Apparatus and method for recommending and configuring LLM models for organizations. For example, LLM model usage requirements of one or more organizations are evaluated, including applications and users associated with each organization. A cost estimation is performed with respect to expected utilization of the plurality of LLM models and a subset of LLM models is recommended for each of the organizations, applications, and users, along with rate limits for each organization and corresponding applications based on a global threshold rate limit specified for the entity. Upon acceptance by an administrator, the global threshold rate limit is partitioned into a corresponding set of per-organization threshold rate limits; each organization threshold rate limit is allocated to a corresponding organization of the one or more organizations, and each respective threshold rate limit is subdivided into portions to be allocated to applications of the respective organization.
A hub-and-spoke architecture for bidirectional connections between a first vendor and a second vendor, wherein the second vendor functions as a partner on a spoke to the first vendor functioning as a hub, an event-based architecture configured to transmit data between the first vendor and the second vendor based on a predefined event, a metadata sharing module configured to share authorized organizational metadata between the vendor and the one or more partners, wherein the metadata sharing module enables the first vendor to submit a request for access to one or more requested data fields from the second vendor, enables the second vendor to approve the request, to identify one or more corresponding data fields corresponding to the one or more requested data fields, and to integrate the granted data fields into the vendor's CRM system.
G06F 21/51 - Monitoring users, programs or devices to maintain the integrity of platforms, e.g. of processors, firmware or operating systems at application loading time, e.g. accepting, rejecting, starting or inhibiting executable software based on integrity or source reliability
Techniques for generating a customized artificial intelligence (AI) assistant that can be integrated into existing systems or services are discussed herein. In some examples, an AI assistant management system may receive access to source code data (e.g., a source code repository) associated with a client application or system. The AI assistant management system may generate a training data based in part on analyzing the source code data. The AI assistant management system may train and/or fine-tune a large language model (LLM) based on the training data such that the customized LLM is equipped with the capability to respond to user queries regarding the particular client application. The customed LLM may be associated with a generated AI assistant component. The AI assistant management system may send the AI assistant component to the client for integration into the client application.
Techniques are described herein for a method of training a target activation sparsity in a neural network. The method includes obtaining a nonlinear portion of a plurality of neurons in a neural network. The neural network is trained to perform a target task. The method further includes substituting the nonlinear portion for a dynamic nonlinear portion in the plurality of neurons in the neural network. The dynamic nonlinear portion is trained to activate or deactivate one or more neurons of the plurality of neurons. The method further includes retraining the neural network using a first loss function that minimizes a loss of the target task and a second loss function that minimizes a number of active neurons.
Techniques (e.g., software, hardware, and/or machine-learned model(s)) may generate a third-party agent broker for validating data access requests on behalf of a resource provider. This third-party agent broker may validate the data access requests against a personalized access model associated with the resource provider. The context of the data access requests may be checked for similarity to the personalized access model (e.g., via embedding the data access request and the personalized access model). The data access requests may then be fulfilled based on validating the context of the data access requests.
A method for performing a dialog session including storing, in a memory a plurality of skills algorithms and a plurality of connected services in a memory, receiving, at a network interface, a first data indicative of a vehicle parameter, and a second data indicative of a vehicle location, and performing, by a processor, a digital twin algorithm to predict an operating condition of the vehicle in response to the first data, performing one of the plurality of skills in response to the operating condition and the second data to identify one of the connected services, and generating a control signal to initiate the connected service on a remote device.
Embodiments described herein provide a method of performing a vision-language task by a neural network multimodal model in response to multiple input images, the method comprising: receiving, via a data interface, a text input and an image input; generating text tokens based on the text input; generating a plurality of image patches, wherein each image patch of the plurality of image patches includes a portion of the image input at substantially a same resolution as the image input; generating a downsized image based on a downsizing of the image input; generating vision tokens based on the plurality of image patches and the downsized image; generating, via a neural network based language model, an output based on the text tokens and the vision tokens; and updating parameters of the neural network based language model based on a loss objective based on the output.
A method and system for providing containerized applications with updated secret values has been developed. An update to a secret from a first secret value to a second secret value is detected at a secrets vault. A configuration map associated with the secret is identified. The configuration map includes a first non-secret that is associated with the first secret value. A second non-secret that is associated with the second secret value is generated. The first non-secret is replaced with the second non-secret in the configuration map. The replacement of the first non-secret with the second non-secret in the configuration map results is an event. A reloader issues a restart signal to a containerized application associated with the secret in response to the event. The secret at the containerized application is updated from the first secret value to the second secret value during a restart process.
Database systems and methods are provided for supporting offline operation of a process flow associated with a native application at a client device. In response to an indication to pause a flow while in an offline mode, the method captures current values for input information for one or more fields for a form associated with the flow received at a client device via one or more graphical user interface (GUI) elements of a GUI display associated with the flow, encodes the current values in a serialized format, and updates a field of an object in a data storage at the client device to include the serialized encoded values. In response to a subsequent indication to resume the flow, the method regenerates the second GUI display associated with the flow using the serialized encoded values from the field of the object to populate the one or more GUI elements.
H04L 67/5683 - Storage of data provided by user terminals, i.e. reverse caching
G06F 16/25 - Integrating or interfacing systems involving database management systems
G06F 16/27 - Replication, distribution or synchronisation of data between databases or within a distributed database systemDistributed database system architectures therefor
11.
Systems and Methods for Cloud-Based Database System Account Management Operations Tracking
A database system may include an account repository, a communication interface, a data model, an operation execution engine, and a record repository. The account repository may store account information for database system accounts including a first subset corresponding with users and a second subset corresponding with machine models. The communication interface may receive a message identifying an operation of the plurality of computing operations and a machine model of the plurality of machine models to perform the operation. The data model may include operation restriction information identifying one or more restrictions regarding database system accounts authorized to perform the operation. The operation execution engine may execute the machine model to perform the operation upon determining that the first account is authorized to assign the operation to a second account of the second subset of accounts corresponding to the machine model.
Embodiments described herein provide for utilizing a large language model (LLM) to automatically generate unit tests, comprising image descriptions and expected answers for specified queries for use in visual programming. Further, text-to-image generation models are utilized to create images that align with the descriptions provided in each unit test. In some embodiments, a system executes only the top-scoring programs, reverts to a baseline model in cases of low scores, uses unit tests for re-prompting, and/or applies unit tests in reinforcement learning scenarios.
