Nippon Telegraph and Telephone Corporation

Ntt Corporation

Dai Ikarashi

Koki Hamada

Ryo Kikuchi

Koji Chida

Naoto Kiribuchi

Hiroki Sudo

Gembu Morohashi

Ibuki Mishina

Hitoshi Fuji

Katsumi Takahashi

Junko Hashimoto

Atsunori Ichikawa

Hiroyoshi Takiguchi

Gen Takahashi

Satoshi Tanaka

Asami Miyajima

Teruko Miyata

Shigeru Kayaguchi

By using [x], . . . , [x], . . . , [x], . . . , [x] obtained by concealing L sets X={x, . . . , x}, . . . , X={x, . . . , x}, [c], . . . , [c] obtained by concealing the number cof elements representing kamong x, . . . , x, . . . , x, . . . , xfor p=0, . . . , m−1 are obtained, [eq], . . . , [eq] obtained by concealing eq=T when cis L and eq=F when cis not L for p=0, . . . , m−1 are obtained, and a data structure including [k] and [eq] associated with each other is output as a concealed operation result of a product set of X, . . . , X.

Secure computation apparatus, secure computation system, secure computation method, and program

A secret equi-join device uses as inputs a first concealed table that is concealed information of a first table having a plurality of keys and a second concealed table that is concealed information of a second table having a plurality of keys, obtains a first concealed additional table that is concealed information of a first additional table which is obtained by adding a subkey column to the first table, obtains by secure computation a second concealed additional table that is concealed information of a second additional table which is obtained by adding a subkey column further to a third table obtained by copying each record in the second table a plurality of times and adding the copied records, and obtains a concealed joined table that is concealed information of a joined table which is obtained by equi-joining the first additional table and the second additional table.

Secure equi-join apparatus, secure equi-join method, and program

A secret share value [y]=[δx+ax] is obtained through secure computation using a secret share value [x] of a real number x, and a secret share value [func(x)]=[y(ζy+b)+cx] of an elementary function approximation value z=func(x) of the real number x is obtained and output through secure computation using secret share values [x] and [y]. Here, x, y, and z are real numbers, a, b, c, δ, and ζ are real number coefficients, and a secret share value of · is [·].

Secure computation apparatus, secure computation method, and program

A secure computation deviceof a secure computation system according to an aspect of this invention includes a first calculation unit, a second calculation unit, a third calculation unit, a fourth calculation unit, a fifth calculation unit, a sixth calculation unit, a seventh calculation unit, and an output unit. By calculation being performed in cooperation of these, a group by count operation can be performed on a table to which a flag is added.

Secret calculation system, apparatus, method and program

A calculation of a gradient descent method in secure computing is performed at high speed while maintaining accuracy. A secure gradient descent computation method calculates a gradient descent method while keeping a gradient and a parameter concealed. An initialization unit initializes concealed values [M], [V] of matrices M, V (S). A gradient calculation unit determines concealed value [G] of a matrix G of a gradient g (S). A parameter update unit calculates [M] β1 [M]+(1−β1) [G] (S-), calculates [V]←β2 [V]+(1−β2) [G]◯[G] (S-), calculates [M{circumflex over ( )}]←β{circumflex over ( )}1, t [M] (S-), calculates [V{circumflex over ( )}]←β{circumflex over ( )}2, t [V] (S-), calculates [G{circumflex over ( )}]←Adam ([V{circumflex over ( )}]) (S-), calculates [G{circumflex over ( )}]←[G{circumflex over ( )}]◯[M{circumflex over ( )}] (S-), and calculates [W]←[W]−[G{circumflex over ( )}] (S-).

Machine learning using secure gradient descent computation

A concealed operation result indicating concealed information of an Intersect operation result of X and Y is obtained while X={{x, . . . , x}} and Y={{y, . . . , y}} are concealed. A secure computation device obtains a sequence ([s], [M]) including [s] including n [B] and m [B] and [M] including [x], . . . , [x] and [y], . . . , [y], performs stable sorting on the sequence ([s], [M]) according to an order relationship of content represented by each of the elements M, . . . , Mto obtain a sequence ([s'], [M′]), obtain [eq] where eq=T when M′=M′and eq=F otherwise and [seq] where seq=T when s′=s′and seq=F otherwise, obtain [f] where f=Dwhen eq=T and seq=F and f=Dotherwise, and outputs [f] and [M′]. Where, [α] is concealed information of α.

A set SP(i) of k secure computation apparatuses selected from a set PA of n secure computation apparatuses repeats processing of multiplying a share according to a secret sharing scheme, by power-of-2 number 2ρ(SP(i)) with the share ρ(SP(i)) as an exponent, the share ρ(SP(i)) being obtained by secret sharing of ρ according to a replicative secret sharing scheme, and redistribution the value obtained in the processing is redistributed into the set SP(i+1) of k secure computation apparatuses selected from the set PA of n secure computation apparatuses. However, the final time is not re-dispersed. Thus, a share of a multiplicative rotation result is obtained. However, the final time is not re-dispersed. Thus, a share of a multiplicative rotation result is obtained.

