20 results about "Bootstrapping" patented technology

A processor-based method comprising: bootstrapping a first ciphertext that is a fully- homomorphic encryption of a plaintext, the plaintext being representable as a vector of coefficients of a polynomial, the bootstrapping comprising: obtaining, from the first ciphertext, an encryption of data indicative of one or more coefficients of the plaintext; based on a bootstrapping key, performing Torus-fully-homomorphic-encryption (TFHE) bootstrappings on the obtained encryption of data, the bootstrappings resulting in one or more bootstrap outputs; deriving, from the bootstrap outputs, a series of one or more encrypted values, each value of the series being based on a coefficient of a respective term of a polynomial; and creating, from, at least, a key-switching-key and the derived series of encrypted values, data indicative of a ring-based or module-based ciphertext, resulting in a second ciphertext that is an encryption, with reduced noise, of the data indicative of one or more coefficients of the plaintext.

Processes, methods, architectures, apparatuses, systems, apparatuses, and computer program products for protocol data units (PDUs) aware booting using multiple accesses or links. For example, uplink and / or downlink traffic may be transmitted using a PDU set aware steering mode. The PDU set aware boot mode may refer to, include, and / or otherwise use one or more boot algorithms applied to (e.g., boot) PDUs of the PDU set. The PDU set aware steering mode may be used to determine one or more accesses from a plurality of different accesses to be used for transmitting PDUs of a PDU set in uplink and / or downlink. A set of PDUs may carry information of application data units (ADUs) processed together by an application. Steering a set of PDUs between multiple accesses may improve quality of experience associated with use of an application, such as in augmented reality (XR) traffic flows.

The present application describes a multi-key fully homomorphic encryption (MK-FHE) scheme that enables secure and efficient multi-party computation by integrating learning with errors (LWE), ring learning with errors (RLWE), and NTRU-based encryption primitives. The scheme supports dynamic key management, parallelizable bootstrapping, and low-overhead homomorphic operations. Key innovations include a hybrid product mechanism for merging ciphertexts across cryptographic structures, a single-key blind rotation algorithm optimized for Fourier domain operations, and a noise-refreshing procedure that bounds error growth during homomorphic evaluations. This scheme achieves quasi-linear time complexity relative to the number of participating parties, making it suitable for resource-constrained environments such as federated learning and secure cloud-based AI inference.

An example system includes a one-time-use secret (OTUS) deployer engine to: provide an OTUS within a container-orchestrated environment (COE). The example system further includes: a non-OTUS provider engine to: provide a non-OTUS in exchange for the OTUS; and, invalidate the OTUS when the non-OTUS is provided. The example system further includes: a first container engine to: in response to bootstrapping, receive the OTUS from the OTUS deployer engine; and receive the non-OTUS from the non-OTUS provider engine in exchange for the OTUS. The example system further includes: a replica of the first container engine, to: in response to bootstrapping, after the first container engine, receive the OTUS from the OTUS deployer engine; attempt to receive the non-OTUS from the non-OTUS provider engine in exchange for the OTUS; receive an indication from the non-OTUS provider engine that the OTUS is invalid; and receive the non-OTUS from the first container engine.

This application discloses a fully homomorphic encryption / decryption method and computing device for performing fully programmable bootstrapping on ciphertext, supporting fully linear operations and expanding its applicability. The fully homomorphic encryption / decryption method includes: selecting noise values ​​from a Gaussian distribution; encoding the message into encoded data using the plaintext modulus and the noise values; encrypting the encoded data into LWE ciphertext using a random vector and a private key; sending the LWE ciphertext, bootstrapping key, and data processing request to a second computing device; performing homomorphic operations on the LWE ciphertext according to the data processing request by the second computing device; and receiving the homomorphic operation result from the second computing device and decrypting the homomorphic operation result using the private key.

A multi-system-on-chip (mSoC) can comprise a substrate with multiple chiplet positions, a main chiplet located on the substrate, and one or more secondary chiplets also located on the substrate. The main chiplet performs a self-identification process before the mSoC boots. The main chiplet then powers up and controls the boot process for the one or more secondary chiplets to identify the chiplet configuration on the mSoC. The main chiplet and each of the one or more secondary chiplets include multiple interfaces with the substrate and a communication mechanism for executing the boot process.

The invention discloses a fully homomorphic encryption high-precision bootstrap method which can give consideration to both efficiency and precision. The method sequentially comprises the steps of modulus improvement, coefficient slot transformation, polynomial approximation, inverse transformation and scaling. According to the method, firstly, the modulus is improved, the modulus space is expanded, and the slot vector is converted, so that large-scale parallel processing is realized; secondly, adopting a double-stage approximation framework, namely, firstly performing global preliminary approximation by using Fourier series, and then performing local refined optimization by using Chebyshev interpolation in a key effective approximation interval so as to effectively balance the calculation efficiency and the approximation precision; and finally, an error-aware adaptive node allocation algorithm is introduced to break through the traditional mode of uniform resource allocation, and remarkably higher overall approximation precision is obtained through dynamic optimization with the completely same calculation cost.

Systems described herein may use machine classifiers to perform a variety of natural language understanding tasks including, but not limited to multi-turn dialogue generation. Machine classifiers in accordance with aspects of the disclosure may model multi-turn dialogue as a one-to-many prediction task. The machine classifier may be trained using adversarial bootstrapping between a generator and a discriminator with multi-turn capabilities. The machine classifiers may be trained in both auto-regressive and traditional teacher-forcing modes, with the maximum likelihood loss of the auto-regressive outputs being weighted by the score from a metric-based discriminator model. The discriminators input may include a mixture of ground truth labels, the teacher-forcing outputs of the generator, and / or negative examples from the dataset. This mixture of input may allow for richer feedback on the autoregressive outputs of the generator. Additionally, dual sampling may improve response relevance and coherence by overcoming the problem of exposure bias.

The present invention relates to a method for performing an equality checking of a first character and a second character in a fully homomorphic encryption scheme wherein programmable bootstrapping is used, at least one of said first and said second character being encrypted, said first and second character each represented as a plurality of chunks. The method comprises: o checking chunk-by-chunk the equality of the first character with the second character by determining a difference between corresponding chunks, and o using the difference as input for a lookup table operation performed through the programmable bootstrap operation, said lookup table returning a first predetermined value if the difference equals zero and returning a second predetermined value different from the first predetermined value otherwise, wherein a lookup table result for a second or further chunk is determined based on the difference between corresponding current chunks and on a lookup table result for the preceding chunk, and wherein a lookup table result for the last chunk yields an equality result for the pair of characters formed by said first and second character.

Some embodiments are directed to a computer-implemented method (1200) of performing a computation on LWE-encrypted values. The invention relates to the TFHE setting in which the encryption supports a programmable bootstrapping operation that evaluates an LWE decryption in an exponent of a GLWE-encrypted monomial. In this setting, a LWE-encrypted product value is computed based on one or more LWE-encrypted input values by applying a cryptographic key switching operation to convert the LWE-encrypted input values to GLWE-encrypted polynomials, wherein a coefficient of an polynomial is a linear combination of the set of input values; applying an encrypted polynomial multiplication in the GLWE domain, and applying a cryptographic sample extraction operation to the resulting GLWE-encrypted polynomial to obtain the LWE-encrypted product value as an encryption of one of its coefficients.

The disclosed method for training a language model involves generating clustered data based on corpus data, performing a variety of bootstrapping operations based on the clustered data to train an initial predictor model and generate a mixed configuration, generating mixed data based on the mixed configuration and the clustered data, and performing one or more training operations based on the mixed data to train an initial untrained language model, and finally generating an initial trained language model, wherein the initial trained language model processes initial input data to generate initial output data.

The disclosed method of training a language model includes generating, based on corpus data, clustered data, performing, based on the clustered data, a plurality of bootstrapping operations to train a first predictor model and generate a mixture configuration, generating, based on the mixture configuration and the clustered data, mixture data, and performing, based on the mixture data, one or more training operations to train a first untrained language model to generate a first trained language model, where the first trained language model processes first input data to generate first output data.

A cloud computing service (e.g., a "Puffin service") is described. The service may maintain service and skill directories corresponding to various services to be deployed to the area (e.g., during area build). The service may hosting multiple user interfaces through which various service and skill metadata may be provided. In some embodiments, such data may include one or more dependencies between skills. Data managed by the cloud computing service may be used to construct a dependency graph. Navigation of the dependency graph may be performed via one or more user interfaces hosted by the cloud computing service. An orchestration service (e.g., a multi-group orchestrator) may manage the bootstrap work of any suitable number of services during area build based at least in part on dependencies between skills.

One example method includes, at a node, installing a default parametrization configuration that facilitates performance of a domain task, obtaining, by the node, a distilled dataset, and obtaining the distilled dataset is either: obtaining the distilled dataset from another node; or leveraging a synthetic state assembled in the node to select the distilled dataset from another node based on state similarity of the node to the another node. The example method further includes training a model at the node, and the training is performed using the distilled dataset, and the trained model is operable to leverage information received by the node to propose changes to the parametrization configuration so as to optimize execution of a task by the node.

A multiple-systems-on-chip (mSoC) can include a substrate comprising a plurality of chiplet positions. A main chiplet disposed on the substrate, and one or more secondary chiplets disposed on the subs

The disclosed systems and methods may launch a first node of a multi-node cluster from an attached storage medium, such as a USB driver, into a launcher OS. The launcher OS may initiate PXE and file sharing services on the first node. The disclosed features may copy one or more software packages for a target OS from the attached storage medium to a persistent storage device of the first node. The method may also include a PXE booting one or more other nodes of the multi-node cluster into the bootstrap OS and connecting one or more of the other nodes to the first node via a high speed switch. One or more software packages including software for the target OS are copied from the persistent storage device to at least one of the one or more other nodes via the high speed switch.

Methods and systems for performing efficient privacy-preserving machine learning (or other computationally intensive private computing operations) based on homomorphic encryption are disclosed. Two computer systems can collectively perform these methods on behalf of a client computer using threshold leveled homomorphic encryption (LHE) or fully homomorphic encryption (FHE), e.g., using the Cheon-Kim-Kim-Song cryptosystem. For example, two computer systems could privately evaluate a client's encrypted inputs using a machine learning model and return an encrypted output to the client, preserving the privacy of the client's data. Methods according to embodiments include efficient bootstrapping methods, ciphertext packing methods, encryption parameter optimization methods, and methods for privately fine-tuning machine learning models using low rank adaptation (LoRA). Embodiments improve the speed and efficiency of private HE-based computation, which are often several orders of magnitude slower than non-private computation.

A system for fully homomorphic encryption (FHE) inference includes: a plurality of server modules; an interconnect operatively coupled to the modules and configured to distribute session-initialization data for a large-language-model session; a first plurality of application-specific integrated circuits (ASICs) on one or more of the modules, each configured to perform bootstrapping on ciphertexts during an initiation phase; and a second plurality of ASICs on the modules comprising: bootstrapping circuitry configured to store auxiliary data used for bootstrapping and to perform the bootstrapping operations on the ciphertexts; processing circuitry configured to perform homomorphic arithmetic operations other than bootstrapping; and support circuitry configured to perform functions that support the bootstrapping and processing circuitry.

A multi-party system comprising a garbler and an evaluator for interactively executing homomorphic SIMD operations using garbled circuits. The garbler and evaluator may each store a unique share of a shared secret key, a ciphertext, and a shared public key. The garbler and evaluator may each partially decrypt the ciphertext using its key share to generate a unique data share. The garbler may linearize and reduce the size of the unique garbler data share. The garbler may send to the evaluator a garbled circuit, a garbling of the linear unique garbled data share, and garbled potential wires for the evaluator to garble its linear unique evaluator data share by oblivious transfer. The evaluator may evaluate the garbled circuit to execute a SIMD program to combine, in parallel, multiple indices of the linear garbler and evaluator unique data shares to efficiently generate an encrypted result of the garbled circuit.

This invention discloses a metadata-driven method and system for constructing a large language model intelligent agent through bootstrapping, relating to the field of artificial intelligence technology. The method includes the following steps: collecting natural language requirements generated from business scenarios and organizing these requirements into a unified expression format; generating readable metadata description content within this unified expression format; performing field splitting and role attribution around the metadata description content; constructing data type definition metadata within the unified expression format; and embedding prompt words into the data type definition metadata to generate content. This invention achieves progressive generation from natural language to data types, rules, and intelligent agent definitions through unified metadata expression, supporting dynamic construction and automatic loading at runtime, reducing manual intervention, and improving automation. Simultaneously, new metadata is generated during runtime and fed back into the generation chain, forming a continuous bootstrapping closed loop, enhancing structural consistency and version evolution controllability.