111 results about "Specific model" patented technology

In various examples, self-supervised learning may be used to pre-train an encoder network of a masked prediction model to reconstruct masked regions of an input representation of 3D detections such as LiDAR point cloud(s). Spatial and / or temporal masking may be applied to a projected representation of 3D detections (e.g., a two-dimensional (2D) projection image), and the masked prediction model (e.g., a masked auto-encoder or joint-embedding predictive architecture) may be used to reconstruct a representation of the masked regions (e.g., reflection characteristic(s) stored in corresponding pixels or cells of the projected representation, a latent representation of the reflection characteristic(s)) during iterations of self-supervised learning. As such, the pre-trained encoder network of the masked prediction model may be used as a foundation model and fine-tuned with a task-specific output head or its pre-trained weights may be used to initialize a task-specific model.

One or more systems, devices, computer program products and / or computer-implemented methods of use provided herein relate to assessing certainty of artificial intelligence models used for detection or segmentation of pathologies. Accordingly, a system can comprise a memory that can store computer executable components. The system can further comprise a processor that can execute at least one of the computer executable components. The computer executable components can comprise a synthetic data generation component that generates biologically-inspired synthetic data that approximates a task-specific data manifold of a medical image from a radiomic features perspective; an artificial intelligence component that uses an artificial intelligence model to learn relevant representations of the synthetic data for an at least one image task; and a training component that utilizes the relevant representations and the artificial intelligence model to generate a task-specific model for the at least one image analysis task.

The invention relates to the technical field of large models, in particular to a reading recommendation large model generation method and system based on big data resources. The method comprises the following steps: acquiring a to-be-assessed text, collecting reading materials with multiple learning segments and multiple styles, labeling the reading materials, carrying out aggregation and consistency verification on labeling results, and constructing a verified recommendation database; performing low-rank adaptive fine tuning on the general pre-training large model based on the recommendation database to obtain a domain model, and performing performance and deviation monitoring and correction in the training process; performing multi-dimensional analysis on a to-be-evaluated text to generate a text feature vector, and performing mixed retrieval on a recommendation database in combination with semantic vector similarity retrieval; according to the invention, through the construction of the large reading recommendation model, the reading recommendation is more accurate and efficient.

The invention discloses a multi-modal large model adaptive iterative reasoning method and system for a long video, and belongs to the technical field of multi-modal artificial intelligence and reinforcement learning training. According to the method, a self-adaptive multi-round iterative reasoning framework is adopted, a model autonomously selects a direct answer according to a current reasoning state, and fine-grained time positioning or representation optimization of self-extraction visual clues is executed. Wherein in the fine-grained time positioning, a key time slice related to a task is dynamically selected for an output target interval through discrete marking; according to representation optimization, visual clues are extracted and injected into visual representation, and the model is supported to integrate dispersed multi-modal information in a long time range. Meanwhile, a reinforcement learning framework based on an entropy guided branch exploration mechanism and a composite reward mechanism is adopted, and an adaptive decision strategy of the model is enhanced. The method does not depend on a specific model structure, has good universality and expansibility, and can significantly improve reasoning integrity, stability and overall efficiency in a long video understanding task.

The application discloses an optical module, comprising: a first shell and a second shell, the first shell and the second shell surround to form a containing space, the containing space is used for containing a circuit board, a boss is arranged in the second shell, and the boss is located in the containing space, an adhesive is arranged on the boss, and the adhesive is used for fixedly connecting the circuit board and the second shell. The optical module does not need to additionally increase a jig, and the jig does not need to be maintained, so that equipment investment and maintenance cost can be reduced, the positioning precision of the jig can be avoided from being affected due to long-term use and abrasion, and the production rhythm can be avoided from being affected due to additional time required for installation and adjustment of the jig, so that the production efficiency can be improved. In addition, the jig is only suitable for a specific model, and the jig needs to be redeveloped when product design is changed, so that the development period is affected, and the boss can be compatible with circuit boards of different size specifications, and the production line flexibility is improved.

The invention discloses an on-demand data cleaning method and system oriented to a support vector machine, relates to the technical field of data cleaning, and provides a new on-demand data cleaning normal form for solving the problem that an existing data cleaning method is low in reliability of a cleaning result. The data and the downstream task model are regarded as a whole, and the data quality oriented to the specific model is further optimized on the basis that the downstream task effect and the generalization requirement are met. The normal form breaks through the limitation that a traditional method only pays attention to the quality of data, and the reliability of a data cleaning result is greatly improved.

