64 results about "Learning abilities" patented technology

The invention relates to the technical field of heart failure auxiliary diagnosis and treatment, in particular to a heart failure auxiliary diagnosis and treatment knowledge distillation method and system.The knowledge distillation method comprises the steps that an age-crossing fusion data set is built by combining ancient medical case text data related to traditional Chinese medicine heart failure; performing field fine tuning on the pre-training model to generate a basic teacher model; performing semantic alignment, and outputting a cross-time contrast database; constructing a curative effect scoring model by using the time sequence model, and outputting a quantitative association database; using a graph neural network to output an intelligent heart failure knowledge graph, and generating a comprehensive teacher model; performing knowledge distillation on the comprehensive teacher model to generate a clinical student model; the knowledge distillation system is applied to the knowledge distillation method, has a dynamic learning ability, can fuse ancient and modern trans-time knowledge and combine traditional Chinese medicine ancient book wisdom to carry out heart failure auxiliary diagnosis and treatment, and can give consideration to model performance and deployment efficiency through knowledge distillation.

The invention relates to the technical field of natural language processing, in particular to a data enhancement and feature enhancement-based Chinese named entity recognition method, which is based on a feature enhancement double-attention named entity recognition model, and is characterized in that the model comprises an embedded layer, a data enhancement module, a multi-scale convolution fusion attention layer and a prediction layer; the embedded layer uses a dual-channel attention fusion module to process texts in parallel, fuses multi-dimensional information features of Chinese characters, combines an error correction type mask language model, a pre-training model and a bidirectional gating loop unit, fuses local and global text features, and obtains text representation from multiple dimensions and multiple levels; according to the method, by introducing an innovative model mechanism or training strategy, challenges such as label sparsity and text noise existing in a Chinese named entity recognition task can be effectively handled, the learning ability and generalization performance of the model for long-tail entity categories are improved, and the robustness of the model in a real and non-ideal data environment is enhanced.

The invention discloses a reinforcement learning intrusion detection method based on space-time attention and dynamic courses, and the method comprises the steps: carrying out the dynamic weight distribution of input features in space and time dimensions through a multi-level space-time attention mechanism, and enhancing the expression capability of key features; a dynamic curriculum learning strategy is adopted to realize progressive difficulty adjustment of training samples, and the learning ability of the model is gradually improved through initial difficulty estimation, dynamic threshold adjustment and curriculum sample selection; designing a composite reward mechanism to optimize a reward signal, and combining a basic classification reward and a stability reward to promote stable convergence of the strategy; a self-evolution target network module is introduced, and dynamic update management of a target network is realized through performance monitoring, emergency update triggering and frequency self-adaption. A space-time cooperative detection closed loop is constructed, and an end-to-end intrusion detection process is realized through environment interaction, strategy optimization and online detection. The method effectively improves the accuracy and robustness of intrusion detection, and adapts to a complex network attack scene.

The invention relates to an RDFS ontology-oriented large language model evaluation and cue word optimization method. The method comprises the following steps of 1, constructing a positive example data set and a negative example data set; 2, designing cue word templates of multiple levels; 3, test cases in the standardized test data set are combined with different types of cue word templates, an input sequence is generated and sent to a to-be-evaluated large language model, and an output judgment result and an explanation text are obtained; 4, analyzing the judgment result and the explanation text to generate a comprehensive evaluation report; and 5, based on a data analysis result of the comprehensive evaluation report, forming a system performance map by constructing a'cue word type-evaluation dimension-axiom type 'ternary mapping relationship, and implementing an optimization process according to the system performance map. According to the scheme, the problems of incomplete evaluation system, imperfect test data, single evaluation dimension and the like in the prior art are solved, and systematic and precise evaluation and improvement of the model RDFS ontology learning ability are realized.