Embodiments described herein provide a generation model comprising a video-specific variational auto-encoder (VAE) for effective compression of video pixel information with reduced spatial and temporal dimensions and a video diffusion transformer (vDiT) to generate latent representations of frames. Specifically, the VAE may, instead of encoding each frame independently, incorporate both temporal and spatial compression. This significantly decreases the token length, improves the computational cost of training and inference, and facilitates the generation of long videos. The encoded training video, in the form of latent representations from a VAE encoder may then be passed to the vDiT to reconstruct the latent representations during training. The trained vDiT may then generate latent representations of a video in response to a text input, and the latent representations may be converted to a video output by a VAE decoder.
Embodiments described herein provide a multi-stage rating and re-ranking pipeline for selecting SWE agents for an input issue description. Specifically, a meta-policy may be selected among available agent policies corresponding to a pool of available SWE agents which maximizes the cumulative reward along the trajectory of states (such as status of a file) and actions taken at a series of time steps, and a context of relevant repository information and issue descriptions. By dynamically choosing the most suitable agent policy for each context, the selection pipeline maximizes the expected cumulative reward across all possible contexts. In this way, software issue resolve rate is improved.
In some systems, users may fine-tune a user-specific machine learning (ML) model. For example, a system may receive a first set of data from a first user and a second set of data from a second user that has a different size than the first set of data. The system may then input the first and second set of data into a probabilistic mixture model to obtain a set of global training parameters that includes a cluster proportions parameter, a cluster means parameter, and a cluster covariance parameter. Further, the system may generate an updated global training parameter for training an ML model for the first user and an updated global training parameter for training an ML model associated with the second user. Moreover, a quantity of updated global training parameters generated for a user may be based on the size of a set of data associated with the user.
A method for building an artificial intelligence (AI) agent. The method includes: receiving a training query; generating, by a first neural network based language model, a training dataset comprising a correct solution and an incorrect solution to the training query; generating, by a second neural network based language model, a first candidate score in response to the correct solution and a second candidate score in response to the incorrect solution; and training the second neural network based language model, based on a training objective. The method also includes building, at a server, an AI agent through a first application programming interface (API) to a third neural network based language model configured to generate a plurality of candidate solutions in response to the user utterance, and through a second API to the trained second neural network based language model configured to generate scores conditioned on the plurality of candidate solutions; ranking.
Database systems and methods are provided for managing usage of large language models (LLMs). One method involves determining a numerical representation of a conversational input to a user interface, identifying a semantically similar subset of prior conversational inputs based at least in part on the numerical representation of the conversational input, and determining numerical representations of respective conversational responses generated by a language model responsive to the respective prior conversational input of the semantically similar subset. When the numerical representations of the respective conversational responses satisfy a semantic similarity threshold, the method automatically generates an automated response to the conversational input based at least in part on one or more prior conversational responses and automatically provides the automated response to the user interface responsive to the conversational input.
Techniques for streaming generative machine learned model (or LLM) output to a virtual space are described herein. A system may receive a request to perform an action. The system may leverage LLMs to assist in performing aspects of the requested action. The system can generate input data to input into the LLM. When generating the input data, the system can identify sensitive data associated with the request. The system can modify the input data to mask and/or anonymize the sensitive data. The system can input the input data into an LLM trained to output a subset (e.g., less than all) of the response at a time. The system can add the output subset to a buffer and upon identifying the masked data in the buffer, the system can demask the sensitive data and output the sensitive data to the user profile in a streaming manner.
A database storing metadata describing a plurality of features may be maintained in association with a computing system implemented via a database system. The metadata may be processable to cause the features to be rendered in a plurality of different user experiences. A user interface may be displayed on a device of an authorized administrator affiliated with an organization implementing the computing system. The user interface May be associated with a first one of the different user experiences, The user interface may be configurable to allow the authorized administrator to enable a plurality of sets of features on behalf of the organization. A request a request to enable a first set of features from the plurality of sets of features may be received. The first set of features may be caused to be enabled for users associated with the organization.
Techniques for filtering out undesirable generative machine learned model (or LLM) output are discussed herein. A system may receive a subset of an LLM output. That is, the system may stream the LLM output to a user device by receiving one or more tokens from the LLM and outputting such token(s) to a user device. However, prior to outputting the token(s) to the user device, the system may determine whether the token(s) include undesirable content that is to be blocked. The system may use synchronous blocking components (e.g., blocks the undesirable token(s) before such token(s) get output to the user device) and/or asynchronous blocking components (e.g., blocks the undesirable token(s) after the token(s) have been output to the user device) to filter out undesirable content. The synchronous and/or asynchronous blocking components may be designed to block one or more undesirable topics such as hateful speech, profanity, bias, toxicity, factualness, etc.
Apparatus and method for integrating external AI services. For example, one embodiment of a method comprises: preparing training data received from various data streams on the cloud-based application service, wherein preparing includes categorizing, filtering, and curating data from the data streams; generating source data model objects (DMOs) based on training data; providing the source DMOs to the external AI service over a secure communication channel, the external AI service to register an AI model based on the source DMOs and to generate a corresponding AI model endpoint; executing an AI model builder on the cloud-based application service, the AI model builder to generate an AI model reference configurable with connection information to communicate with the AI model endpoint, the AI model builder configurable to automatically trigger an inference when data mapped to an input of the AI model is changed in one or more of the source DMOs.
Disclosed are some implementations of systems, apparatus, methods and computer program products for implementing page context aware actions. Input submitted via an input device is obtained. A context is determined based upon a web age rendered via a display device, where the context includes one or more field placeholders. The input is transmitted to a large language model (LLM). The determined context is transmitted to the LLM, wherein values of the one or more field placeholders are not transmitted to the LLM. A plan including at least one of the field placeholders is received from the LLM, where the plan includes one or more actions, at least one action being with respect to one of the field placeholders. One or more field values are substituted into the field placeholders of the plan. The plan is executed by executing the one or more actions of the plan, at least one action of the one or more actions being with respect to a field value of the one or more field values.