This invention discloses a rapid method for detecting the moisture content of fine aggregates used in concrete, belonging to the field of building materials testing technology. The method involves collecting samples of different types of fine aggregates, preparing multiple moisture content samples from oven-dry to saturated surface-dry states, and then measuring the resistivity values ​​after compaction under constant pressure to form a dataset. Based on this dataset, a specific model is established using nonlinear function fitting, and a general prediction model is established using machine learning algorithms. During on-site testing, the resistivity values ​​of the fine aggregates to be tested are measured after compaction under the same constant pressure, and the moisture content is calculated according to the type of aggregate using the appropriate model. This invention achieves rapid and non-destructive testing of the moisture content of fine aggregates, providing real-time data for adjusting water usage in concrete production.

The present disclosure describes a method of unknown model based ML collaboration by configuring a set of the matching model condition identifiers with the associated parameter values in a wireless communication system including base station e.g., gNB, TN, NTN and mobile station e.g., UE. In AI / ML model is applied to radio access network, model performance is significantly impacted without specific model condition information. Therefore, model operation e.g., model training / inferencing / monitoring / updating is set up between network and UE by using the matching model condition information.

This invention discloses a system and method for measuring the moisture content of cow manure during collection and storage, specifically relating to the field of agricultural waste treatment technology. It includes a sampling unit, a weighing unit, and a calculation and display unit. By weighing a fixed volume of wet cow manure sample, its wet density is calculated. Then, using a wet density-dry matter content relationship model and a specific model for cow manure, the moisture content of the collected cow manure is quickly calculated and displayed. This invention overcomes the bottleneck of time-consuming traditional drying methods and the disadvantage of electrical methods being susceptible to interference. It can complete accurate measurements within minutes at the production site, providing efficient and reliable data support for real-time control of cow manure collection, transportation, solid-liquid separation, composting, and other treatment processes.

The application discloses an AI generated text detection method, system, device and medium, and relates to the technical field of text processing. The application converts a Chinese text sentence to be detected into a tone category sequence composed of tone categories. In this process, the categories of Chinese tones are used as a quantitative index of phonological structure, and text analysis is converted from a high-dimensional and high-cost semantic space to a low-dimensional and high-efficiency phonological feature space. The application is not dependent on training data of a specific model and is adaptable to text detection of multiple models. Then, N-gram analysis is performed on the generated tone category sequence, and the occurrence frequency of each N-gram combination is calculated to form a feature vector. Finally, an AI generated text is judged through a pre-trained classification model. In this process, the calculation type in processing is mainly string processing and frequency statistics, and the inconsistency problem of the model in multiple model invocations is avoided, so that the AI generated text can be accurately detected.

The invention belongs to the technical field of software engineering, and particularly relates to a probe generation method, system, equipment and medium for model coverage analysis, and the method comprises the following steps: traversing source codes of a whole model to identify operators needing instrumentation; aiming at each operator, generating a mapping file carrying a unique identifier according to a logic path recorded in the operator; wherein the mapping file defines a mapping relationship between the coverage state of each logic path and the bit of the integer value in the log entry; and generating one or more probes according to the number of logic paths recorded in the mapping file, and modifying source codes of the model according to source files of the probes to realize automatic instrumentation. According to the method, pre-compiling is not carried out during instrumentation, one or more probes with accurate matching interfaces are dynamically generated according to the monitoring requirements of a specific model, and compared with a limited preset traditional scheme, the number of the probes can be infinitely increased, and the actual requirements are met.

The present application relates to the technical field of feature vector merging in large model inference acceleration, in particular to a feature vector similarity measurement method and device based on transformation domain, by constructing a new type of calculation path of "transformation-measurement", using an orthogonal transformation matrix to project the feature vector to the transformation domain and then calculating the absolute value error sum, effectively extracting the internal structured features, solving the defect of insufficient accuracy of traditional lightweight measurement indicators, designing a multiplierless hardware acceleration architecture supporting the algorithm, by strictly constraining the element values of the orthogonal transformation matrix within the integer set of {-1, 0, 1}, completely equivalent to the matrix multiplication into simple addition and subtraction operation, completely avoiding multiplication, division and square root operation, greatly reducing the chip area and power consumption overhead, in addition, the pareto optimality of model accuracy and hardware efficiency is also realized, and the configuration capability of searching and customizing the transformation matrix for specific models is also provided.

