Complexity of signed total k-Roman domination problem in graphs. AIMS Mathematics, 2021, 6(1): 952-961. doi: 10.3934/math.2021057 |
[10] | Muhammad Kamran Jamil, Muhammad Imran, Aisha Javed, Roslan Hasni . On the first general Zagreb eccentricity index. AIMS Mathematics, 2021, 6(1): 532-542. doi: 10.3934/math.2021032 |
This paper deals with a class of fractional-order octonion-valued neural networks (FOOVNNs) with impulsive effects. Firstly, although the multiplication of octonion numbers does not satisfy the commutativity and associativity, we don't need to separate an octonion-valued system into eight real-valued systems. Secondly, by applying the appropriate Lyapunov function, and inequality techniques, we obtain the global asymptotical synchronization of FOOVNNs. Finally, we give two illustrative examples to illustrate the feasibility of the proposed method.
It is well-known that the chemical compounds can be represented by graphs (or chemical graphs) in which vertices correspond to the atoms while edges represent the covalent bonds between atoms. Predicting physicochemical properties of chemical compounds is considered to be a fascinating issue in theoretical chemistry. Many predictive methods have been and are being developed for correlating molecular structures with their physicochemical properties. One of the simplest such methods is topological indices. In chemical graph theory, graph invariants are usually referred as topological indices. Nowadays the most active problem in the theory of topological indices is finding the extremal structures under certain constraints, that maximize or minimize the given topological index.
Suppose that G is a connected graph with vertex set V(G) and edge set E(G). The degree of vertex u is denoted by d(u). A vertex u in G is called pendant if d(u)=1. The distance from u to v, denoted by d(u,v), is defined as the smallest length of all u−v paths in G. The eccentricity of v, denoted as ecc(v), is the maximum distance between v and other vertices in G. For other notation and terminology not defined here, the readers are referred to the book by West [24].
A topological index is a numerical descriptor of the molecular structure derived from the corresponding molecular graph. The first Zagreb index (the second Zagreb index as well) was first introduced by Gutman and Trinajstić [13] and has been closely related to many chemical properties [13,22]. The first Zagreb index of a graph G is defined by:
M1(G)=∑v∈V(G)d2(v). |
Some mathematical properties of the first Zagreb index can be found in [15,16,17,21,23] and the comprehensive surveys [5,12,20].
Roman dominating function RDF of G is a function f:V(G)→{0,1,2} satisfying the condition that every vertex with label 0 is adjacent to a vertex with label 2. The weight of an RDF f is w(f)=∑v∈V(G)f(v). The Roman domination number of G, denoted by γR(G), is the minimum weight among all RDF in G. The concept of Roman domination number of graphs was introduced by Cockayne et al. [8], and further studied in [1,2,3,9,11,14,19] and references therein.
Relationship between various topological indices and domination number of graphs is in the focus of interest of the researchers for quite many years, and this direction is continuously vital, see [4,6,10,18], the surveys [5,20] and references therein. In this paper, we determine the trees with the maximum first Zagreb index when Roman domination number is fixed.
Let G−v be the graph obtained from T by deleting the vertex v and its incident edges, for v∈V(G). Let T(n,γR) be the set of trees whose order is n and Roman domination number is γR, where n≥max{⌈3γR−52⌉,γR+3}.
In this section, we will determine the trees with the maximum first Zagreb index in T(n,γR) step by step.
Let us introduce some properties about the Roman domination number of trees which will be used later.
The first lemma reveals the effect on the Roman domination number of trees after deleting a pendant vertex (there are two possibilities for the effect).
Lemma 2.1. Let T be a tree with a pendant vertex v. Then
γR(T−v)≤γR(T)≤γR(T−v)+1. |
Proof. First, we prove the left inequality γR(T−v)≤γR(T). If v is labeled as 0 in an RDF (Roman dominating function) of T with weight γR(T), then delete such label 0 of v, it leads to an RDF of T−v still with weight γR(T), so γR(T−v)≤γR(T). If v is labeled as 1 in an RDF of T with weight γR(T), then similarly we have γR(T−v)≤γR(T)−1<γR(T). Assume that v is labeled as 2 in an RDF of T with weight γR(T), and denote by u the unique neighbor of v in T. Obverse that u must be labeled as 0, otherwise we may obtain another RDF of T with a smaller weight than γR(T), by decreasing the label 2 of v. At this time, we may change the labels of u and v both into 1, and keep the labels of other vertices unchanged. Clearly, the weight under this new RDF of T is still γR(T). But now v is labeled as 1, so by the above comments, we obtain γR(T−v)<γR(T) again.
