Loading [MathJax]/jax/output/SVG/jax.js
Research article Special Issues

Generalized Reed-Solomon codes over number fields and exact gradient coding

  • This paper describes generalized Reed-Solomon (GRS) codes over number fields that are invariant under certain permutations. We call these codes generalized quasi-cyclic (GQC) GRS codes. Moreover, we describe an application of GQC GRS codes over number fields to exact gradient coding.

    Citation: Irwansyah, Intan Muchtadi-Alamsyah, Fajar Yuliawan, Muhammad Irfan Hidayat. Generalized Reed-Solomon codes over number fields and exact gradient coding[J]. AIMS Mathematics, 2024, 9(4): 9508-9518. doi: 10.3934/math.2024464

    Related Papers:

    [1] Shaofang Hong, Rongjun Wu . On deep holes of generalized Reed-Solomon codes. AIMS Mathematics, 2016, 1(2): 96-101. doi: 10.3934/Math.2016.2.96
    [2] Jing Huang, Jingge Liu, Dong Yu . Dimensions of the hull of generalized Reed-Solomon codes. AIMS Mathematics, 2024, 9(6): 13553-13569. doi: 10.3934/math.2024661
    [3] Xiaofan Xu, Yongchao Xu, Shaofang Hong . Some results on ordinary words of standard Reed-Solomon codes. AIMS Mathematics, 2019, 4(5): 1336-1347. doi: 10.3934/math.2019.5.1336
    [4] Xiaofan Xu, Yongchao Xu . Some results on deep holes of generalized projective Reed-Solomon codes. AIMS Mathematics, 2019, 4(2): 176-192. doi: 10.3934/math.2019.2.176
    [5] Claude Carlet . Identifying codewords in general Reed-Muller codes and determining their weights. AIMS Mathematics, 2024, 9(5): 10609-10637. doi: 10.3934/math.2024518
    [6] Xuesong Si, Chuanze Niu . On skew cyclic codes over M2(F2). AIMS Mathematics, 2023, 8(10): 24434-24445. doi: 10.3934/math.20231246
    [7] Wei Qi . The polycyclic codes over the finite field Fq. AIMS Mathematics, 2024, 9(11): 29707-29717. doi: 10.3934/math.20241439
    [8] Guanghui Zhang, Shuhua Liang . On the construction of constacyclically permutable codes from constacyclic codes. AIMS Mathematics, 2024, 9(5): 12852-12869. doi: 10.3934/math.2024628
    [9] Adel Alahmadi, Tamador Alihia, Patrick Solé . The build up construction for codes over a non-commutative non-unitary ring of order 9. AIMS Mathematics, 2024, 9(7): 18278-18307. doi: 10.3934/math.2024892
    [10] Ismail Aydogdu . On double cyclic codes over Z2+uZ2. AIMS Mathematics, 2024, 9(5): 11076-11091. doi: 10.3934/math.2024543
  • This paper describes generalized Reed-Solomon (GRS) codes over number fields that are invariant under certain permutations. We call these codes generalized quasi-cyclic (GQC) GRS codes. Moreover, we describe an application of GQC GRS codes over number fields to exact gradient coding.



    Data-intensive machine learning has become widely used, and as the size of training data increases, distributed methods are becoming increasingly popular. However, the performance of distributed methods is mainly determined by stragglers, i.e., nodes that are slow to respond or are unavailable.

    Raviv et al. [11] used coding theory and graph theory to reduce stragglers in distributed synchronous gradient descent. A coding theory framework for straggler mitigation, called gradient coding, was first introduced by Tandon et al. [14]. Gradient coding consists of a system with one master and n worker nodes, where the data are partitioned into k parts, and one or more parts are assigned to each worker. In turn, each worker computes the partial gradients on each given partition, combines the results linearly according to a predefined vector of coefficients, and sends this linear combination back to the primary node. By choosing the coefficients at each node appropriately, it can be guaranteed that the primary node can reconstruct the full gradient even if a machine fails to do its job.

    The importance of straggler mitigation is demonstrated in [8,16]. Specifically, it was shown by Tandon et al. [14] that stragglers run up to 5 times slower than the performance of typical workers (8 times in [16]). In [11], for gradient calculations, a cyclic maximum distance separable (MDS) code is used to obtain a better deterministic construction scheme than existing solutions, both in the range of parameters that can be applied and in the complexity of the algorithms involved.

