Kernels
KERNELS202 math · 16 application components · 2 privacy-preserving applications
SPECIFICATIONS2611 answered · 15 with nothing yet
IMPLEMENTATIONS21across 4 libraries
LAST 30 DAYS
25 submissions · 11 kernels moved
OPEN CHALLENGES15prizes on 1 kernel
ALL 20L1 · Math 2L2 · Cryptographic Components emptyL3 · Advanced Cryptography emptyL4 · Application Components 16L5 · Privacy-Preserving Applications 2
L1 · Math2 kernels · 2 specifications · 7 implementations
KERNELSUMMARYSPECS / BENCHMARKABLEIMPLEMENTATIONSTAGSLAST 30 DAYS
Element-wise MultiplyThe element-wise product of two length-N vectors: c_i = a_i * b_i.15ARITHMETICVECTORREFERENCE+1+4Polynomial MultiplicationCompute the product of two polynomials. The coefficient domain, polynomial ring and representation are fixed by individual specifications, not by the kernel.12ARITHMETICPOLYNOMIALRING+2+1L4 · Application Components16 kernels · 22 specifications · 14 implementations
KERNELSUMMARYSPECS / BENCHMARKABLEIMPLEMENTATIONSTAGSLAST 30 DAYS
Encrypted AXPYr = a·x + y over vectors: one scale, one add. The BLAS level-1 classic, and the first kernel measured over encrypted data.13FHECKKSARITHMETIC+1+5Encrypted MaximumCompute the maximum of a collection of encrypted values without revealing the individual inputs. The plaintext type, size, scheme and exactness are fixed by individual specifications.21FHECOMPARISONBGV+1+4Encrypted Signsign(x) for every element of a secret vector — the comparison primitive FHE does not have.12FHECOMPARISONML+1+3Encrypted Logisticσ(x) = 1/(1+e⁻ˣ) for every element of a secret vector — the probability gate of machine learning, encrypted.12FHEMLACTIVATION+1+2Encrypted ReLUmax(0, x) for every element of a secret vector — the activation function of deep learning, encrypted.12FHEMLACTIVATION+1+2Encrypted LookupA[i] with both the table and the index secret — random access without revealing the address.21FHEDATA-ACCESSPIR+1+1Encrypted Matrix MultiplicationC = A·B over square matrices — the workhorse of linear algebra, encrypted.21FHELINEAR-ALGEBRAML+1+1Encrypted Parityx mod 2 for every element of a secret vector — one exact bit out of an approximate scheme.21FHEBITWISECHALLENGE+1Encrypted Shift Leftx << n with the value and the count both secret — an EVM opcode, evaluated blind.21FHEBLOCKCHAINEVM+1+1Encrypted GELUEvaluate the GELU activation on an encrypted real vector under CKKS.1—CKKSno activityEncrypted Matrix InverseCompute the inverse of an encrypted square matrix under CKKS.1—CKKSno activityEncrypted ModuloCompute the elementwise modulo of two encrypted integer vectors under CKKS.2—CKKSno activityEncrypted SVDCompute the singular values of an encrypted matrix under CKKS.1—CKKSno activityEncrypted SoftmaxMap an encrypted real vector to a probability distribution over its slots under CKKS.1—CKKSno activityEncrypted SortReturn an encrypted real array in non-decreasing order under CKKS.1—CKKSno activityEncrypted k-Nearest NeighboursReturn the k nearest neighbours of an encrypted query over an encrypted dataset under CKKS.1—CKKSno activityL5 · Privacy-Preserving Applications2 kernels · 2 specifications · 0 implementations
KERNELSUMMARYSPECS / BENCHMARKABLEIMPLEMENTATIONSTAGSLAST 30 DAYS
Encrypted Image ClassificationAssign an encrypted image to one of a fixed set of classes under CKKS.1—CKKSno activityEncrypted Sentiment ClassificationClassify the sentiment of an encrypted feature vector under CKKS.1—CKKSno activity20 kernels shown · L2 and L3 have none yet