The invention discloses a pulveryte lithofacies identification method based on deep learning, and relates to the technical field of deep learning and sedimentary rock lithofacies identification, and the method comprises the steps: collecting lithofacies curve data and microstructure image data, screening feature parameters, and generating a standardized multi-modal feature set through an exclusive analysis engine; calling a specific model to perform hierarchical feature extraction on the microstructure image to obtain a deep microstructure feature vector; inputting the two types of features into an exclusive network, and outputting a preliminary recognition result through cross-modal attention weight distribution; and performing inversion correction and dynamic adjustment on the preliminary result by adopting a specific algorithm, and constructing a lithofacies classification system in combination with feedback information to complete accurate division. According to the method, through multi-source feature deep fusion, cross-modal correlation enhancement and multi-link collaborative optimization, efficient and accurate identification of the lithofacies of the pulveryte rock is realized, complex lithologic features are adapted, and support is provided for technical application in related fields.

A method for metrics-based anomaly detection involves receiving an anomaly analysis request for an asset and obtaining metrics associated with the asset. Each of the metrics includes time series data. The method further involves detecting that one of the metrics is a counter. The detection involves seasonally differencing the metric, obtaining a regression line by performing a linear regression on the metric, and determining that an angle of the regression line exceeds a predetermined threshold angle. The method also involves training models for the metrics, the training including training a counter-specific model for the metric that is a counter. The method further involves determining, using the models after the training, at least one metric that is anomalous.

The application discloses a fine-grained sedimentary rock facies identification method based on deep learning, relates to the technical field of deep learning and sedimentary rock facies identification, and comprises the following steps: collecting facies curve data and microstructure image data and screening characteristic parameters, generating a standardized multi-modal feature set through a special analysis engine; calling a specific model to perform hierarchical feature extraction on the microstructure image to obtain a deep microstructure feature vector; inputting the two types of features into a special network to output a preliminary identification result through cross-modal attention weight distribution; and adopting a specific algorithm to perform inversion correction and dynamic adjustment on the preliminary result, combining feedback information to construct a facies classification system to complete accurate division. Through multi-source feature deep fusion, cross-modal correlation enhancement and multi-link collaborative optimization, the application realizes efficient and accurate identification of fine-grained sedimentary rock facies, adapts to complex lithological characteristics, and provides support for technical application in related fields.

The invention discloses a concrete material creep model parameter updating method based on monitoring data driving, and belongs to the technical field of civil engineering structure health monitoring. The method comprises the following steps: firstly, acquiring actual monitoring data of a target structure in a long-term service process; then calculating a theoretical response value by using a preselected creep model based on structural design and material parameters; constructing a linear regression relationship between the theoretical value and the measured value, and recognizing a regression coefficient by adopting a least square method; and finally, systematically correcting the native parameters of the creep model by using the coefficient to obtain an updated model suitable for a specific structure. According to the method, accurate conversion from a general model to a structure exclusive model is realized through a data driving mode, the accuracy of long-term performance prediction of the concrete structure is remarkably improved, and the method is particularly suitable for pre-stress loss evaluation and life prediction of major infrastructures such as a nuclear power station containment.

This invention provides a method, apparatus, device, and storage medium for generating teaching data. The method includes: acquiring user-inputted unit instructional design requirements; calling a pre-set domain-specific model; acquiring teaching intentions based on the unit instructional design requirements; retrieving knowledge information related to the teaching intentions from a domain knowledge base; calling the domain-specific model; and generating unit instructional design content based on the unit instructional design requirements and the knowledge information related to the teaching intentions. Furthermore, it can generate lesson-level teaching activity content based on the unit instructional design content, and can generate lesson-level teaching activity courseware based on the lesson-level teaching activity content. The teaching data generation method provided by this invention can intelligently generate teaching data, which can be directly used or modified by the user, and can also serve as a reference for the user's instructional data design, thereby greatly reducing the user's burden.

The invention relates to the technical field of intelligent classroom analysis and multi-modal data processing, in particular to a classroom intelligent analysis data acquisition method and system for cross-modal synchronization algorithm optimization. The method comprises the following steps: deducing a teaching event probability from multi-modal data through a specific model, and calculating a comprehensive acquisition value index in combination with cross-modal consistency and event time sequence logicality; the system performs real-time evaluation according to the value index, compares the value index with a preset threshold value, and generates a hierarchical acquisition instruction to adaptively control a data storage strategy; the method has the core advantages that through intelligent evaluation, only high-value teaching fragments are selectively stored, collection of invalid data can be effectively avoided, the cost of data storage and subsequent analysis is remarkably reduced, and the utilization efficiency of storage resources is greatly improved.