The second inequality γR(T)≤γR(T−v)+1 is straightforward. Starting from the tree T−v, when we add a new vertex v, we may always label it as 1. So there is an RDF of T with weight γR(T−v)+1, which results in γR(T)≤γR(T−v)+1.
From the above lemma, after deleting a pendant vertex v from a tree T, either γR(T−v)=γR(T) or γR(T−v)=γR(T)−1. In the following lemma, we present a condition in which γR(T−v)=γR(T) holds.
Lemma 2.2. Let T be a tree. Assume that there are (at least) three pendant vertices sharing the same neighbor in T. In particular, one of the three pendant vertices is denoted by v. Then γR(T)=γR(T−v).
Proof. From Lemma 2.1, we already have γR(T−v)≤γR(T). So we are left to show the reversed counterpart that γR(T)≤γR(T−v). Assume that v,u,w are the three pendant vertices sharing the same neighbor z. If there is an RDF of T−v with weight γR(T−v) such that z is labeled as 2, then in addition to such RDF of T−v, assign the label 0 to the new added vertex v, an RDF of T with weight γR(T−v) follows, which leads to γR(T)≤γR(T−v) immediately.
Next let us analyze if we can always have such ideal RDF of T−v (with weight γR(T−v) such that z is labeled as 2). If u or w is labeled as 0, then z must be labeled as 2, from the definition of Roman domination number. Otherwise, the labels of u and w are both at least 1, means that the sum of labels of u and w is at least 2, then we modify such RDF of T−v, such that u and w are both labeled as 0, z is labeled as 2, keep the labels of other vertices unchanged, clearly it is just our desired RDF of T−v (such revised RDF of T−v would not increase the weight).
The proof is completed now by noting that γR(T)≤γR(T−v) always holds.
The maximum degree of trees in T(n,γR) plays an important role in our subsequent deduction, which is at most n−γR+1 established as following.
Lemma 2.3. [7] If G is an n-vertex graph with maximum degree Δ(G), then γR(G)≤n−Δ(G)+1.
Based on this lemma, subsequently we partition our consideration into two parts: Trees with maximum degree exactly n−γR+1 and trees with maximum degree less than n−γR+1.
Actually the trees with the maximum first Zagreb index in T(n,γR) are of maximum degree n−γR+1. We focus on such trees (with maximum degree n−γR+1) in the following several lemmas, which is crucial for our final results.
First, we observe that such trees have a small eccentricity.
Lemma 2.4. Let T∈T(n,γR).Assume that v is a vertex in T with degree n−γR+1. Then ecc(v)≤3.
Proof. First, we introduce a realizable RDF of T with weight γR. Let us label v as 2, label all the neighbors of v (there are n−γR+1 such neighbors) as 0, and label the remaining vertices (there are γR−2 vertices) as 1. Such RDF of T gives the weight γR. It is worth mentioning that there is no RDF of T with a smaller weight than γR, from the definition of Roman domination number (since γR(T)=γR).
Assume that x,y,z∈V(T) such that xy and yz are edges of T. We claim that at least one of x,y,z is adjacent to v. Suppose that it is not true (none of x,y,z is adjacent to v). Under the above-mentioned RDF of T, each of x,y,z should be labeled as 1. But if we change the labels of x,y,z such that the vertex y is labeled as 2, vertices x,z are both labeled as 0, and the labels of other vertices keep unchanged, it leads to a new weight γR−1 (less than γR), which is a contradiction to the definition of Roman domination number.
Finally, ecc(v)≤3 follows from the above claim immediately (otherwise we can find the vertices x,y,z satisfying the above condition).
Lemma 2.5. Let G be a tree of order n with maximum degree Δ(G). If Δ(G)=n−γR+1, then for every vertex v of maximum degree have to labeled by 2.
Proof. Let G be a graph with Δ(G)=n−γR+1 and let v a vertex of maximum degree. Consider an RDF f that assigns the value 2 to v, 0 to every neighbor of v and 1 to the remaining vertices. Clearly w(f)=n−γR+1 and thus f is a γR-function. Now suppose to the contrary that v has a neighbor w having at least three neighbors in N(w). Then reassigning w a 2 instead of 0 and each vertex of N(w) a 0 instead of 1 produces an RDF with smaller weight than γR, a contradiction.