Secure computation system, secure computation apparatus, secure computation method, and program

The present invention implements high-speed right shift computation and division in secure computation. According to the present invention, a public value multiplication part calculates [a']=[2a] from a distributed value [a] of a value 'a.' A first conversion part converts [a′] into additive secret sharing. A right shift computation part calculates <s>=<a′>>>b+u. A second conversion part converts <s> into linear secret sharing. A first bit conversion part converts lower u bits of <a′>into {a′mod 2}. A quotient transfer partobtains lower u bits of −Σ{a′mod 2} as {q}. A second bit conversion part converts lower b+u bits of <a′>into {a′}={a′mod 2}. An addition part calculates {z}=Σ{a′}+{q}, and obtains a bit sequence {z} of a (b+u)-th bit and after of {z}. A third conversion part converts {q} and {z} into linear secret sharing. An output computation part outputs [s]−[q]+[z] as [a>>b].

Secure right shift computation system, secure division system, methods therefor, secure computation apparatus, and program

A secret share value [f(x)−f'(x)] of f(x)−f′(x) is obtained through secure computation using a secret share value [x] of a real number x, and a secret share value [f(x)−f′(x)]of (f(x)−f′(x))obtained by right-shifting f(x)−f′(x) by the predetermined number of bits is obtained through secure computation using the secret share value [f(x)−f′(x)]. Here, [μ] is a secret share value of μ, n is an integer equal to or greater than 1, t=0, . . . , n−1, u=1, . . . , n−1, f(x) is a function of the real number x, f′(x) is an approximation function of the function f(x), a secret share value [f′(x)] of an approximation function f′(x) is [f′(x)]=c+c[x], a secret share value [f′(x)] of an approximation function f′(x) is [f′(x)]=c+c[x]+c[f(x)]+ . . . +c[f(x)], cis a public value, and c, . . . , care coefficients.

A public value 2/m is obtained, and secure computation of public value division [x]/(2/m) using a secret share value [x] and the obtained public value 2/m is performed, so that a secret share value [mx]of a value obtained by right-shifting mx by σ bits is obtained and output. Here, x is a real number, [•] is a secret share value of •, σ is a positive integer that is the number of bits indicating a right shifting amount, and m is a real number.

The present invention provides techniques to calculate the number of surviving and the number of deaths while still concealing survival time data. The present invention includes: a group data position calculation means configured to calculate a share [[g]] of a sequence gand a share [[g]] of a sequence grepresented by predetermined equations from a share [[g]] of a sequence g of values of group of survival time data included in a survival time data set D; a group data number calculation means configured to calculate a share [[s]] and a share [[s]] from a share [[t]] of a sequence t of values of time of survival time data included in the survival time data set D, the share [[g]], and the share [[g]], by [[s]]=GroupSum ([[g]], [[t]]), [[s]]=GroupSum ([[g]], [[t]]); and a survival number calculation means.

Secret survival data processing system, secret survival data processing apparatus, secret survival data processing method, and program

A technique for performing secure computation of a sigmoid function with high speed and precision is provided. A secret sigmoid function calculation system is a system, in which g(x) is assumed to be a secure computable function, which calculates, from a share [[x]] of an input value x, a share [[σ'(x)]] of a value of a sigmoid function for the input value x, and includes: a first comparing means that generates a first comparison result [[c]]=less_than([[x]], t); a second comparing means that generates a second comparison result [[d]]=greater_than([[x]], t); a first logical computation means that generates a first logical computation result [[e]]=not([[c]]); a second logical computation means that generates a second logical computation result [[k]]=and([[c]], [[d]]) or [[k]]=mul([[c]], [[d]]); and a function value calculating means that calculates the share [[σ′(x)]]=mul([[k]], [[g(x)]])+[[e]].

Secret sigmoid function calculation system, secret logistic regression calculation system, secret sigmoid function calculation apparatus, secret logistic regression calculation apparatus, secret sigmoid function calculation method, secret logistic regression calculation method, and program

A secure joining system is a secure joining system including a plurality of secure computing apparatuses. The plurality of secure computing apparatuses include a vector joining unit, a first vector generation unit, a first permutation calculation unit, a first permutation application unit, a second vector generation unit, a third vector generation unit, a second permutation calculation unit, a second permutation application unit, a fourth vector generation unit, a fifth vector generation unit, a first inverse permutation application unit, a first vector separation unit, a second inverse permutation application unitand a second vector separation unit, a third permutation application unit, a fourth permutation application unit, and a first joined table generation unit.

Secure joining information generation system, secure joining system, methods therefor, secure computing apparatus and program

A secret share value [q] of a quotient q of a/p is obtained through secure computation using a secret share value [a] and a modulus p and [a/d]=[(a+qp)/d]−[q]p/d, . . . , [a/d]=[(a+qp)/d]−[q]p/dare obtained and output through secure computation using secret share values [a] and [q], divisors d, . . . , d, and a modulus p. Here, [μ] is a secret share value of μ, a is a real number, n is an integer equal to or greater than 2, d, . . . , dare divisors of real numbers, p is a modulus of a positive integer, and q is a quotient of a positive integer.

A calculation process is efficiently performed to a text file in which one or more records are included, each of the records includes one or more cells having an arbitrary length, and each of the cells includes arbitrary pieces of characters. A parameter setting apparatus sets a maximum value Sand a minimum value sof a size of character strings for one record by using attribute information as an input, a maximum value Sof a total size of encode information, a maximum value Sof a total size of a calculation value obtained by performing specific calculation to the encode information, and a total size Sof reference information, obtains a function value of C/(S+S+S) as the number of records which is a process unit of encoding and calculation, and obtains a function value of f/I·r·Sas the number of parallels in the calculation process. Here, C is a cache memory size, M is a main memory size, and fis a function value of s·M/(s+S+max(S,S)).

Growing community of inventors

Musashino, Japan

Dai Ikarashi