    One well-known family of MDS codes is generalized Reed-Solomon (GRS) codes. GRS codes have interesting mathematical structures and many real-world applications, such as mass storage systems, cloud storage systems, and public-key cryptosystems. On the other hand, although more complex than cyclic codes, quasi-cyclic codes satisfy the condition of the Gilbert-Varshamov lower bound at minimum distances, as shown in [6]. Quasi-cyclic codes are also equivalent to linear codes with circulant block generator matrices. This type of matrix has circular blocks of the same size, such as m, which denotes the co-indexes of the associated quasi-cyclic code. From this point of view, one way to generalize quasi-cyclic codes is to let the generator matrix have circular blocks of different sizes. This code is called a generalized quasi-cyclic code with shared indices (m1,m2,,mk), where m1,m2,,mk represents the size of the circular block in the generator matrix.

    In [10], a generalized quasi-cyclic code without block length limitations is studied. By relaxing the conditions on block length, several new optimal codes with small lengths could be found. In addition, the code decomposition and dimension formulas given by [3,12,13] have been generalized.

    In this paper, we describe the construction of generalized quasi-cyclic GRS codes over totally real number fields, as well as their application in exact gradient coding. The construction method is derived by integrating known results from the inverse Galois problem for totally real number fields. Furthermore, methods in [2,4,11,14] will be adapted to generalized quasi-cyclic GRS codes to mitigate stragglers.

    Let F be a Galois extension of Q and choose non-zero elements v1,,vn in F and distinct elements a1,,an in F. Also, let v=(v1,,vn) and a=(a1,,an). For 1kn, define the GRS codes as follows:

    GRSn,k(a,v)={(v1f(a1),,vnf(an))|f(x)F[x]k},

    where F[x]k is the set of all polynomials over F with degree less than k. The canonical generator of GRSn,k(a,v) is given by the following matrix:

    G=(v1v2vjvnv1a1v2a2vjajvnanv1a21v2a22vja2jvna2nv1ai1v2ai2vjaijvnainv1ak11v2ak12vjak1jvnak1n) (2.1)

    Theorem 2.1. [7] Let vFn be a tuple of non-zero elements in F and aFn be a tuple of pairwise distinct elements in F; then,

    a) The GRSn,k(a,v) is a [n,k,nk+1] code, i.e., GRS codes are MDS codes.

    b) The dual code of GRSn,k(a,v) is as follows:

    GRSn,k(a,v)=GRSn,nk(a,u),

    where u=(u1,,un) with

    u1i=viji(aiaj).

    Proof. (a) See the proof of [7, Theorem 6.3.3]. (b) See the proof of [7, Theorem 6.5.1].

    Let ¯F=F{} and a be an n-tuple of mutually distinct elements of ¯F, and let c be an n-tuple of non-zero elements of F. Also, define

    [ai,aj]=aiaj,[,aj]=1[ai,]=1for allai,ajF.

    Definition 2.2. ([9]) Let B(a,c) be the k×(nk) matrix with the following entries:

    cj+kci[aj+k,ai],for1ik,1jnk.

    The generalized Cauchy code Ck(a,c) is an [n,k,nk+1] code defined by the generator matrix (Ik|B(a,c)).

    The following proposition shows that the GRS codes are also generalized Cauchy codes.

    Proposition 2.3. [9, Proposition C.2] Let a be an n-tuple of mutually distinct elements of ¯F, and let c be an n-tuple of non-zero elements of F. Also, let

    ci={bikt=1,ti[ai,at],if1ik;bikt=1[ai,at],ifk+1in.

    Then, GRSn,k(a,b)=Ck(a,c).

    Let Gal(F/Q) be the Galois group of F over Q and PΓL(2,F) denote the group of semilinear fractional transformations given by

    f:¯F¯Fxaγ(x)+bcγ(x)+d,

    where adbc0 and γGal(F/Q). Let Sn be the symmetric group on a set of n elements and Per(C)={ξSn|ξ(C)=C}, where n is the length of the code C. The set Per(C) is called the permutation group of the code C. We have the following theorem that is related to the permutation group of a Cauchy code.

    Theorem 2.4. [1, Corollary 2] Let C=Ck(a,y) be a Cauchy code over F, where 2kn2 and a=(a1,,an). Also, let L={a1,,an}. Then, the map

    ω:{fPΓL(2,F)|f(L)=L}Per(C)fσ,

    where aσ(i)=f(ai) for i=1,,n is a surjective group homomorphism.