We present the definition of branches of a tree, which helps us describe the local structures of trees with the maximum first Zagreb index.
Definition 2.1. Let T be a tree with v∈V(T). Let u be some neighbor of v in T. Based on T, delete all the neighbors of v except u, the component of the resulting forest containing v is said to be a branch of T at v.
For a tree with the maximum first Zagreb index in T(n,γR), there are exactly five possibilities for the branches at a vertex of maximum degree n−γR+1.
Lemma 2.6. Let T be a tree with the maximum first Zagreb index in T(n,γR).Assume that v is a vertex in T with degree n−γR+1, and H is a branch of T at v. Then H≅Hi, for 1≤i≤5 (all the Hi are depicted in Figure 1).
Proof. Let H be a branch of T at v, and set s=max{d(v,x):x∈V(H)}. From Lemma 2.4, s=1,2,3. In particular, the cases s=1,2 corresponding to H1,H2,H3 are clear. Assume that s=3 in the following.
Just like the proof of Lemma 2.4, we deduce that there are no two pendant vertices, at distances 3 from v, sharing a common neighbor in H (otherwise we have a weight less than γR after modifying the labels of the two pendant vertices and their common neighbor, just like the proof of Lemma 2.4), which implies the existence of a pendant path of length 2 or 3 in H.
If H is the pendant path of length 3 (at v), then reformulate H as H3, keep other parts unchanged, the weight of T keeps the same, but the first Zagreb index becomes larger, which is a contradiction to our choosing of T (with the maximum first Zagreb index).
Suppose that there is a pendant path of length 2 in H. Denote by u the root vertex of such pendant path (u is adjacent to v). If u is of degree at least 4, then u has (at least) three neighbors (each one is labeled as 1) besides v, modify the labels of u and its neighbors such that u is labeled as 2, all the neighbors of u except v are labeled 0, keep the labels of other vertices except v unchanged, such a new RDF of T would have a smaller weight than γR, a contradiction. Thus, u is of degree 3, H≅H4 or H5 follows easily.
At the next stage, let us go ahead under the same hypothesis as Lemma 2.6. Let T be a tree with the maximum first Zagreb index in T(n,γR). Assume that v is a vertex in T with degree n−γR+1. Lemma 2.6 has revealed that each branch of T at v can only be Hi, for 1≤i≤5. Furthermore, we determine the possible occurrence numbers for each Hi in T, 1≤i≤5.
Denote by ni the number of branches of T at v with the form Hi, for 1≤i≤5. Then
n1+n2+n3+n4+n5=n−γR+1 | (2.1) |
and
n2+2n3+3n4+4n5=γR−2. | (2.2) |
The former equation follows from the degree n−γR+1 of v, and the latter equation comes from the Roman domination number γR of T, where the uncounted vertex v is labeled as 2.
For convenience, we use T=(n1,n2,n3,n4,n5) to represent a tree T∈T(n,γR) having ni branches Hi of T at v, for 1≤i≤5. When T=(n1,n2,n3,n4,n5), it is trivial to get that
M1(T)=(n−γR+1)2+n1+5n2+11n3+15n4+19n5. | (2.3) |
Lemma 2.7. Let T=(n1,n2,n3,n4,n5) be a tree with the maximum first Zagreb index in T(n,γR). Then none of the following conditions would occur:
(i) n1≥1 and n5≥1;
(ii) n2≥2;
(iii) n2≥1 and n4≥1;
(iv) n1≥1 and n4≥2;
(v) n2≥1 and n5≥1.
Proof. Define
T′={(n1−1,n2,n3+2,n4,n5−1),ifn1≥1 and n5≥1;(n1+1,n2−2,n3+1,n4,n5),ifn2≥2;(n1,n2−1,n3+2,n4−1,n5),ifn2≥1 and n4≥1;(n1−1,n2,n3+3,n4−2,n5),ifn1≥1 and n4≥2;(n1,n2−1,n3+1,n4+1,n5−1),ifn2≥1 and n5≥1. |
In each case, it is easy to see that T′ has the same number of vertices and Roman domination number as T from Eqs (2.1) and (2.2), but
M1(T′)−M1(T)=2 |
from Eq (2.3). It means that under each of such conditions, we may always construct another tree on the same number of vertices and Roman domination number but having a larger M1 value, which contradicts with our choosing for T (with the maximum first Zagreb index).