A computer system and associated processes for grouping similar real estate properties into contiguous neighborhoods and generating neighborhood-specific models capable of estimating property values within their neighborhoods. An artificial intelligence system directed to using a graph neural network framework to identify relationships between different parcel groups based on similar property features and embed the parcel groups into low dimensional space vectors. The method can include generating a graph and features relevant to the parcel groups that can train an embedding function that generate an embedding vector for each parcel group in a geographic unit grouping, such as a census tract. Embedding vectors of two or more parcel groups can then be compared to each other to determine whether the parcel groups are similar or to determine a housing valuation of a parcel group.

This invention discloses a collaborative confidence fusion perception method for autonomous driving, belonging to the field of autonomous driving. This method eliminates the differences in perception models between different intelligent agents, makes full use of the advantages of sensor data from different modalities, and avoids the operation of forcibly converting visual images that are not good at extracting BEV features into BEV features in order to achieve feature unification. Based on fully leveraging the advantages of different modal data features, learning to adapt to the features that specific models are good at, and the inherent collaborative advantages of collaborative perception, it achieves better collaborative perception results.

The invention discloses an intelligent voice interaction method and device, which are applied to the technical field of data processing, and the method comprises the steps: obtaining multi-modal original data of a user image, voice and text, processing through a specific model and a tool, detecting and cutting a human face through RetinaFace, then extracting an image emotion feature through a ResNet-50 network model, extracting a voice MFCC feature through a torchaudio library, and carrying out the recognition of the MFCC feature through the RetinaFace; the method comprises the following steps: extracting semantic features of a text by a Sension-BERT Chinese model, and generating three pieces of standardized single-mode feature data; and the dimension is unified through linear projection, a cross-modal attention mechanism and a Transform encoder fusion feature are combined, and an emotion classifier is input to obtain an emotion recognition result. Then, according to the Plutchik emotional wheel theory, a service scene and role constraint matching text and voice double-response strategy is combined, multi-modal interaction response content is generated, finally, continuous interaction is achieved in combination with intelligent hardware and circulating monitoring, and an intelligent interaction result meeting the emotion and scene requirements of a user is output.

This invention discloses a dynamic modeling and tracking control method for an adaptive wire-laying and compaction mechanism, belonging to the technical field of specific model calculation systems. The method includes the following steps: S1: Establishing a kinematic model of the adaptive wire-laying and compaction mechanism; S2: Establishing a workpiece surface model and solving for the coordinates of the tangent points and positions of the driving and driven wheels with the ground; S3: Solving for the relationship between the reaction forces and external forces of each support based on the force balance equations and moment balance equations of each rod; S4: Establishing a dynamic model of the driving device; S5: Performing error analysis and optimization to ensure that the main pressure is within the fluctuation range. This invention can ensure a constant roller output pressure when laying steeply varied surface structures, avoiding forming defects caused by uneven pressure.

This invention discloses a method and system for predicting the remaining lifespan of individual sensors based on Bayesian statistics, relating to the field of equipment health status monitoring and lifespan prediction technology. The invention provides a method comprising: establishing a general degradation model and prior distribution using historical degradation data; collecting monitoring data of a specific target sensor under real or accelerated stress in the field; updating the posterior distribution of model parameters based on Bayesian statistical inference and Gibbs sampling to generate a specific degradation model for that individual sensor; and calculating the remaining lifespan based on this specific model and a failure threshold. This invention achieves a breakthrough from "group lifespan assessment" to "accurate prediction of individual lifespan" through data-driven adaptive correction, significantly reducing over-maintenance costs and enhancing equipment operational safety.

A system is provided.SOLUTION: A system comprising: a default setting unit configured to allow a user to register; a unit configured to collect and store conversation data with the user; a unit configured to analyze the collected conversation data and learn a manner of speaking and a context of the user; a unit configured to train a user-specific model based on the learned data; a unit configured to generate a reply of the conversation using the generated user-specific model; and a unit configured to transmit the generated reply to a user terminal.SELECTED DRAWING: Figure 1

This invention relates to the fields of computer vision and public safety technology, and in particular to an adaptive multi-scale dynamic fire detection method and system based on few-shot learning. The method includes: constructing a multi-scale semantic attribute library; generating a multi-scale dynamic feature map based on video sequences, fusing color, temporal motion, and multi-scale optical flow information; acquiring visual and textual feature vectors, and constructing a basic model through cross-modal semantic alignment via contrastive learning; calculating and fusing visual and semantic prototypes to obtain a scene prototype, and fine-tuning the basic model to generate a scene-specific model; acquiring real-time video sequences and calculating real-time multi-scale dynamic feature maps, and extracting the cosine similarity between real-time feature visual vectors and scene prototypes, thereby performing real-time fire detection. Because multi-scale optical flow information is extracted, it can accurately capture dynamic characteristics across all scales, from microscopic flickering to macroscopic spread, enabling efficient fire detection.

The invention relates to a method for realizing face recognition confrontation sample generation based on frequency domain decomposition and an attention mechanism, and the method comprises the following steps: carrying out the data preprocessing, and carrying out the model construction; generating an attention mask M; performing frequency domain decomposition by using a discrete wavelet transform algorithm; constructing a joint loss function J; frequency domain disturbance iterative optimization is carried out, and disturbance variables are initialized in a frequency domain space; and repeating until an attack stop condition is met, and outputting a final confrontation sample image. The method, the system, the device, the processor and the computer readable storage medium for generating the face recognition adversarial sample based on the frequency domain decomposition and the attention mechanism have the advantages of high mobility, high concealment and high robustness, and the adversarial sample generated based on the frequency domain has higher resistance to common image processing operation; the algorithm is universal, and the proposed framework does not depend on a specific model structure and can be flexibly applied to various CNN-based face recognition systems.

According to one embodiment, a method, computer system, and computer program product for generating synthetic business documents for data augmentation is provided. The embodiment may include identifying a first set of key value pairs (KVPs) within a set of business documents spanning multiple domains. The embodiment may include creating domain-agnostic models of spatial distribution and content distribution of KVPs within the set. The embodiment may include grounding the domain-agnostic models of spatial distribution and content distribution using a second set of domain-specific KVPs to derive domain-specific models of spatial distribution and content distribution of KVPs within the second set. The embodiment may include generating a set of synthetic domain-specific business documents using the derived domain-specific models of spatial distribution and content distribution. The embodiment may include augmenting a training data set of a large language model (LLM) with the set of synthetic domain-specific business documents.

A time-shifting high-density electrical method data temperature and humidity noise reduction method fused with a KNN algorithm comprises the following steps: 1, arranging a field acquisition system, and obtaining original observation data; 2, data preprocessing; and 3, constructing a noise reduction model and outputting. The method comprises the following steps: establishing an apparent resistivity and temperature and humidity data set aligned in time and depth dimensions by constructing a dynamic noise reduction system; by introducing a time shift reference and machine learning driven interpolation strategy, noise reduction is realized through coupling calculation of time shift data and influence factors, the problem of error accumulation in traditional static correction is effectively reduced, time-space dynamic accurate correction of temperature and humidity interference is realized, and the accuracy of temperature and humidity interference correction is improved. The data quality and the interpretation accuracy of the high-density electrical method in a complex environment are obviously improved; a model with site specificity is fitted by using a long-term monitoring time-shifting data set, the dynamic relationship between temperature and humidity and apparent resistivity is quantified, and the separation of environmental noise is realized, so that the data signal-to-noise ratio of a high-density electrical method and the detectability of weak signals are improved.

Existing segmentation models have the disadvantage that they cannot be deployed on edge devices as they are not compact and require large space. Embodiments disclosed herein provide a method and system for generating a class focused segmentation model using a two stage model training approach, which enables generation of a compact, task specific model. Using a first stage of training, a Machine Learning (ML) segmentation model that is trained on a master training dataset is generated. Further, in a second stage of training, a fine-tuned task specific segmentation model is generated, which can be used for task specific class segmentation. The fine-tuned task specific segmentation model being task specific and in turn compact, can be used edge device deployment and such applications.

A method for testing single event effects in a SiP device. The method comprises: on the basis of depth information of a sensitive area of a SiP device, decapping or thinning the SiP device; on the basis of the structures and models of embedded chips of the SiP device, determining a sensitive module of each embedded chip therein, and performing irradiation test analysis on each sensitive module, so as to obtain single event effects test items for the SiP device; determining a test plan, and performing a targeted design on test software and hardware; injecting an error mode into the designed software and hardware to perform simulation verification, and if the simulation verification fails, modifying test procedures, and re-performing the simulation verification; performing a single-event test on each embedded chip; performing an irradiation test on the SiP device; and after the test is completed, uploading, saving and analyzing test data, and ending the test. The present method solves the problems of a ground test of SiP single event effects being inaccurate, and the test method and a test apparatus being limited to specific models.