    A number field F is a finite Galois extension of the rational field Q. In this section, we describe a way to construct a number field F with Gal(F/Q)σ for σSn, where σ is a cyclic subgroup generated by σ.

    Let σ=σ1σ2σt be a permutation in Sn, where σ1,σ2,,σt are disjoint cycles. Also, let σ be the cyclic group generated by σ. Let l(σj) be the length of the cycle σj, and define a set P={p:pprime andj{1,,t}p|l(σj)}. Since P is finite, assume that p1<p2<<p|P| are all elements in P. For any j, we have

    l(σj)=|P|i=1pαiji, (3.1)

    where αijZ0. Based on Eq (3.1), we have

    ord(σ)=|σ|=|P|i=1pmaxj{αij}i, (3.2)

    where ord(σ) is the order of the permutation σ. Since σ contains the element of order pmaxj{αij}i for all i=1,,|P|, by the structure theorem for finite Abelian groups, we have

    σ|P|i=1Zpmaxj{αij}iZZ|P|i=1pmaxj{αij}iZ. (3.3)

    Let ζp be the primitive p-th root of unity and Q(ζp) be the corresponding cyclotomic extension of Q. The following theorem shows a Galois extension of Q, where its Galois group is isomorphic to σ. The proof of the theorem is similar to the proof of [5, Theorem 3.1.11]. We write the proof here to give a sense of how to construct the related Galois extension.

    Theorem 3.1. There exists a totally real Galois extension K of Q such that Gal(K/Q)σ.

    Proof. By Eq (3.3), we have

    σ|P|i=1Zpmaxj{αij}iZZ|P|i=1pmaxj{αij}iZ.

    Now, choose a prime p such that

    p1mod2|P|i=1pmaxj{αij}i.

    Let ζp be the p-th root of unity. By [5, Theorem C.0.3], Q(ζp) is a Galois extension of Q, with its corresponding Galois group being isomorphic to G=(Z/pZ)×, where (Z/pZ)× is the multiplicative group of Z/pZ{¯0}. Since p is a prime number, G is a cyclic group. Moreover, we can find a unique subgroup H of G such that

    |H|=p1|P|i=1pmaxj{αij}i.

    Let Q(ζp)H be a subset of Q(ζp) which is invariant under the action of H. By the fundamental theorem of Galois theory ([15, Theorem 25]), Q(ζp)H is also a Galois extension of Q, with the corresponding Galois group isomorphic to G/H. Moreover, |G/H|=|P|i=1pmaxj{αij}i, and, as a consequence,

    G/HZ|P|i=1pmaxj{αij}iZσ.

    Also, by using a similar argument as in the proof of [5, Theorem 3.1.11], we have that Q(ζp)H is a totally real Galois extension of Q. The following algorithm provides a way to construct Q(ζp)H in the proof of Theorem 3.1. The algorithm is based on Theorem 3.1 and [5, Proposition 3.3.2].

    Algorithm 3.2. Suppose that σSn and G=Gal(Q(ζp)/Q)(Z/pZ)×, where p is a prime number such that p1mod2ord(σ).

    1) Choose HG, where H is the subgroup of G with order p1ord(σ).

    2) Calculate

    α=λHλ(ζp).

    3) Find minimal polynomial mα(x) of α over Q.

    4) Construct the splitting field F of mα(x) by using Algorithm A.1.

    5) Then, F=Q(ζp)H.

    In this section, we describe a way to construct an invariant GRS code under a given permutation in Sn. We call this GRS code the GRS generalized quasi-cyclic (GQC) code. Let σ=σ1,σ2,,σt be a permutation in Sn, where σ1,σ2,,σt are disjoint cycles. Also, let G=σ be a cyclic group generated by σ.

    Theorem 4.1. If σ is a permutation in Sn, then there exists a GQC GRSn,k(¯α,b) over F, with its corresponding permutation being σ for some totally real number field F.

    Proof. We can find the number field F and its corresponding minimal polynomial mα(x) with Gal(F/Q)σ by using Algorithm 3.2. Since Gal(F/Q)σ, there exists γGal(F/Q) to be associated with σσ. Let L={α1,,αn} be the roots of mα(x) and some additional elements from linear combinations of the roots. We can see that γ is a permutation on L, i.e., γ(L)=L. Note that the orbit of L under H can be used to rearrange the elements of L such that

    γ(αi)=ασ(i), (4.1)

    for all i=1,2,,n. Let ¯α=(α1,α2,,αn) and b=(b1,b2,,bn) be an n-tuple of non-zero elements in F. Define a Cauchy code Ck(¯α,c), where c=(c1,c2,,cn), with

    ci={bikt=1,ti[αi,αt],if1ik;bikt=1[αi,αt],ifk+1in. (4.2)

    Then, by Proposition 2.3, Ck(¯α,c) is a GRSn,k(¯α,b) code. Moreover, according to Theorem 2.4 and Eq (4.1), ω(γ)=σ is an element in Per(Ck(¯α,c)).