In virtue of Lemma 2.7, the number of candidates (with the maximum first Zagreb index) is depleted.
Remark 2.1. Roman domination number is unchanged after using transformations in the proof of Lemmas 2.6 and 2.7.
Lemma 2.8. Let T∈T(n,γR), where n≥max{⌈3γR−52⌉,γR+3}. Assume that the maximum degree of T is n−γR+1.Then
M1(T)≤{n2−(2γR−3)n+γ2R+2γR−8,ifγR is even;n2−(2γR−3)n+γ2R+2γR−9,ifγR is odd. |
In particular:
● When γR is even, the equality holds if and only if T=(2n−3γR+42,0,γR−22,0,0).
● When γR is odd, the equality holds if and only if T=(2n−3γR+32,1,γR−32,0,0) or T=(2n−3γR+52,0,γR−52,1,0).
Proof. Assume that T=(n1,n2,n3,n4,n5) is a tree with the maximum first Zagreb index in T(n,γR). We partition the proof into two parts: n1≥1 and n1=0.
Case 1. n1≥1.
From Lemma 2.7 (ⅰ), (ⅱ) and (ⅳ), n2=0 or 1, n4=0 or 1, n5=0. So there are actually three candidates (n2=n4=1 is impossible from Lemma 2.7 (ⅲ)):
(i) n1=2n−3γR+42, n2=0, n3=γR−22, n4=0, n5=0;
(ii) n1=2n−3γR+32, n2=1, n3=γR−32, n4=0, n5=0;
(iii) n1=2n−3γR+52, n2=0, n3=γR−52, n4=1, n5=0.
Under the condition (ⅰ), we have that γR is even, and
M1(T)=n2−(2γR−3)n+γ2R+2γR−8. |
As to the conditions (ⅱ) and (ⅲ), in either case, γR is odd, and
M1(T)=n2−(2γR−3)n+γ2R+2γR−9. |
Case 2. n1=0.
Notice that n2=0 or 1, from Lemma 2.7 (ⅱ).
First suppose that n2=1. Then n4=n5=0 from Lemma 2.7 (ⅲ) and (ⅴ). Furthermore, n3=n−γR (from Eq (2.1)) and 2n=3γR−3 (from Eq (2.2)), implying that γR is odd, which coincides with condition (ⅱ) mentioned in Case 1.
Next suppose that n2=0. Then by Eqs (2.1) and (2.2),
{n3+n4+n5=n−γR+1,2n3+3n4+4n5=γR−2. |
Clearly,
2n3+3n4+4n5≥2(n3+n4+n5), |
equivalently γR−2≥2(n−γR+1), i.e., 3γR−4≥2n. Recall that 2n≥3γR−5, from the hypothesis about the lower bound of n. So either 2n=3γR−4 or 2n=3γR−5. The former is actually the above condition (ⅰ) occurring in Case 1, while the latter is identical to condition (ⅲ) in Case 1.
The proof is completed now.
Remark 2.2. In order to provide a visual representation for the three trees attaining the equality cases, we draw such three trees in Figure 2.
All the trees in T(n,γR) with maximum degree n−γR+1 have been done now. In fact, the trees attaining the equality cases obtained above are also the ones with the maximum first Zagreb index in T(n,γR), which will be proved subsequently.
In the next theorem, we obtain the upper bound for the first Zagreb index of trees with given Roman domination number, and the trees attaining the upper bound are also characterized (they coincide exactly with the ones occurred in Lemma 2.8).
Theorem 3.1. Let T∈T(n,γR), where n≥max{⌈3γR−52⌉,γR+3}.Then
M1(T)≤{n2−(2γR−3)n+γ2R+2γR−8,if γR is even;n2−(2γR−3)n+γ2R+2γR−9,if γR is odd. |
In particular:
● When γR is even, the equality holds if and only if T=(2n−3γR+42,0,γR−22,0,0).
● When γR is odd, the equality holds if and only if T=(2n−3γR+32,1,γR−32,0,0) or T=(2n−3γR+52,0,γR−52,1,0).
Proof. For convenience, set
f(n,γR)={n2−(2γR−3)n+γ2R+2γR−8,if γR is even;n2−(2γR−3)n+γ2R+2γR−9,if γR is odd. |
That is to say, we would like to show that M1(T)≤f(n,γR), and characterize the equality cases.