Techniques are disclosed for managing metadata of a distributed database system in a hybrid manner. A computer system may receive, from a computing device, a request to access a set of data stored in nodes of a distributed storage system that is a caching layer of the system. The system retrieves metadata for a set of data specified in the request, including accessing a reversemap storing a reverse-ordered copy of original metadata stored in a metadata store of the system, where the reversemap is stored on durable storage of the system. Based on retrieving the metadata from the reversemap, the system accesses nodes of the distributed storage system, where the reversemap specifies the nodes of the distributed storage system that store the set of data. The system transmits, to the computing device, information indicating a result of accessing data stored in nodes of the distributed storage system.
G06F 11/20 - Error detection or correction of the data by redundancy in hardware using active fault-masking, e.g. by switching out faulty elements or by switching in spare elements
A computing device executing a browser application obtains a dataset for rendering a data visualization, the dataset including a plurality of data points. The device selects, from the plurality of data points, a first subset of data points according to a statistical data distribution of the dataset. The device recursively applies a first algorithm to the first subset of data points to obtain a final subset of data points. Each of first subset of data points and the final subset of data points has a fewer number of data points than the plurality of data points. The device renders a data visualization using the browser application. The data visualization has a plurality of data marks corresponding to the final subset of data points. The device displays, on the browser application, the data visualization including the plurality of data marks.
A computer system obtains an initial dataset for rendering a data visualization. The initial dataset includes a plurality of data points, and each data point has a respective spatial location in the visualization. The system dynamically generates a data structure, including (i) recursively dividing the plurality of data points into a plurality of nodes until each node satisfies a set of criteria and (ii) allocating a respective subset of data points to each node according to a spatial location of a respective data point in the data visualization. For each node, the system recursively applies a linearization algorithm to an initial subset of data points to obtain a reduced subset of data points. The system obtains a reduced dataset, generates a data visualization according to data in the reduced dataset, and causes display of the visualization on a browser application.
Systems, devices, and techniques are disclosed for efficient knowledge graph indexing and retrieval. Document chunks may be generated from documents. Summarizations may be generated from document chunks. Entity types, entity properties, relations, and relation properties may be generated from a subset of the summarizations. A schema including entity types, entity properties, relations, and relation properties may be generated. Entity property triplets and entity relation triplets may be generated from the summarizations based on the schema and linked to the document chunks. A knowledge graph including nodes representing entities from the entity property triplets and entity relation triplets and edges representing the entity property triplets and the entity relation triplets may be generated. A search query may be received. Nodes and edges of the knowledge graph that include the entities, the entity property triplets and the entity relation triplets most similar to keywords of the search query may be determined.
Disclosed are some implementations of systems, apparatus, methods and computer program products for providing recommendations in a recommendation system. A server system applies a large language model (LLM) to identify a first one of a plurality of items based, at least in part, on a first user profile. The system recommends the first item and a machine learning model is generated or updated based, at least in part, on training data including the first item and the first user profile. The system then applies the machine learning model to identify a second one of the plurality of items. The system determines whether the trained machine learning model has predicted the second item with a confidence that is greater than a predetermined threshold. The system then returns the second item according to whether the trained machine learning model has predicted the identified second item with a confidence that is greater than the predetermined threshold.
Embodiments described herein provide knowledge graph synthesis pipeline to generate a knowledge graph from long documents so as to serve a retrieval augmented generation (RAG) large language model (LLM) based AI chat agent. Specifically, each document is decontextualized by substituting entity references with their explicit mentions. Subsequently, to enhance coverage, the document is segmented into chunks and entities and relations are extracted from each chunk independently, e.g., by an LLM. The extracted entities and relations are then synthesized into a knowledge graph for the document. Therefore, the retrieval component may search the knowledge graph based on a received user query to retrieve entities and relations, which are in turn input to an LLM to generate a response.
According to some implementations, while a proxy routes production traffic to a first application (app) version that runs in a plurality of container orchestration system (cos) pods having first app version containers, configuration information is received including an identification of a second app version container image for a second app version. The second app version is an updated version of the first app version. Cos pods having second app version containers are brought up based on the second app version container image identified in the configuration information. Test and/or warmup traffic is caused to be routed to the second app version containers. Responsive to an indication regarding the routing of the test and/or warmup traffic to the second app version, causing a transition to sending production traffic to the second app version containers instead of to the first app version.
Methods and corresponding systems and apparatuses for integrating data from disparate data sources are described. A list of accounts that are accessible to a user of a customer relationship management (CRM) database may be determined. At least one segment may be determined. The at least one segment may represent a group of individuals that satisfy membership criteria associated with the at least one segment. One or more individuals that are (i) associated with a given account in the list of accounts and (ii) included in the at least one segment may be determined based at least in part on one or more queries involving data tables that reside in a data cloud platform separate from the CRM database. Information describing at least the one or more individuals may be provided for presentation in a graphical user interface (GUI).
A computer system determines whether a selected data field matches a first bin concept of a plurality of bin concepts in a semantic bin lookup table. Each bin concept is associated with one or more bin options and each bin option includes a plurality of bin intervals. When the data field matches the first respective bin concept, the computer system selects, from the semantic bin lookup table, a bin option of one or more bin options associated with the bin concept. The selected bin option includes a first plurality of associated bin intervals. The computer system displays a data distribution for the selected data field according to the first plurality of associated bin intervals. When the data field does not match a bin concept, the computer system generates default bin intervals based on statistical properties of the distribution of data values for the data field.
In some embodiments, a method initiates a set of threads at an engine to perform a set of migration tasks for a migration of data for a plurality of organizations. A refill task is started to monitor a status of migration tasks in the set of migration tasks at a first interval cycle. The refill task is running outside of a context of the engine. At a time in the first interval cycle, the method determines a status of migration tasks in the set of migration tasks. A new migration task is sent to the engine for assignment to a thread that has finished its respective migration task before one of the threads has finished executing a migration task in the set of migrations tasks.
Fine-tuning AI models is described. According to some aspects, a set of one or more data objects are selected. Based on the selection, a set of one or more of a plurality of categories is selected. Also, one of a number of pre-trained AI models is selected based on the set of categories and implicit input. In addition, one of a number of fine-tuning methods is selected. The selected set of categories identify a selected subset of categorized data items in the selected set of data objects. The selected AI model is fine-tuned using the selected fine-tuning method and a version of the selected subset of categorized data items.
Techniques for masking data based on annotations are discussed herein. A system may receive a request to perform an action and leverage an LLM to assist in performing the requested action. When generating the input data to input to the LLM, the system can use a template to organize the input data. The template may include static data and/or slot(s) which can include a reference to data to input into such slots. The system may retrieve data to input to the slot based on the reference, retrieve annotations that define a classification of the data, and receive a policy that defines which types of data classifications are to be masked. Based on the data classification and the policy, the system can determine whether to mask the data. The system can generate the input data using the template, the data, and/or the mask(s) and input such data into the LLM.
A computer implemented method for managing and remediating security drift in a public cloud network is disclosed. A security drift event may be received at a contextual impact classification engine of a server. An impact tier for the received security drift event may be assigned at the contextual impact classification engine. A queue shaping orchestrator at the server may reorder a queue with entries that include the received security drift event based on the assigned impact tier. A remediation engine of the server may determine a remediation for the received security drift event based on the assigned impact tier, and/or one or more contextual inputs received by the server.
Embodiments described herein provide A method of item recommendation at an electronic recommendation system, the method comprising: receiving, via a data interface, a query; generating, via a base neural network based recommender model, a first recommendation of at least a first item based on the query, wherein the first item is associated with a vector of tags; computing a scalar value indicative of a relevance level between the first item and a specific domain based on a multiplication of the vector of tags with a learnable coefficient vector associated with the specific domain; and displaying the first item via a user interface in response to the scalar value surpassing a threshold.
Methods, systems, apparatuses, and computer program products are described. A data processing system may ingest asset service action history data from data sources of different data organization models via clicks from a user, generate at least an asset service action training data set from the ingested data, and train an asset management model to predict future asset service actions using the asset service action training data set. The different data organization models may include data objects including cases, work orders, asset identifiers, and products. The data processing system may transform the ingested asset history into the asset service action training data set according to a configurable unified data model. The future asset service action predictions may be based on a binary classification system, and may indicate a likelihood of performing a second asset service action type on an asset type based on an inputted first asset service action type.
G06Q 10/20 - Administration of product repair or maintenance
G06F 16/25 - Integrating or interfacing systems involving database management systems
G06Q 10/04 - Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
47.
System And Methods Of Defense Against DDoS Attacks Via Autonomous Agents For Applications On A Multi-Substrate Multi-Ingress Shared Infrastructure With Multiple Cloud Architectures
A computing services environment may include application gateways receiving application-layer request messages from various sources. The computing services environment may also include an autonomous agent platform configured to instantiate and execute an autonomous agent to evaluate network traffic associated with a portion of the computing services environment. The computing services environment may also include an orchestration engine configured to determine one or more mitigation policies corresponding with one or more of the application gateways based on identification of the application-layer distributed denial of service attack by the autonomous agent. The computing services environment may also include application-layer web application firewalls corresponding to the plurality of application gateways and implementing the one or more mitigation policies to prevent a subset of subsequent application-layer request messages from the subset of the sources from reaching one or more components of the computing services environment.
Techniques may generate realistic synthetic data by programmatically generating a configuration file object type and relationship data. This configuration file may be used to retrieve source data matching the object type(s) and/or specific records indicated by the configuration file. The techniques may detect and anonymize private/proprietary information and may determine statistical characteristic(s) of the source data. A batch of prompt(s) may be generated using the source data, the statistical characteristic(s), and the configuration file and may be transmitted to one or more instances of a transformer-based machine-learned model. Sets of synthetic data received from the model instance(s) may be de-duplicated, checked for similarity to the source data (e.g., via embedding the synthetic data and the source data), and may be used to generate synthetic object(s) using the relationship(s) and/or other data indicated by the configuration file. These synthetic object(s) may then be deployed in a software environment.
A computing services environment may include application gateways receiving application-layer request messages from a plurality of sources. The computing services environment may also include an orchestration engine configured to identify an application-layer distributed denial of service attack based on input data characterizing network traffic received at the application gateways and to determine a mitigation plan update to address the application-layer distributed denial of service attack. The computing services environment may also include an autonomous AI agent platform configured to instantiate and execute an autonomous AI agent instance configured to determine whether to approve or reject the mitigation plan update by evaluating the mitigation plan update via a generative language model. The computing services environment may also include application-layer web application firewalls corresponding to application gateways. The orchestration engine may instruct the application-layer web application firewalls to implement the mitigation plan update upon approval by the autonomous AI agent instance.
H04L 41/16 - Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks using machine learning or artificial intelligence
50.
System And Methods Of Defense Against DDoS Attacks For Applications On A Multi-Substrate Multi-Ingress Shared Infrastructure With Multiple Cloud Architectures
A computing services environment may provide computing services to a plurality of recipients via the Internet. The computing services environment may include application gateways receiving application-layer request messages from various sources. The computing services environment may also include an orchestration engine determining mitigation policies corresponding with the application gateways based on a classification of a subset of the application-layer request messages as being sent from sources associated with a distributed denial of service attack. The computing services environment may also include application-layer web application firewalls corresponding to the application gateways and being configured to transition from a deactivated state to an activated state upon receipt of an instruction from the orchestration engine. The activated application-layer web application firewalls may implement the mitigation policies prevent a subset of subsequent application-layer request messages from the subset of the sources from reaching one or more components of the computing services environment.
Apparatus and method for predictive protection of cloud-based applications and services. For example, a web application firewall (WAF) detects vulnerabilities in the data traffic and collects relevant information including, for example, the payload type, structure, and specific vulnerability information. For certain vulnerabilities associated with views, the view document object model (DOM) and history of views may be collected. For vulnerabilities related to API calls, the corresponding the API path and history of API calls may be retrieved. The WAF includes a signature controller which decodes the payload (e.g., the JavaScript Object Notation (JSON) structure) and creates a signature of the vulnerability based on the relevant information including, but not limited to, the API path, the web application domain, and/or the URL path. The WAF distributes collected and generated vulnerability information to web browser extensions of the web application which perform mitigations such as generating notifications when a vulnerability is encountered.
A computer services environment may include web servers providing access domains and a network ingress paths receiving application-layer request messages. The application-layer request messages may each be received from a respective source via a respective ingress path and may be directed to a domain. The computing services environment may also include an orchestration engine configured to determine and implement mitigation policies corresponding with the ingress paths based on a classification of a subset of the plurality of application-layer request messages as being sent from a subset of the sources associated with a distributed denial of service attack. The mitigation policies may include rules to prevent a subset of subsequent application-layer request messages from the subset of the sources from reaching one or more components of the computing services environment.
Systems and methods are provided for determining, at a server communicatively coupled to a database, whether any past conversation texts in the database include a message segment that used by one or more service agents in a conversation with a customer. The server samples m-furthest neighbors for at least one of the determined past conversation texts to determine usage data for the at least one of the determined past conversation texts. The samples are indexed for at least one of the determined past conversation texts for retrieval based on a context from the determined usage data to generate an index. A representation of a current conversation between the service agent and the customer is determined, and index is queried using the representation. A large language model (LLM) generates a response that is transmitted to the service agent to be used in the conversation with the customer.
Methods, systems, apparatuses, and computer program products are described. A system may obtain a set of metadata associated with multiple entities of a datastore, the multiple entities including at least two different data types. The system may augment, using a generative artificial intelligence (AI) model, the set of metadata resulting in a set of augmented metadata for the multiple entities of the datastore. The system may generate, using the set of augmented metadata, a set of entity objects for the multiple entities of the datastore and in accordance with a reference schema. The system may identify, using the generative AI model and the set of entity objects as inputs into the generative AI model, first relationships between two or more first entities of the multiple entities of the datastore. Based on the identified relationships and a reference schema relationship model, the system may generate a schema relationship model.
Techniques are disclosed relating to implementing a statement-level INSTEAD OF trigger. In one embodiment a computer system receives a database operation statement specifying performance of a particular database operation on one or more database views of selected data from a database, wherein a given database trigger of one or more statement-level database triggers is executable to initiate execution of at least one trigger instruction for the database instead of performing the particular database operation on a specified database view of the one or more database views. For a given database view of the one or more database views, the computer system stores a trigger function call in a function call data structure for each statement-level database trigger defined for the database operation statement and the given database view. The computer system executes the stored trigger function calls instead of executing the database operation statement.
Embodiments described herein provide a parallel adapter-based training paradigm that trains multiple adapters in parallel for specific tasks or domains. The trained adapters are then selectively merged with a base neural network to produce a new finetuned neural network that is finetuned to perform the specific tasks. In this way, the parallel training largely improves computational efficiency to train or adapt a neural network for different tasks without repeated retraining of the entire neural network.
An application server may receive, from a user and at an interface for accessing a large language model, a request for a response from the large language model. In some cases, the request may include a prompt for the large language model and data access role information associated with the user. The application server may retrieve, from a data source including a set of data objects, one or more data objects for inputting to the large language model based on comparing the data access role information associated with the user with data access policy information associated with the one or more data objects. The application server may then input, via a model interface, the one or more data objects to the large language model, and may receive, via the model interface, an output of the large language model based on the one or more data objects.
Techniques are described herein for a method of decreasing the likelihood of out-of-domain LLM responses. The method includes determining, by a block of a LLM, a representation of the text input. The method further includes determining a set of coefficients based at least on a reconstruction of the text input using a dictionary and the representation of the text input. The method further includes performing a sparsity check using the set of coefficients. The method further includes generating a response to the text input based at least on the sparsity check.
A method receives audio data from a call. Services are performed to process the audio data to automatically generate a response, wherein the services include converting the audio data to input text, inputting the input text into a model to automatically generate a text response, and converting the text response to an audio response. Supplemental words are selected based on the input text. The method determines a type of service based on services performed to generate the audio response and determines a position in the response to insert the supplemental words based on the type of service. The supplemental words are provided for insertion in the call at the position to supplement the audio response.
G10L 13/08 - Text analysis or generation of parameters for speech synthesis out of text, e.g. grapheme to phoneme translation, prosody generation or stress or intonation determination
G10L 15/18 - Speech classification or search using natural language modelling
G10L 15/22 - Procedures used during a speech recognition process, e.g. man-machine dialog
G10L 15/30 - Distributed recognition, e.g. in client-server systems, for mobile phones or network applications
60.
SYSTEMS AND METHODS FOR CONSTRUCTING NEURAL NETWORKS
Embodiments provide a merging framework that selectively merges pretrained model parameters of an LLM and retrained adapter weights. Specifically, the merging framework measures a similarity metric between a pretrained base LLM and an adapter that is retrained for a specific task or domain, and then prunes one or more components (weights or layers) of the adapter that have a high similarity with the base LLM and thus are likely to be redundant. The pruned adapter with only sparse features that are most dissimilar to the base LLM is then merged with the base LLM to produce a new neural network model that is adapted for the specific task or domain. In this way, redundant features may be pruned from adapter modules before merging.
Embodiments also provide an LLM adapter training and merging framework that builds a new neural network model by merging a first LLM (stronger) with an adapter that has been trained in conjunction with a second LLM (weaker). Specifically, the adapter may be trained in conjunction with a smaller LLM to perform a specific task or adapt to a particular domain. The trained adapter is then merged with a different (larger) LLM to produce a new model. In this way, developers may select compatible LLMs as base models to merge with trained adapters to produce new models without additional training and/or finetuning the adapter with different LLMs. The one-time domain specific adapter training may be applied to any subsequent developments in merging compatible models with the trained specific adapter, thus enhancing computational efficiency of neural network model adaptation.
A system may receive a configuration associated with a tenant of a multi-tenant generative artificial intelligence (AI) system and tenant-specific training data, where the configuration includes a first indication of a first communication channel over which a tenant-specific conversational agent is to communicate with users and where the tenant-specific training data includes context information associated with the tenant that is expressed in natural language. The system may determine an intent of a query received from the tenant based at least in part on an analysis of the query. The system may transmit the query to a first generative AI model of a plurality of generative AI models, wherein the first generative AI model is selected based at least in part on the determined intent. The system may transmit, to the tenant over the first communication channel, a response to the query generated by the first generative AI model.
Techniques are disclosed pertaining to modifying storage properties of application pods in a computing environment. A computer system may receive an update to a set of storage properties associated with a deployment of application pods coupled to storage volumes that satisfy the storage properties. The computer system performs a volume conversion process to replace the application pods with ones coupled to storage volumes that satisfy an updated set of storage properties corresponding to the update. The volume conversion process involves transitioning a particular application pod into a suspended state in which the pod is unavailable for data access and replicating data associated with the particular application pod to at least one other application pod. After replicating the data, the computer system deletes the particular pod to trigger a deployment system to provision a replacement application pod coupled to a storage volume satisfying the updated set of storage properties.
Embodiments described herein provide a method for configuring an artificial intelligence (AI) conversation bot to respond to a user query based on retrieved contextual documents. The method includes: receiving, via a communication interface, a user query comprising a natural language description of a topic; generating, by a first neural network based language model, one or more subtopics of the topic based on a first input prompt combining the topic and a first instruction to generate the one or more subtopics; generating, by the first neural network based language model, one or more statements for at least one of the subtopics based on a second input prompt combining the one or more subtopics and a second instruction to generate the one or more statements; and generating, by the first neural network based language model, at least one document containing a set of randomly selected statements from the one or more statements.
Embodiments described herein provide a method for time series forecast. The method includes: obtaining a set of time series data comprising a first segment of past time series data and a second segment of predicted time series data; generating, by a first neural network based language model, a text description describing a forecast explanation based on a first input prompt combining the set of time series data; generating, by a second neural network based language model, a third segment of predicted time series data based on a second input prompt combining the first segment of past time series data and the text description of forecast explanation; determining a performance metric based on a comparison between the second segment of predicted time series data and the third segment of predicted time series data; and generating a control command based on the text description to cause an action with a control system.
Embodiments described herein provide a method for building an artificial intelligence (AI) agent to respond to a user query. The method includes: receiving a user query; retrieving a set of documents that are ranked based on respective relevance scores of a first type to the user query; generating a core question that filters out irrelevant texts from the user query; generating a first summary of a first document from the set of documents and a first reasoning output explaining how the first summary addresses the core question; generating a relevance score of a second type and a corresponding reranking for the first document based at least in part on a combination of the core question and the first reasoning output; generating a response to the user query using one or more top-ranked documents according to generated rerankings of the set of documents.
Apparatus and method for managing cross-domain connectivity, interoperability, and authentication. For example, some implementations rely on an external credential manager which securely stores user credentials for each respective service domain for which integration is to be performed. An introspection service searches an existing codebase to locate connector code that can be used for the integration, which are presented as options to the user within an integration development application. The connector code includes identifiers which indicate the credentials to request from the credential management service. The credentials are never provided directly to the integration platform, which generates a mapping between the placeholders and unique identifiers which identify the corresponding credentials. The mapping and the associated connector code are provided to the integration, which uses the mapping to inject the credentials into the connector code in accordance with the placeholders. The credentials are used for authenticating with service domains.
Embodiments described herein provide A method of fine-tuning a neural network based model. In some embodiments, a system receives, via a data interface, a training dataset including a plurality of input samples. The system generates, via a pre-trained neural network based model, a first response based on a first input sample of the plurality of input samples, and a second response based on the first input sample. The system generates, via a trained reward model, a first reward score based on the first input sample and the first response, and a second reward score based on the first input sample and the second response. The system computes a loss function based on the first prompt, the first response, the second response, the first reward score, and the second reward score. The system updates parameters of the neural network based model based on the loss function.
Fine-tuning AI models is described. According to some aspects, one of a number of pre-trained AI models is selected based on the explicit input and the implicit input. In addition, one of a number of fine-tuning methods is selected. Also, a set of one or more of a plurality of categories is selected, where a categorized data set associated with an organization was classified into the categories using a classifier, and where the selected set of categories identify a selected subset of the categorized data set. A version of the selected subset is used to fine-tune the selected AI model using the selected fine-tuning method.
Disclosed herein are mobile device, method, and computer program product embodiments for an improved integrated mobile AI assistant. The mobile device may launch a mobile application including an integration component, where the integration component is in communication, through the mobile application, with a data service and a user interface (UI) service. The integration component may receive a response including data and a data type from the data service generated by a large language model responsive to a natural language query. The integration component may customize an interface at the integration component using a rendering configuration received from the UI service to display the data, the rendering configuration generated by decomposing the data type into a predefined type.
H04L 51/02 - User-to-user messaging in packet-switching networks, transmitted according to store-and-forward or real-time protocols, e.g. e-mail using automatic reactions or user delegation, e.g. automatic replies or chatbot-generated messages
71.
SYSTEMS AND METHODS FOR TRAINING AND EVALUATING MULTIMODAL NEURAL NETWORK BASED LANGUAGE MODELS
Embodiments described herein provide a method of building an artificial intelligence (AI) agent to respond to a task request from a user. The method includes: receiving a set of single-modal data samples of a plurality of modalities; selecting a first single-modal data sample of a first modality and a second single-modal data sample of a second modality; generating a question associated with the first single-modal data sample and the second single-modal data sample; generating an answer with a reasoning to the question based on a second input prompt; training, a second neural network based language model, using a dataset comprising the question and the answer to generate a candidate answer in response to a training query; building the AI conversation bot through an application programming interface to the trained second neural network language model; and generating, using the AI conversation bot, a response to the task request.
Disclosed are some implementations of systems, apparatus, methods and computer program products for synchronizing data. A source device processes an update to data in a database. The source device transmits, via a message bus, a first event message pertaining to the update, the first event message having an associated indicator. A target device accessing the message bus detects the indicator. Responsive to detecting the indicator, the target device skips the first event message on the message bus and identifies a snapshot link in a second event message subsequent to the first event message. The target device accesses a snapshot event identified by the snapshot link, stores data of the snapshot event, and processes one or more event messages subsequent to the snapshot event.
G06F 16/27 - Replication, distribution or synchronisation of data between databases or within a distributed database systemDistributed database system architectures therefor
G06F 11/14 - Error detection or correction of the data by redundancy in operation, e.g. by using different operation sequences leading to the same result
Methods, systems, apparatuses, devices, and computer program products are described. A processing device may support a large language model (LLM) for automatically improving pull requests to a codebase. To use the LLM, the processing device may create and maintain a vector space tracking information relating to historical pull requests to the codebase. The processing device may receive a new pull request indicating a change to code in the codebase and may determine, from the vector space, a vector corresponding to a code chunk affected by the pull request. The processing device may send, as an input to the LLM, a prompt including the code chunk affected by the pull request and one or more comments from a set of historical comments relating to the code chunk and indicated by the determined vector. The processing device may modify the pull request based on the one or more comments.
A policy-based approach to execution of commands in a distributed environment involves applying policies to determine permissions for executing commands. In some implementations, a user inputs a command at a web portal, causing a request to be sent to a computer system. The web portal also sends an indication of one or more machine components of a remote system to which the command is to be applied. After identifying a policy associated with the user, the computer system evaluates a rule in the policy to determine whether the user is permitted to execute the command with respect to the one or more machine components. The computer system routes the command to the remote system for execution based on determining that the rule is satisfied. This enables the command to be executed without providing the user with direct or unrestricted access to the remote system.
Techniques are disclosed relating to database query optimizers. In some embodiments, a system receives, from a query optimizer, a plurality of query plans for a database maintained by the database system. The system retrieves a set of database statistics for the database and generates, via a data synthesizer, a plurality of synthetic datasets, where generating a given synthetic dataset is performed based on a given query plan of the plurality of query plans and the set of database statistics, and includes generating a plurality of synthetic data tuples. The system executes the plurality of query plans on the plurality of synthetic datasets and updates the query optimizer based on results of executing the plurality of query plans on the plurality of synthetic datasets. The disclosed data synthesis may advantageously improve query performance due to more efficient query plans being selected for execution of requested queries.
Techniques for generating and displaying a summary of multiple virtual spaces are discussed herein. A communication platform may determine whether to generate a summary of the content posted across a set of virtual spaces. For instance, the communication platform can identify a set of virtual spaces that the user is a member of. For a pre-determined period, the communication platform can determine a first number of content items posted to the set of virtual spaces. The communication platform may also determine, over the same period, a second number indicating the number of the content items the user has yet to view. Based on the second number meeting or exceeding a threshold, the communication platform may generate a summary for the user. Accordingly, the communication platform may generate a summary of the content posted to the set of virtual spaces and display the summary via a user interface of the user.
G06F 3/01 - Input arrangements or combined input and output arrangements for interaction between user and computer
G06F 3/0481 - Interaction techniques based on graphical user interfaces [GUI] based on specific properties of the displayed interaction object or a metaphor-based environment, e.g. interaction with desktop elements like windows or icons, or assisted by a cursor's changing behaviour or appearance
78.
Applied Artificial Intelligence Technology for Narrative Generation Based on Explanation Communication Goals
Artificial intelligence (AI) technology can be used in combination with composable communication goal statements to facilitate a user's ability to quickly structure story outlines using “explanation” communication goals in a manner usable by an NLG narrative generation system without any need for the user to directly author computer code. This AI technology permits NLG systems to determine the appropriate content for inclusion in a narrative story about a data set in a manner that will satisfy a desired explanation communication goal such that the narratives will express various ideas that are deemed relevant to a given explanation communication goal.
A method for avoiding exposure of sensitive data in a web application running in a browser due to rendering content generated by a generative artificial intelligence platform (“GenAI content”). A Web Application Firewall or Gateway (“WAF”) receives, from a web application, a first response to forward to the browser. The WAF modifies the first response by inserting detection code that causes the browser to: obscure a rendered version of the GenAI content in a first browser window, render the GenAI content in a second browser window that is not visible, convert the rendered GenAI content in the second browser window to an image, obtain an assessment from a sensitive data scanning engine on whether the image contains sensitive data, and based on the assessment, determine whether to unobscure the rendered version of the GenAI content in the first browser window. The WAF sends the modified response to the browser.
A method of generating a code output in response to a natural language problem description. The method includes: receiving the natural language problem description; generating, by a neural network based language model, a first candidate code snippet based on a first input prompt combining the natural language problem description and a first instruction; executing, at a code execution environment, the first candidate code snippet based on a unit test thereby producing a first feedback reflecting a correctness of the first candidate code snippet; generating, by the neural network based language model, a second candidate code snippet based on a second input prompt combining the natural language problem description, the first candidate code snippet, and the first feedback; and executing, at the code execution environment, the second candidate code snippet based on a runtime test thereby producing a second feedback reflecting a runtime efficiency of the second candidate code snippet.
Embodiments described herein provide a video generation framework built on a decoupled multimodal cross-attention module to simultaneously condition the generation on both an input image and a text input. The video generation may thus be conditioned on the visual appearance of a target object reflected in the input image. In this way, zero-shot video generation may be achieved with little fine-tuning efforts.
System, method and interface for generating data visualizations are provided. The system receives a user input to specify a natural language command directed to a data source. The system also generates a prompt for generating a data visualization based on relevant data fields and data values, rules that characterize the data visualization, and a context free grammar. The system also prompts a trained large language model using the prompt to generate a structured document following a domain-specific schema based on a shorthand notation. The system also uses a parser that uses the context free grammar to map the structured document to a visual specification. The visual specification specifies the data source, visual variables, and data fields from the data source. The system also generates and displaying a data visualization based on the visual specification, including displaying visual marks representing data, retrieved from the data source, for the data fields.
Embodiments described herein provide a method of jointly generating a code output. A first language model (LM) generates a code output in response to a task description. Second and third LMs generate critiques based on the task description and the generated code. The second LM may critique the accuracy of the generated code, and the third LM may critique the safety of the generated code (e.g., susceptibility to hacks). The first LM may revise the generated code based on the critiques. The revised code may be executed, and based on the results of the execution, the first LM may revise the code again. The process of critiques, revisions, and execution may be repeated. The final generated code is output to a user (e.g., in a programming environment).
Embodiments described herein provide a reinforcement learning framework for neural network models to generate outputs that align with desired human preference. In at least one embodiment, cross-prompts are generated from an original prompt to elicit a response from the neural network model.
Embodiments described herein provide A method of training a neural network based model for predicting time series data. The method may include receiving, via a data interface, multi-variate time-series data; generating a plurality of tokens based on flattening the multi-variate time-series data; generating a first intermediate representation via a first cross-attention layer of the neural network based model with a plurality of dispatcher tokens as the query, and the plurality of tokens as the key and value; generating a second intermediate representation via a second cross-attention layer of the neural network based model with the plurality of tokens as the query, and the first intermediate representation as the key and value; generating a predicted time-series value based on the second intermediate representation; computing a loss based on a comparison of the predicted time-series value and a ground-truth value; and training the neural network based model based on the loss.
Database trigger firing techniques for reducing unnecessary trigger firings are disclosed. In one embodiment a computer system stores trigger information relating to initiating execution of at least one trigger instruction for a database in connection with a database operation statement. The trigger information includes a first set of one or more database field identifiers for a first set of one or more fields in the database and a second set of one or more database field identifiers for a second set of one or more fields in the database. The computer system receives a database operation statement and makes determinations that at least one field within the first set of fields and at least one field within the second set of fields is specified by the database operation statement. Based at least in part on the determinations, the computer system initiates execution of the at least one trigger instruction.
The disclosed techniques for generating a migration plan include identifying one or more entities that are eligible for data migration to a destination database from a source database. The techniques include generating, using planning procedures that include a workload balancing procedure, a data migration plan for the eligible entities and executing the migration plan. The workload procedure includes mapping, based on data metric values of the eligible entities, different ones of the eligible entities to instances in the destination database, where the mapping is performed based on utilization metric values of the instances, and where the instances are of a storage service that collectively implements the destination database. The workload balancing procedure includes altering the mappings of entities to instances in the destination database, where the remapping is based on a standard deviation of data for entities mapped to instances in the destination database not meeting a threshold standard deviation.
Methods, systems, apparatuses, devices, and computer program products are described. A processing device may receive a natural language query asking a question about a data metric. The processing device may use a large language model (LLM) to generate a summary of the natural language query for vector embedding. The processing device may determine one or more query response portions indicating possible answers to the query based on the summary and a vector database including vector representations of data summaries. To expand the scope of the answers, the processing device may recursively expand a set of data metrics for analysis. For example, the processing device may determine additional data metrics adjacent to the data metric of the query and may search the vector database for additional query response portions based on the additional data metrics. The processing device may use the query response portions to answer the natural language query.
Methods, apparatuses, and computer-program products are disclosed. The method may include transmitting, to a first generative artificial intelligence (AI) model, a request to convert a natural language description of a generative AI behavioral policy into a pseudo-code expression of the generative AI behavioral policy, where the generative AI behavioral policy describes conditions and actions to be performed based on the conditions; transmitting, to a second generative AI model, a prompt generated based on the pseudo-code expression of the generative AI behavioral policy, a user request, and an instruction that the second generative AI model is to generate a response to the user request; and receiving, from the second generative AI model and based on the prompt, an output of the second generative AI model and the output conforms with the user request and the pseudo-code expression of the generative AI behavioral policy.
Techniques are disclosed that pertain to profiling database statement execution. A computer system receives a request to profile the execution of a database statement by a database process. The request specifies a process identifier (ID) associated with the database process. The computer system initializes a profiling process to establish a profiling session in which the profiling process profiles the execution of the database statement to generate profiling results that identify a set of performances metrics associated with the execution of the database statement. The process ID is provided to the profiling process to establish the profiling session. The computer system detects an occurrence of a trigger event indicating that the profiling process should be terminated. The computer system terminates the profiling process in response to the occurrence of the trigger event. The profiling results may be stored in a storage repository accessible to the computer system.
Systems, devices, and techniques are disclosed for transaction coordination across services. A service including virtual threads receives a transaction that is part of a multi-part transaction. The service determines the transaction does not already exist in transaction state data. The service updates the status of the transaction in the transaction state data to pending. The service processes the transaction. The service updates the status of the transaction in the transaction state data to completed. The service uses receives at and endpoint a request for the status of the transaction. The service responds to the request for the status of the transaction. The service sends a request for status to an endpoint another service that is processing another transaction of the multi-part transaction. The service receives a response from the another service of a status of completed.
Methods, apparatuses, and computer-program products are disclosed. The method may include generating a first system message indicative of a role for the generative artificial intelligence (AI) model; generating a query-response message pair that includes a query message that may include an action invocation and a response message that includes information responsive to the action invocation; obtaining one or more interaction messages; generating a second system message that includes an instruction for the generative AI model to generate an utterance and an indication of one or more actions available to the generative AI model; transmitting, to the generative AI model, a prompt that may include the first system message, the query-response message pair, the one or more interaction messages, and the second system message; and receiving, from the generative AI model and based on the prompt, an output of the generative AI model.
Techniques for generating tasks from message data are discussed herein. A user profile may view content displayed within a communication-based virtual space. The user profile may request that the communication platform generate a task based on a content item displayed therein. That is, the user profile may be viewing a communication-based virtual space and select a message to use as the basis for creating a task. The communication platform may display one or more lists with which the user profile can associate the task. The user profile may select a list from the one or more of the lists and based on the selection, the communication platform may generate a task associated with the list. The communication platform may cause the task to be displayed via a user interface of a mobile device used by the user profile.
Techniques for modifying the display of task data are discussed herein. A user profile may request to modify a user interface object from a compressed state to an expanded state. Based on the request, the communication platform may convert the user interface object into an expanded state such that one or more columns of the task are presented in a single user interface. The communication platform may determine an updated organization of the columns and use such data to determine a modified size of the user interface object. The communication platform can generate a modified user interface object in an expanded state that includes an increased number of the columns. The communication platform may cause the modified user interface object to be displayed via a user interface of a mobile device.
H04M 1/72469 - User interfaces specially adapted for cordless or mobile telephones for operating the device by selecting functions from two or more displayed items, e.g. menus or icons
A method may include receiving, at a database that may include extension-based functionality, a database query request to perform a machine learning inference operation on data stored in the database, the machine learning inference operation to be performed at the database in accordance with the extension-based functionality. The method may include instantiating, in accordance with the extension-based functionality, a user-defined function (UDF) for performing machine learning inference operations. The method may include calling, with the UDF, the machine learning inference operation to process, at the database, the data retrieved from a table of the database. The method may include transmitting a response to the database query request, the response that may indicate an output of the machine learning inference operation, the output that may include a processed version of the data.