    Consider the following example.

    Example 4.2. Let σ=(1,2,3,4)(5,6) in S6. We would like to construct a GRS code of length 6 over a totally real number field that is invariant under the action of σ. We can see that ord(σ)=4 and σ=Z/4Z. Choose p=17 so that p1mod2×4. The corresponding subgroup H of Gal(Q(ζ17)/Q) will have the order equal to 4. Since the unique subgroup of (Z/17Z)× with order 4 is {1,4,13,16}, we have

    H={λk|k=1,4,13,16},

    where λk:ζ17ζk17. Then, we have

    α=λHλ(ζ17)=ζ17+ζ417+ζ1317+ζ1617.

    From [5, Example 3.3.3], the minimal polynomial of α is as follows:

    mα(x)=x4+x36x2x+1.

    The roots of mα(x) given by

    r1=14(11734+17),r2=14(117+34+17),
    r3=14(1+173417).,r4=14(1+17+3417).

    Let γ be a map such that

    r1r2,r2r3,r3r4,r4r1.

    We can see that γ=Gal(Q(ζ17)H/Q)Z/4Z.

    Choose L={α1,,α6}, where αi=ri for i=1,2,3,4,α5=r1+r3, and α6=r2+r4. We can check that

    γ(αi)=ασ(i),

    for all i=1,,6. Take ¯α=(α1,,α6), any n-tuple of non-zero elements b (from the set of linear combinations of roots of mα(x)), and c=(c1,,c6), where ci is as in Eq 4.2. We have that Ck(¯α,c) is a GQC GRS code with corresponding permutation σ.

    In Section 4, we described the construction of GRS code, which is invariant under the action of a given permutation in Sn. Moreover, the alphabet for the corresponding codes is a totally real number field, not a complex number field. This feature can be useful for bandwidth reduction in exact gradient coding schemes.

    Algorithm 1 describes the process of gradient coding. The algorithm is a slight modification of [11, Algorithm 1].

    Algorithm 1 Gradient coding
    Input:
      Data S={zi=(xi,yi)}mi=1, number of iterations t>0, learning rate {η}tr=1,
      straggler tolerance parameter {sr}tr=1, a matrix BCn×n, a function
      Λ:P(n)Cn, a vector of non-zero elements ¯β=(β1,,βn)Cn
    Initialize:
      w(1)(0,0,,0)
    Partition S=ni=1Si and send {Sj|jsupp(bi)} to Wi for every i[n]
    for r=1 to t do
      M broadcasts w(r) to all nodes
      Each Wj sends isupp(bj)bj,iLSi(w(r))βi to M
      M waits until at least nsr nodes have responded
      M computes vr=Λ(Kr)C, where the i-th row of C is 1n times the response from Wi if it has responded, and 0 otherwise; also, Kr is the set of non-stragglers in the current iteration r
      M updates w(r+1)w(r)ηrvr
    end for
    return 1ttr=1w(r+1)

    Algorithm 1 works in the following way. In order to execute the gradient descent process, the master node M distributes a particular partition of the training set S to all worker nodes Wj, where j=1,,n. In the r-th iteration of the gradient descent process, the master M broadcasts the parameter w(r) to all worker nodes. Using the received parameter w(r), the worker node Wj calculates the partial gradient LSi(w(r)) and sends its linear combination isupp(bj)bj,iLSi(w(r))βi to M. The linear combination is chosen from the entries bj,i of a particular matrix B. In this work, B is constructed by using GRS codes which are invariant under the action of a particular permutation. After M has received the linear combinations of partial gradients from some number of worker nodes, M updates the parameter w by using the decoding vector Λ(Kr),w(r), and some other additional vectors (mentioned in the algorithm). Note that we will see later that the decoding vector Λ(Kr) can be computed by using Algorithm 2[11, Algorithm 2].