Recall that the maximum degree of T is at most n−γR+1, from Lemma 2.3. In Lemma 2.8, we have considered the upper bound of the first Zagreb index when the maximum degree of T is exactly n−γR+1. So we are left to show that M1(T)<f(n,γR) when the maximum degree of T is less than n−γR+1. We will realize it by using induction on γR.
When γR=2,3, there is no tree whose maximum degree is less than n−γR+1. When γR=4, the trees under consideration should be double stars, the result can be checked easily. Now assume that the result holds for any tree with order less than n and Roman domination number less than γR, and T is a tree on n vertices with Roman domination number γR.
Recall an obvious observation about a tree T: There is some vertex, say u, having d(u)−1 pendant neighbors, where d(u)≥2. Let v be a pendant neighbor of u in T.
When d(u)≥4 (u has at least three pendant neighbors), from Lemma 2.2, we get γR(T)=γR(T−v)=γR. Observe that
f(n,γR(T))−f(n−1,γR(T−v))=f(n,γR)−f(n−1,γR)=2(n−γR+1), |
and d(u)<n−γR+1 (since the maximum degree of T is less than n−γR+1). Therefore, together with induction, it follows that
M1(T)=M1(T−v)+2d(u)<f(n−1,γR(T−v))+2(n−γR+1)<f(n,γR(T)), |
i.e., M1(T)<f(n,γR).
We may assume that d(u)=2 or 3 in the following. Furthermore, we may assume that γR(T−v)=γR(T)−1=γR−1, otherwise γR(T−v)=γR(T)=γR from Lemma 2.1, which can be settled by the method used for the case d(u)≥4 again.
For γR(T−v)=γR(T)−1=γR−1, it is easy to see that
f(n,γR(T))−f(n−1,γR(T−v))=f(n,γR)−f(n−1,γR−1)={6,if γR is even;4,if γR is odd. |
Thus, when d(u)=2, by induction, we have
M1(T)=M1(T−v)+4≤f(n−1,γR(T−v))+4≤f(n,γR(T)). |
In particular, if M1(T)=f(n,γR(T)), then T−v is a tree with maximum degree (n−1)−γR(T−v)+1=n−γR+1, which contradicts with the hypothesis that the maximum degree of T is less than n−γR+1. So M1(T)<f(n,γR(T))=f(n,γR).
When d(u)=3, assume that v and w are the two pendant neighbors of u in T. Then
M1(T)=M1(T−v−w)+10≤f(n−2,γR(T−v−w))+10. |
Notice that either γR(T−v−w)=γR(T)−1=γR−1, or γR(T−v−w)=γR(T)−2=γR−2. If γR(T−v−w)=γR−1, then
f(n,γR(T))−f(n−2,γR(T−v−w))=f(n,γR)−f(n−2,γR−1)={2n−2γR+8,if γR is even;2n−2γR+6,if γR is odd, |
and thus,
M1(T)≤f(n−2,γR(T−v−w))+10≤f(n,γR(T))−(2n−2γR+6)+10<f(n,γR(T)), |
since n−γR+1>3 (considering the hypothesis about the maximum degree of T). If γR(T−v−w)=γR−2, then
f(n,γR(T))−f(n−2,γR(T−v−w))=f(n,γR)−f(n−2,γR−2)=10, |
and thus,
M1(T)≤f(n−2,γR(T−v−w))+10=f(n,γR(T)). |
In particular, if M1(T)=f(n,γR(T)), then T−v−w is a tree with maximum degree (n−2)−γR(T−v−w)+1=n−γR+1, which contradicts with the hypothesis that the maximum degree of T is less than n−γR+1. So M1(T)<f(n,γR(T))=f(n,γR).
The proof is completed.
This paper is devoted to the investigation of relationship between the first Zagreb index and Roman domination number of trees. More precisely, we establish an upper bound of the first Zagreb index of trees in terms of Roman domination number, and all the tree(s) attaining the equality case are characterized.
As a possible sequel in the future, it is interesting to investigate the relationship between the first Zagreb index (or other well-known topological indices, e.g., Randić index, Wiener index) and other types of domination number of graphs.
To end this paper, we pose the following problem:
Problem 1. Study the extremal topological indices in the class of all connected graphs with a given Roman domination number.
This work was supported by the National Natural Science Foundation of China (Grant No. 11701505) and the Research Intensified Grant Scheme (RIGS), Phase 1/2019, Universiti Malaysia Terengganu, Malaysia with Grant Vot. 55192/6.
The authors declare no conflicts of interest.
[1] |
M. Alfaro-Ponce, A. Argüelles, I. Chairez, Pattern recognition for electroencephalographic signals based on continuous neural networks, Neural Networks, 79 (2016), 88–96. https://doi.org/10.1016/j.neunet.2016.03.004 doi: 10.1016/j.neunet.2016.03.004
![]() |
[2] |
T. Liang, Z. Liu, W. Wang, Pattern recognition of decorative elements based on neural network, J. Intell. Fuzzy Syst., 39 (2020), 8665–8673. https://doi.org/10.3233/jifs-189262 doi: 10.3233/jifs-189262
![]() |
[3] |
A. Chakravarty, J. Mentink, S. Semin, T. Rasing, Training and pattern recognition by an opto-magnetic neural network, Appl. Phys. Lett., 120 (2022), 022403. https://doi.org/10.1063/5.0073280 doi: 10.1063/5.0073280
![]() |
[4] |
A. Azad, L. Wang, N. Guo, H. Tam, C. Lu, Signal processing using artificial neural network for BOTDA sensor system, Opt. Express, 24 (2016), 6769–6782. https://doi.org/10.1364/oe.24.006769 doi: 10.1364/oe.24.006769
![]() |
[5] |
D. Auge, J. Hille, E. Mueller, A. Knoll, A survey of encoding techniques for signal processing in spiking neural networks, Neural Process. Lett., 53 (2021), 4693–4710. https://doi.org/10.1007/s11063-021-10562-2 doi: 10.1007/s11063-021-10562-2
![]() |
[6] |
B. Traore, B. Kamsu-Foguem, F. Tangara, Deep convolution neural network for image recognition, Ecol. Inform., 48 (2018), 257–268. https://doi.org/10.1016/j.ecoinf.2018.10.002 doi: 10.1016/j.ecoinf.2018.10.002
![]() |
[7] |
Y. Li, R. Zhou, R. Xu, J. Luo, W. Hu, A quantum deep convolutional neural network for image recognition, Quantum Sci. Technol., 5 (2020), 044003. https://doi.org/10.1088/2058-9565/ab9f93 doi: 10.1088/2058-9565/ab9f93
![]() |
[8] |
T. Zheng, Q. Wang, Y. Shen, X. Ma, X. Lin, Batch covariance neural network for image recognition, Image Vision Comput., 122 (2022), 104446. https://doi.org/10.1016/j.imavis.2022.104446 doi: 10.1016/j.imavis.2022.104446
![]() |
[9] |
H. Wei, R. Li, C. Chen, Z. Tu, Stability analysis of fractional order complex-valued memristive neural networks with time delays, Neural Process. Lett., 45 (2017), 379–399. https://doi.org/10.1007/s11063-016-9531-0 doi: 10.1007/s11063-016-9531-0
![]() |
[10] |
Y. Ke, Finite-time stability of fractional order BAM neural networks with time delay, J. Discret. Math. Sci. C., 20 (2017), 681–693. https://doi.org/10.1080/09720529.2017.1339435 doi: 10.1080/09720529.2017.1339435
![]() |
[11] |
L. Zhang, Y. Yang, Stability analysis of fractional order Hopfield neural networks with optimal discontinuous control, Neural Process. Lett., 50 (2019), 581–593. https://doi.org/10.1007/s11063-019-10054-4 doi: 10.1007/s11063-019-10054-4
![]() |
[12] |
A. Pratap, R. Raja, J. Cao, C. Huang, M. Niezabitowski, O. Bagdasar, Stability of discrete-time fractional-order time-delayed neural networks in complex field, Math. Method. Appl. Sci., 44 (2021), 419–440. https://doi.org/10.1002/mma.6745 doi: 10.1002/mma.6745
![]() |
[13] |
C. Chen, S. Zhu, Y. Wei, C. Chen, Finite-time stability of delayed memristor-based fractional-order neural networks, IEEE T. Cybernetics, 50 (2020), 1607–1616. https://doi.org/10.1109/tcyb.2018.2876901 doi: 10.1109/tcyb.2018.2876901
![]() |
[14] |
S. Tyagi, S. Martha, Finite-time stability for a class of fractional-order fuzzy neural networks with proportional delay, Fuzzy Set. Syst., 381 (2020), 68–77. https://doi.org/10.1016/j.fss.2019.04.010 doi: 10.1016/j.fss.2019.04.010
![]() |
[15] |
N. Thanh, P. Niamsup, V. Phat, New results on finite-time stability of fractional-order neural networks with time-varying delay, Neural Comput. Appl., 33 (2021), 17489–17496. https://doi.org/10.1007/s00521-021-06339-2 doi: 10.1007/s00521-021-06339-2
![]() |
[16] |
F. Du, J. Lu, New criteria for finite-time stability of fractional order memristor-based neural networks with time delays, Neurocomputing, 421 (2021), 349–359. https://doi.org/10.1016/j.neucom.2020.09.039 doi: 10.1016/j.neucom.2020.09.039
![]() |
[17] |
Z. Yang, J. Zhang, J. Hu, J. Mei, New results on finite-time stability for fractional-order neural networks with proportional delay, Neurocomputing, 442 (2021), 327–336. https://doi.org/10.1016/j.neucom.2021.02.082 doi: 10.1016/j.neucom.2021.02.082
![]() |
[18] |
A. Pratap, R. Raja, C. Sowmiya, O. Bagdasar, J. Cao, G. Rajchakit, Robust generalized Mittag-Leffler synchronization of fractional order neural networks with discontinuous activation and impulses, Neural Networks, 103 (2018), 128–141. https://doi.org/10.1016/j.neunet.2018.03.012 doi: 10.1016/j.neunet.2018.03.012
![]() |
[19] |
P. Anbalagan, R. Ramachandran, J. Cao, G. Rajchakit, C. Lim, Global robust synchronization of fractional order complex valued neural networks with mixed time varying delays and impulses, Int. J. Control Autom., 17 (2019), 509–520. https://doi.org/10.1007/s12555-017-0563-7 doi: 10.1007/s12555-017-0563-7
![]() |
[20] |
L. Zhang, Y. Yang, Finite time impulsive synchronization of fractional order memristive BAM neural networks, Neurocomputing, 384 (2020), 213–224. https://doi.org/10.1016/j.neucom.2019.12.056 doi: 10.1016/j.neucom.2019.12.056
![]() |
[21] | C. Popa, Octonion-valued neural networks, In: A. Villa, P. Masulli, A. Pons Rivero, Artificial Neural Networks and Machine Learning–ICANN 2016, Lecture Notes in Computer Science, Cham: Springer, 2016,435–443. https://doi.org/10.1007/978-3-319-44778-0_51 |
[22] | C. Popa, Global asymptotic stability for octonion-valued neural networks with delay, In: F. Cong, A. Leung, Q. Wei, Advances in Neural Networks–ISNN 2017, Lecture Notes in Computer Science, Cham: Springer, 2017,439–448. https://doi.org/10.1007/978-3-319-59072-1_52 |
[23] | C. Popa, Exponential stability for delayed octonion-valued recurrent neural networks, In: I. Rojas, G. Joya, A. Catala, Advances in Computational Intelligence–IWANN 2017, Lecture Notes in Computer Science, Cham: Springer, 2017,375–385. https://doi.org/10.1007/978-3-319-59153-7_33 |
[24] |
C. Popa, Global exponential stability of octonion-valued neural networks with leakage delay and mixed delays, Neural Networks, 105 (2018), 277–293. https://doi.org/10.1016/j.neunet.2018.05.006 doi: 10.1016/j.neunet.2018.05.006
![]() |
[25] |
C. Popa, Global exponential stability of neutral-type octonion-valued neural networks with time-varying delays, Neurocomputing, 309 (2018), 117–133. https://doi.org/10.1016/j.neucom.2018.05.004 doi: 10.1016/j.neucom.2018.05.004
![]() |
[26] |
J. Wang, X. Liu, Global μ-stability and finite-time control of octonion-valued neural networks with unbounded delays, arXiv, 12 (2020), 1950016. https://doi.org/10.48550/arXiv.2003.11330 doi: 10.48550/arXiv.2003.11330
![]() |
[27] |
S. Shishegar, R. Ghorbani, L. Saoud, S. Duchesne, G. Pelletier, Rainfall-runoff modelling using octonion-valued neural networks, Hydrolog. Sci. J., 66 (2021), 1857–1865. https://doi.org/10.1080/02626667.2021.1962885 doi: 10.1080/02626667.2021.1962885
![]() |
[28] |
U. Kandasamy, R. Rajan, Hopf bifurcation of a fractional-order octonion-valued neural networks with time delays, Discrete Cont. Dyn. Syst.-Ser. S, 13 (2020), 2537–2559. https://doi.org/10.3934/dcdss.2020137 doi: 10.3934/dcdss.2020137
![]() |
[29] |
L. Zhang, Y. Yang, Bipartite synchronization analysis of fractional order coupled neural networks with hybrid control, Neural Process. Lett., 52 (2020), 1969–1981. https://doi.org/10.1007/s11063-020-10332-6 doi: 10.1007/s11063-020-10332-6
![]() |
[30] |
B. Zheng, C. Hu, J. Yu, H. Jiang, Finite-time synchronization of fully complex-valued neural networks with fractional-order, Neurocomputing, 373 (2020), 70–80. https://doi.org/10.1016/j.neucom.2019.09.048 doi: 10.1016/j.neucom.2019.09.048
![]() |
[31] |
M. Hui, N. Yao, H. Iu, R. Yao, L. Bai, Adaptive synchronization of fractional-order complex-valued neural networks with time-varying delays, IEEE Access, 10 (2022), 45677–45688. https://doi.org/10.1109/access.2022.3170091 doi: 10.1109/access.2022.3170091
![]() |
[32] |
W. Zhang, H. Zhang, J. Cao, H. Zhang, F. Alsaadi, A. Alsaedi, Global projective synchronization in fractional-order quaternion valued neural networks, Asian J. Control, 24 (2022), 227–236. https://doi.org/10.1002/asjc.2485 doi: 10.1002/asjc.2485
![]() |
[33] |
W. Zhang, H. Zhao, C. Sha, Y. Wang, Finite time synchronization of delayed quaternion valued neural networks with fractional order, Neural Process. Lett., 53 (2021), 3607–3618. https://doi.org/10.1007/s11063-021-10551-5 doi: 10.1007/s11063-021-10551-5
![]() |
[34] |
J. Xiao, J. Cao, J. Cheng, S. Zhong, S. Wen, Novel methods to finite-time Mittag-Leffler synchronization problem of fractional-order quaternion-valued neural networks, Inform. Sci., 526 (2020), 221–244. https://doi.org/10.1016/j.ins.2020.03.101 doi: 10.1016/j.ins.2020.03.101
![]() |
[35] |
H. Li, H. Jiang, J. Cao, Global synchronization of fractional-order quaternion-valued neural networks with leakage and discrete delays, Neurocomputing, 385 (2020), 211–219. https://doi.org/10.1016/j.neucom.2019.12.018 doi: 10.1016/j.neucom.2019.12.018
![]() |
[36] |
G. Narayanan, M. Syed Ali, M.Alam, G. Rajchakit, N. Boonsatit, P. Kuamr, et al., Adaptive fuzzy feedback controller design for finite-time Mittag-Leffler synchronization of fractional-order quaternion-valued reaction-diffusion fuzzy molecular modeling of delayed neural networks, IEEE Access, 9 (2021), 130862–130883. https://doi.org/10.1109/access.2021.3113915 doi: 10.1109/access.2021.3113915
![]() |
[37] |
C. Wang, H. Zhang, H. Zhang, W. Zhang, Globally projective synchronization for Caputo fractional quaternion-valued neural networks with discrete and distributed delays, AIMS Math., 6 (2021), 14000–14012. https://doi.org/10.3934/math.2021809 doi: 10.3934/math.2021809
![]() |
[38] |
H. Yan, Y. Qiao, L. Duan, J. Miao, New inequalities to finite-time synchronization analysis of delayed fractional-order quaternion-valued neural networks, Neural Comput. Appl., 34 (2022), 9919–9930. https://doi.org/10.1007/s00521-022-06976-1 doi: 10.1007/s00521-022-06976-1
![]() |
[39] | I. Podlubny, Fractional differential equations, San Diego: Academic Press, 1998. |
[40] |
S. Liang, R. Wu, L. Chen, Adaptive pinning synchronization in fractional order uncertain complex dynamical networks with delay, Phys. A., 444 (2016), 49–62. https://doi.org/10.1016/j.physa.2015.10.011 doi: 10.1016/j.physa.2015.10.011
![]() |
1. | Jian Yang, Yuefen Chen, Zhiqiang Li, Some sufficient conditions for a tree to have its weak Roman domination number be equal to its domination number plus 1, 2023, 8, 2473-6988, 17702, 10.3934/math.2023904 |