The invention provides a training method of a face recognition model, and a face recognition method and system, and relates to the technical field of face recognition, and the method comprises the steps: pre-training a teacher network; constructing a student network and establishing a knowledge distillation framework; dividing the feature vector output by the teacher network into blocks, and applying different weights according to the importance of each block so as to form weighted block mean square error loss; the cosine similarity of feature vectors output by the teacher network and the student network is used as a difficult sample indicator; the angle interval in additive angle interval loss is dynamically adjusted according to the indicator, and differential constraints are applied to samples with different difficulty levels. According to the method, the learning ability for key features is enhanced through weighted block mean square error loss, attention distribution for difficult and simple samples is optimized through dynamic additive angle interval loss, the identification precision, convergence speed and model compression effect of a student network are effectively improved, and the method is suitable for resource-limited edge device deployment.

The invention is suitable for the technical field of post matching, and provides a multi-dimensional intelligent matching method and system for employment posts, and the method comprises the following steps: carrying out the comprehensive skill matching of posts and talents, and determining a comprehensive skill matching score; calling educational background information and certificate information in the talent resume, and reasoning to determine the learning ability of the talent; extracting post skills which are matched unsuccessfully in the employment posts, determining the demand emergency degree and learning difficulty of each post skill which is matched unsuccessfully, and determining the association degree between the post skills and talent skills and the association degree between the post skills and professions; and based on the learning ability, the demand emergency degree, the learning difficulty and the correlation degree, determining the cultivation cost cardinal number of each post skill which fails in matching. According to the invention, the learning ability of talents is incorporated into a matching evaluation system, so that potential suitable talents with relatively strong learning ability and development potential although the current skills are not completely matched with posts can be identified, the talent selection range is expanded for enterprises, and the talent recruitment efficiency is improved.

The application discloses a kind of teacher teaching ability promotion training method and system, its method includes: determining multiple evaluation dimensions of teacher teaching ability, obtaining the quantitative evaluation index of each evaluation dimension, obtain multiple high-quality training course resources according to quantitative evaluation index;The historical teaching state parameter of each teacher is input and the teaching ability of each teacher is evaluated;According to the evaluation result, the teaching short board index and short board state description of each teacher are analyzed and the ability promotion demand parameter is determined;According to the ability promotion demand parameter, the target training course resource is filtered from multiple high-quality training course resources, and the training plan is formulated based on the learning ability information of each teacher, and the teaching ability promotion training is carried out according to the training plan for each teacher.It can accurately aim at different teaching level teachers for accurate short board related ability training, which improves the efficiency and stability of training, and also guarantees the adaptability of training resources, improves the practicability.

An object of the system according to the embodiment is to support creation of a reading comment, improve academic ability of a child, and reduce a burden on a guardian.SOLUTION: A system includes a basic rule presentation unit, a support unit, a correction unit, and a recommendation unit. A basic rule presentation part presents a basic rule and a writing method for creating a reading comment. The support unit supports creation of a comment based on the basic rule and the writing style presented by the basic rule presentation unit. The correction unit corrects the comment created by the support unit. The recommendation unit provides a recommendation of a writing style according to the content based on the information of the book registered in the database.SELECTED DRAWING: Figure 1

Embodiments of the present application provide a BERT model fine-tuning method and device, equipment and a storage medium, and relate to the technical field of computers. The method comprises: generating an original token list according to a text corpus, and obtaining word frequency statistical information of the original token list; determining a mask generation strategy based on the word frequency statistical information, and performing mask processing on the original token list according to the mask generation strategy to obtain a mask processing result; and performing iterative learning on a pre-trained BERT model based on the mask processing result until a preset condition is met, to obtain a fine-tuned BERT model. The present application statistically obtains word frequency information corresponding to a text corpus, and adaptively adjusts a mask generation probability according to the word frequency, so that the mask distribution generated by tokens of different frequencies is more uniform, and the semantic information of the text corpus is better preserved, thereby improving the learning ability of the BERT model for masked corpus.

The invention provides an attack technology prediction method and device based on a large language model. The method comprises the steps of obtaining historical attack data of an attacker and performing data modeling to construct a bipartite graph; generating a structured context according to the bigraph, and inputting the structured context into a large language model; and constructing a prediction cue word, guiding the large language model to carry out attack technology prediction according to the structured context by the prediction cue word to obtain a preliminary prediction result, and carrying out attack stage filtering and sorting screening on the preliminary prediction result according to a set standard to obtain a final attack technology prediction result. By applying the method, the recommendation ability of the large language model can be stimulated, the adaptability and learning ability of the large language model to a specific security task can be improved, the scene learning ability of the large language model can be fully utilized, a new attack mode can be effectively dealt with, the dependence on a large amount of annotation data is reduced, and the user experience is improved. Particularly, under the condition that data is scarce or the cold start problem is serious, the accuracy of the generated result can be improved.

The invention discloses a learning ability dynamic evaluation method and system based on a stylus writing track. The method comprises the steps that N sample objects with object IDs are selected; constructing N writing behavior sequences; encoding N sequence representation vectors; performing unsupervised clustering on the N sequence representation vectors; determining the most similar behavior cluster of the target object; extracting a standard writing behavior sequence of the most similar behavior cluster, and calculating a sequence deviation between the standard writing behavior sequence and the real writing behavior sequence; if the sequence deviation is greater than a first threshold value, determining that the writing track of the target object is unqualified; otherwise, extracting the writing rhythm ratio of the target object to perform learning abnormity label evaluation; according to the method, the standard sequence can better represent the stable behavior pattern of the excellent students of the cluster, and the sequence deviation can more accurately reflect the deviation degree of the target object and the normal behavior, so that misjudgment caused by standard distortion is avoided, and the evaluation effectiveness of the learning ability is improved.

The invention relates to an intelligent reading recommendation method and system based on dynamic learning ability evaluation, and relates to the field of intelligent education technology.The method comprises the steps that current scanning content is extracted based on scanning behavior data, knowledge content mastered by a user is recognized in combination with historical learning records, and the knowledge content is removed to obtain an unmastered knowledge range; analyzing the current scanning content to identify knowledge entries and evaluating the learning difficulty of the entries; based on historical learning records and knowledge entries, evaluating expected time required for understanding the current scanned content; determining an actual understanding time based on the scanning behavior data; the real-time learning ability of the user is generated in combination with the actual understanding time, the expected time and the entry learning difficulty; and generating learning recommendation content based on the real-time learning ability and the unmastered knowledge range, and performing content recommendation based on the learning recommendation content. The method has the effect of improving the learning efficiency and experience of the user.

The invention discloses a radiation source individual identification method and system based on large language model reprogramming. The radiation source individual identification method comprises the steps that received radio frequency signals are preprocessed and partitioned; through a signal reprogramming module, by utilizing a trainable text prototype library and a multi-head cross attention mechanism, mapping the signal fragments into a representation sequence aligned with the semantic space of the large language model; constructing a natural language prompt, converting the natural language prompt into a vector, splicing the vector with the signal representation, and inputting the spliced vector into a parameter-frozen large language model for feature extraction and aggregation; and finally, outputting an identification result through the lightweight trainable classification layer. According to the method, the general mode recognition and few-sample learning ability of a large language model are utilized, only a few parameters such as a reprogramming module and a classification layer need to be trained, and the problems that a traditional method is poor in generalization, depends on a large amount of annotated data and lacks knowledge guidance are effectively solved; rapid, accurate and explainable identification of radiation source individuals in a high-dynamic electromagnetic environment is realized.

The present disclosure belongs to the technical field of nuclear power and specifically relates to a vulnerability information identification method and device based on a large language model. The present disclosure collects PoC information of certain CVEs and manually labels part of the data, thereby enhancing the accuracy and reliability of the training data in a manual labeling manner, so as to improve the learning ability of the subsequent model. The GPT-3.5-turbo model is used to identify and summarize the CVE and PoC data, each piece of information is described based on the TTP content form, the CVE and PoC information are unified into the TTP standard format, the noise interference in the original information is eliminated, the identification and learning ability of the large language model for the key information features can be enhanced, and the instruction fine-tuning technology is used to improve the task completion effect of the GPT model. Based on the Llama-2 model after fine-tuning, the input target CVE and PoC information are identified and processed, and the correlation between the CVE and the PoC is determined based on the output. Thus, the problem of relying on manual analysis in the CVE and PoC correlation identification is solved.

The present invention relates to a method for providing a learner-personalized language learning service and, more specifically, to a method for providing a learner-personalized language learning service, which is performed by a server and provides a Korean learning service to a foreign learner. The method comprises the steps of: preparing, by the server, an evaluation corpus including a learning corpus for allowing a learner to learn Korean and an error corpus in which an error made by the learner is expressed; receiving learner information input using a learner terminal; setting a Korean language ability level of the learner on the basis of the learner information; and allowing the learner to learn by using the evaluation corpus according to the set Korean language ability level of the learner. The present invention provides language learning data according to an individual language learning ability level of a learner and a subject desired by the learner, and enables personalized language learning to be performed for each learner, thereby improving learning efficiency and satisfaction.

The application belongs to the field of intelligent education, and particularly relates to an artificial intelligence education system based on education big data, which comprises a student terminal and a server, and is characterized in that a user verification module, a learning module, a handwork module, an entertainment module, a feedback module and an analysis module are arranged in the student terminal; the student terminal comprises a main body and a time module; the time module is used for controlling the working time of the learning module, the handwork module and the entertainment module; the learning module and the entertainment module are arranged to allow students to have proper rest during the learning process, so as to achieve the combination of work and rest; the handwork module is arranged to improve the attention and the hand-on ability of the students, so as to improve the learning ability of the students; the learning result, the attention and the hand-on ability of the students are tested to measure the personal comprehensive level of the students, and the personal comprehensive level of the students is compared with the level of the students of the same age, so that the personal characteristics of the students can be better judged, and the teaching plan can be formulated in a targeted manner, so as to improve the education effect.

The invention discloses a mathematical question answering auxiliary system based on artificial intelligence, and relates to the technical field of artificial intelligence, the mathematical question answering auxiliary system comprises a management center, and the management center is connected with a question input module, a language processing module, an intelligent engine module and an answering output module; performing solution decomposition on the collected to-be-solved mathematical question to obtain analysis knowledge points, and setting learning reference points to perform resource supplementary matching to obtain recommended learning materials; performing question detection on the solving process of the to-be-answered mathematical question by visiting students to obtain question learning points, and feeding back the question learning points to the management center for feedback supplement to obtain question supplement data; performing learning recording on the visited students through the question supplement data to obtain a learning feedback result, and performing optimization answering on the to-be-answered mathematical question through the learning feedback result to obtain an optimal answering process; the accuracy and efficiency of mathematical question answering are improved, and the self mathematical literacy and learning ability are improved by supplementing learning resources.

The invention discloses a learning material recommendation method and device based on an AI agent, and relates to the technical field of education, and the method comprises the steps: obtaining the historical learning data of a student and a knowledge point learning grid used by the student from a learning accompanying system; based on historical learning data of the students, determining learning ability data of the students and learning state information of the students on knowledge points in the knowledge point learning grids; determining a to-be-learned plan type of the student based on the learning ability data and the learning state information, and determining a to-be-learned knowledge point set in the knowledge point learning grid according to the to-be-learned plan type; and generating a to-be-learned path of the student in the knowledge point learning grid based on the to-be-learned knowledge point set, and determining a to-be-learned material set matched with the to-be-learned path from the learning materials to recommend the student to learn the to-be-learned knowledge point set based on the to-be-learned material set. According to the method, collaborative matching of the learning materials and the learning paths is realized, and the pertinence of learning material recommendation is improved.

The invention provides a cue word optimization method, which specifically comprises the following steps that: a plurality of language models are selected as teacher models, the learning ability of the trained teacher models is distilled to student models, and the parameter quantity of the student models is far smaller than that of the teacher models; comment information of a user is input into the BM25 model, the BM25 model extracts a quality label and a word-of-mouth label which are highly related to the comment information, the extracted quality label and word-of-mouth label are put into a cue word template, and the distilled student model extracts the quality label and the word-of-mouth label corresponding to the comment information based on cue words. In order to reduce the dependence on hardware resources, the model parameter quantity is reduced through a model distillation mode, and meanwhile, in order to balance the model capacity loss possibly caused by the reduction of the parameter quantity, a model integrated distillation mode is adopted; in order to improve the accuracy of multi-label extraction, the length of cue words is adaptively adjusted by introducing BM25 and other technologies to relieve the instruction following problem.

The application belongs to the technical field of data processing, and provides a learning ability evaluation model construction method for individual learning of an online learning platform, which comprises the following steps: firstly, collecting behavior data and learning resource metadata of users in a learning process; constructing an individual behavior heterogeneous graph, a knowledge system hypergraph and a knowledge point sequential graph based on the collected data; then, presetting a learning scene attention matrix, combining the knowledge point sequential graph and determining a context perception mechanism; finally, constructing a learning ability evaluation model according to the context perception mechanism, the individual behavior heterogeneous graph and the knowledge system hypergraph. The application realizes accurate modeling of the complex interaction relationship among the user, the learning resource and the knowledge point by constructing the individual behavior heterogeneous graph, the knowledge system hypergraph and the knowledge point sequential graph, deeply fusing the behavior data generated by the user in the learning process with the metadata such as the type, the difficulty and the associated knowledge point of the learning resource, fully showing the systematicness and the dynamics of the knowledge system, and effectively improving the accuracy of the learning ability evaluation.

This invention discloses a training method for a reward model, an optimization method for a large language model, and related methods. The training method for the reward model includes: obtaining preference training sample pairs and a reward model to be trained, wherein the preference training sample pairs include preferred response samples and non-preferred response samples; calculating the reward score difference between the preferred response samples and non-preferred response samples based on the reward model to be trained; constructing a cost matrix based on the reward score difference and the semantic correlation between the preferred response samples and non-preferred response samples; calculating the loss margin based on the cost matrix; calculating the pairwise preference loss value with the margin based on the loss margin, and updating the parameters of the reward model to be trained with the goal of minimizing the loss value with the margin, thereby obtaining a trained reward model. This improves the model's learning ability on difficult samples and its overall generalization performance, avoids over-reliance on simple samples, and thus improves the generation quality of large language models in complex tasks.

The application discloses a model backdoor attack confrontation method based on frequency domain feature fusion reconstruction. The method filters the feature map in the frequency domain by using Fourier convolution on the feature map set of the student model to remove the backdoor attack mode injected in the time domain; semantic information is cascaded and fused from the deep feature map to the shallow feature map in sequence, so that the output of the whole student model becomes a whole, thereby increasing the semantic information that can be learned by the student model in the process of matching the feature map of the teacher model and weakening the attack backdoor based on local information that may exist; attention operation is used on the fused feature map to enhance the shallow semantic information density by using deep high-order semantic information between adjacent output feature map layers, thereby improving the learning ability of the student model and obtaining higher training precision. The application can learn and obtain a student model with high precision and capable of removing the time domain attack backdoor and the attack backdoor based on local information on the basis of a pre-trained model from an untrustworthy source.

The invention discloses a learning ability evaluation method based on label aggregation and behavior analysis and a medium. Obtaining a current to-be-evaluated task and current joint description data; performing aggregation standardization processing on the multi-source tag description data through a tag aggregation processing method to obtain at least one current standardized tag; processing through a sub-label behavior trend analysis and calculation method to obtain each current weighted learning intensity index, a current behavior trend slope and a current behavior trend confidence coefficient; and obtaining and performing calculation processing through a sub-label capability difference value contribution degree calculation method according to a standard capability requirement to obtain and feed back a target sub-label capability difference value and a target sub-label difference value contribution degree so as to complete evaluation processing of the learning capability of the current to-be-evaluated task. The problems of label redundancy, poor ability evaluation accuracy, insufficient interpretability of the user ability development trend and poor flexibility are solved, and the accuracy, comprehensiveness and flexibility of ability evaluation are improved.

The invention discloses an education platform resource optimization management system and method based on a big data model, and belongs to the technical field of education resource management. The system comprises a test question database, an education platform, a data acquisition module, a model analysis module and an intelligent query push module. The test question database is used for storing test questions and corresponding knowledge points; the test question database sends test questions to the education platform; the education platform is used for the user to answer the test questions in the test question database; the data acquisition module is used for acquiring learning data of a user; the model analysis module is used for determining the distinction degree and difficulty parameters of each test question, and determining the learning ability of a user and the mastery degree of different knowledge points; the intelligent query and push module is used for querying homework test questions conforming to the test question difficulty of the user and pushing the homework test questions to the education platform; according to the method, the homework test questions conforming to the user are pushed through intelligent query, user requirements are accurately matched, and the learning efficiency of the user is improved.

The invention discloses an automatic driving reinforcement teaching learning method, device and equipment, a medium and a product. The method comprises the following steps: in a simulation platform, constructing a reinforcement learning abstract environment containing a state, an action and a reward function based on an automatic driving scene; obtaining multi-source trajectory data in the reinforcement learning abstract environment, wherein the multi-source trajectory data comprises a preset teaching state transition trajectory, a different-orbit state transition trajectory and a same-orbit state transition trajectory; clustering the multi-source trajectory data to obtain a clustering result, performing probabilistic sampling according to the clustering result, and constructing a mixed experience playback pool; and carrying out automatic driving reinforcement teaching learning based on the mixed experience playback pool. According to the scheme, teaching knowledge used in imitation learning is integrated into a traditional reinforcement learning method, and the sampling and utilization mode of a probabilistic sampling optimization sample is combined, so that the learning ability of the traditional reinforcement learning method is kept, and meanwhile, the automatic driving learning process is more stable.

This invention discloses an intelligent identification method for creep-type landslide hazards that combines image processing and semantic understanding. First, high-precision geometric modeling and filtering techniques are used to suppress irrelevant phases in InSAR interferograms while preserving deformation information related to landslides. Next, signal features are enhanced through phase gradient, RGB channel mapping, and generative adversarial networks to improve signal representation capabilities. Subsequently, convolutional operations, self-attention mechanisms, and structural reparameterization are used to optimize the deep learning model, improving its learning ability for landslide hazards. In the feature encoding stage, a multimodal caption generation model is used to obtain visual and linguistic features, which are then mapped to the same embedding space for seamless integration. Finally, an autoregressive language generation model and a multilayer perceptron are used to predict landslide hazards. This approach, by integrating visual, linguistic, and deep learning technologies, significantly improves the efficiency and accuracy of landslide hazard monitoring, providing effective support for disaster response and management.

[Problem] To provide improved dictionary lookup learning materials and a dictionary lookup learning method that improve upon existing dictionary lookup learning methods utilizing Japanese language dictionaries and sticky notes, and are linked to digital technology in order to adapt to advances in educational digital transformation. [Solution] A vocabulary list table is used as a learning material to be used in a dictionary lookup learning method for increasing vocabulary and improving academic ability by referencing a dictionary, wherein: grid squares are provided; a vocabulary word has been entered into each grid square or a new word can be entered thereinto; and grid squares in which vocabulary words are entered that have been indicated as known by the learner through dictionary lookup or grid squares in which the vocabulary words that have been indicated as already known have been entered can be colored using colors corresponding to hues determined by the parts of speech of the vocabulary words indicated as known and to the learner's degree of understanding of the vocabulary words indicated as known, with greater brightness used for greater understanding, and greater colorfulness used for a higher degree of utilization corresponding to the learner's degree of utilization of the vocabulary words indicated as known.