    Definition 5.1. A matrix BCn×n and a function Λ:P(n)Cn satisfy the exact computation (EC) condition with respect to ¯βCn, where ¯β is an n-tuple of non-zero elements in Cn if, for all K[n] such that |K|maxr[t]sr, we have that Λ(K)B=¯β.

    Note that Definition 5.1 is a slight modification of [11, Definition 2]. Let ¯β=(β1,,βn) be an n-tuple of non-zero elements of Cn and

    N¯β(w)=1n(LS1(w)β1LS2(w)β2LSn(w)βn).

    Lemma 5.2. If Λ and B satisfy the EC condition with respect to ¯β, then, for all r[t], we have that vr=LS(w(r)).

    Proof. Given r[t], let B be the matrix whose i-th row bi equals to bi if iKr, and 0 otherwise. The matrix C in Algorithm 1 can be written as C=BN¯β(w(r)). Since supp(Λ(Kr))Kr, we have that Λ(Kr)B=Λ(Kr)B. Therefore, we have

    vr=Λ(Kr)C=Λ(Kr)BNβ(w(r))=βNβ(w(r))=1nni=1LSi(w(r))=1nni=11m/nzSil(w(r),z)=1mzSl(w(r),z)=LS(w(r)).

    For a given n and s, let C=GRSn,ns(¯α,¯β) GQC code over a number field F with corresponding permutation π of order n. Clearly, the vector ¯β is in C. Moreover, by [11, Lemma 8], there exists a codeword c1 in C whose support is {1,2,,s+1}. Let ci=πi1(c1) for i=2,,n and B=(cT1,cT2,,cTn).

    Theorem 5.3. The matrix B satisfies the following properties:

    a) Each row of B is a codeword in σ(C), where σ is a permutation such that

    σ1=(123innπn1(n)πn2(n)πn(i1)(n)π(n)). (5.1)

    b) wH(b)=s+1 for each row b in B.

    c) The column span of B is the code C.

    d) Every set of ns rows of B are linearly independent over F.

    Proof. (a) Let c1=(c1,,cn). Notice that the i-th row of B is as follows:

    (ci,cπn1(i),cπn2(i),,cπ(i)).

    Since ord(π)=n, the i-th row of B is a permutation of c1 for all i=1,,n. Moreover, by considering the last row of B, we can see that all rows of B constitute a codeword in σ(C), where σ is the permutation as in Eq (5.1).

    (b) By part (a), we have that the Hamming weight of every row of B is $ s+1.

    (c) Let σ=(1,2,,n) be a cyclic permutation and G1 be a cyclic group generated by σ. Also, let G2 be a cyclic group generated by π. Define ¯S1=span(G1c1) and ¯S2=span(G2c1), where Gc1={λ(c1)|λG}. Since ord(σ)=ord(π)=n, we have that G1G2 by the following group isomorphism:

    τ:G1G2σiπi.

    Define the following map:

    ¯τ:¯S1¯S2ni=1αiσi(c1)ni=1αiπi(c1).

    The map ¯τ is a linear map. Since it is induced by τ,¯τ is a bijective map. So, ¯S1¯S2. By [11, Lemma 12 B3], ¯S1=C. Since ¯S2C and dim¯S2=ns, we have that ¯S2=C.

    (d) Similar to [11, Lemma 12 B4].

    Let G be the canonical generator for the C=GRSn,ns(¯α,¯β) GQC code, as in Eq (2.1). By Theorem 2.1(b), the canonical generator for the dual code C is G=GD, where D=diag(u1,,un), with

    ui=1β2iji(αiαj)

    for all i=1,,n. Using this setting, Algorithm 2[11, Algorithm 2] can be used to compute the decoding vector a(K).

    The authors declare they have not used Artificial Intelligence (AI) tools in the creation of this article.

    This research was funded by Hibah PPMI KK Aljabar Institut Teknologi Bandung 2023.

    The authors declare no conflict of interest.

    The following algorithm provides a way to construct the unique splitting field of a given polynomial f(x) in Q[x].

    Algorithm A.1. Given a polynomial f(x) in Q[x], we will construct the splitting field L of f(x) based on the construction of a chain of number fields:

    K0=QK1K2Ks1Ks=L

    such that Ki is an extension of Ki1 containing a new root of f(x).

    1) Factorize f(x) over Ki into irreducible factors f1(x)f2(x)ft(x).

    2) Choose any non linear irreducible factor g(x)=fj(x) for some j{1,,t}.

    3) Construct the field extension Ki+1=Ki[x]g(x).

    4) Repeat the process for Ki+1 until f(x) completely factors.

    The following algorithm can be used to compute the decoding vector in the exact gradient coding scheme [11, Algorithm 2].

    Algorithm 2 Computing decoding vector Λ(K)
    Data: any vector xCn such that xB=β
    Input:
      A set K[n] of ns non-stragglers
    Output: a vector Λ(K) such that supp(Λ(K))K and Λ(K)B=β
    find fCs such that fGKc=xKcD1Kc
    yfGD
    return Λ(K)y+x



    [1] A. Dür, The automorphism groups of Reed-Solomon codes, J. Comb. Theory A, 44 (1987), 69–82. https://doi.org/10.1016/0097-3165(87)90060-4 doi: 10.1016/0097-3165(87)90060-4
    [2] S. Dutta, V. Cadambe, P. Grover, Short-dot: Computing large linear transforms distributedly using coded short dot products, IEEE T. Inform. Theory, 65 (2019), 6171–6193. https://doi.org/10.1109/TIT.2019.2927558 doi: 10.1109/TIT.2019.2927558
    [3] C. Güneri, F. Özbudak, B. Özkaya, E. Sacikara, Z. Sepasdar, P. Solé, Structure and performance of generalized quasi-cyclic codes, Finite Fields Th. App., 47 (2017), 183–202. https://doi.org/10.1016/j.ffa.2017.06.005 doi: 10.1016/j.ffa.2017.06.005
    [4] W. Halbawi, Error-correcting codes for networks, storage, and computation, PhD Thesis, California Institute of Technology, 2017. https://doi.org/10.7907/Z9J67F08
    [5] S. Kalyanswamy, Inverse Galois problem for totally real number fields, PhD Thesis, Cornell University, 2012.
    [6] T. Kasami, A Gilberd-Varshamov bound for quasi cyclic codes of rate 1/2, IEEE T. Inform. Theory, 20 (1974), 679. https://doi.org/10.1109/TIT.1974.1055262 doi: 10.1109/TIT.1974.1055262
    [7] S. Loepp, W. K. Wooters, Protecting information: from classical error correction to quantum cryptography, Cambridge: Cambridge University Press, 2006. https://doi.org/10.1017/CBO9780511813719
    [8] M. Li, D. G. Andersen, A. Smola, K. Yu, Communication efficient distributed machine learning with the parameter server, Proceedings of the 27th International Conference on Neural Information Processing Systems, 1 (2014), 19–27.
    [9] I. Márquez-Corbella, R. Pellikaan, A characterization of MDS codes that have an error-correcting pair, Finite Fields Th. App., 40, (2016), 224–245. https://doi.org/10.1016/j.ffa.2016.04.004
    [10] I. Muchtadi-Alamsyah, Irwansyah, A. Barra, Generalized quasi-cyclic codes with arbitrary block lengths, Bull. Malays. Math. Sci. Soc., 45 (2022), 1383–1407. https://doi.org/10.1007/s40840-022-01251-x doi: 10.1007/s40840-022-01251-x
    [11] N. Raviv, I. Tamo, R. Tandon, A. G. Dimakis, Gradient coding from cyclic MDS codes and expander graph, IEEE T. Inform. Theory, 66 (2020), 7475–7489. https://doi.org/10.1109/TIT.2020.3029396 doi: 10.1109/TIT.2020.3029396
    [12] G. E. Séguin, The algebraic structure of codes invariant under a permutation, In: Information theory and applications II, Heidelberg: Springer, 1995, 1–18. https://doi.org/10.1007/BFb0025131
    [13] I. Siap, N. Kulhan, The structure of generalized quasi-cyclic codes, Appl. Math. E-Notes, 5 (2005), 24–30.
    [14] R. Tandon, Q. Lei, A. Dimakis, N. Karampatziakis, Gradient coding: Avoiding stragglers in distributed learning, Proceedings of the 34th International Conference on Machine Learning, 70 (2017), 3368–3376.
    [15] H. G. J. Tiesinga, The inverse Galois problem, PhD Thesis, University of Groningen, 2016. Available from:
    [16] N. J. Yadwarkar, B. Hariharan, J. E. Gonzales, R. Katz, Multitask learning for straggler avoiding predictive job scheduling, J. Mach. Learn. Res., 17 (2016), 3692–728.
  • Reader Comments
  • © 2024 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(1291) PDF downloads(107) Cited by(0)

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog