12results about How to "Reduce forgetting" patented technology

The invention provides a relation extraction model construction method and system based on semantic enhancement and knowledge distillation, relates to the technical field of relation extraction, and aims to solve the problem that in the prior art, when continuous learning is processed, a model is often difficult to keep memorizing an old relation when a new relation is learned, and forgetting is caused. Model overfitting is easily caused by excessive dependence on a small amount of data, so that a new relationship cannot be effectively generalized. The method comprises the steps of obtaining a relationship prototype of a training sample and a typical sample, performing semantic enhancement on the relationship prototype to obtain an expanded enhanced data set and an enhanced memory set, and performing feature extraction, relationship classification prediction and relationship extraction based on a student model and a multi-teacher model. According to the method, the overfitting problem caused by scarcity of small sample data is effectively relieved through a semantic enhancement technology, old relation forgetting caused by new relation learning in the continuous learning process is reduced through a multi-teacher knowledge distillation method, and the classification accuracy and generalization ability of the model facing different relations are improved.

The invention relates to a class increment image classification method and system based on knowledge distillation. The method comprises the following steps: generating output representations of old class training samples among old classes through a teacher model, and respectively carrying out Z score standardization to obtain first standard output, generating old class training samples, inter-class output expressions of the new class training samples between new classes and old classes and intra-class output expressions of the old class training samples between the old classes through a student model, and respectively performing Z score standardization to obtain second standard output and third standard output; calculating inter-class distillation loss according to the first standard output and the second standard output, calculating intra-class distillation loss according to the first standard output and the third standard output, and obtaining total loss according to the inter-class distillation loss, the intra-class distillation loss and the classification loss; and optimizing the total loss of the image classification model to obtain a trained image classification model, and carrying out image classification by adopting the trained image classification model. By adopting the method, the accuracy of image classification can be improved.

The invention relates to a home-based care technology for the aged, is used for solving the problems of difficulty in shopping, damage to the body caused by large activity and difficulty in distinguishing food quality in home-based care for the aged, and particularly relates to a home-based care smart cloud system based on the Internet of Things, which comprises a shopping auxiliary module, a health management module and a service evaluation module, according to the invention, the shopping auxiliary module is combined with the health management module to judge the activity amount of the elderly, so that the situation of shopping difficulty caused by non-support of the activity amount of the elderly is avoided; the health management module compares the activity data of the elderly with the optimal activity to judge whether the elderly is suitable for self-shopping, so as to prevent the physical health of the elderly from being damaged due to excessive activity; whether an article is fresh or not is judged through the service evaluation module according to quality assurance information of an article package and color difference comparison, the situation that old people eat deteriorated food to affect body health is avoided, and service scoring evaluation is conducted on stores in a set range in the aspects of pre-stocking and expenditure amount through the service evaluation module.

The invention belongs to the technical field of computer vision, and particularly discloses a generative playback-based category increment target detection method, which comprises the following steps of: obtaining a fine-tuned Stable Diffusion model; using the fine-tuned Stable Diffusion model to generate image-level playback data corresponding to the old task and the new task, performing pseudo-label labeling on the image-level playback data, and performing confidence sampling to obtain initial image-level playback data; screening the initial image-level playback data to obtain final image-level playback data; and combining the final image-level playback data with the real data of the new task to obtain a complete training data set to train the detection model to obtain a trained detection model. According to the invention, the quality of the generated data in generative category increment target detection can be improved. The problem that in the prior art, the quality of generated data is not high when a model is used in a generative category increment target detection method is solved.

The invention relates to a communication signal modulation identification method and system based on task incremental learning, belongs to the field of communication modulation identification, and solves the problem of low communication signal modulation identification precision in a dynamic spectrum environment caused by a catastrophic forgetting problem. Comprising the following steps: respectively generating modulation signal data of a plurality of tasks in a dynamic spectrum environment, and taking corresponding modulation modes as sample labels to obtain modulation signal data sets under different tasks; a communication signal modulation identification model is initialized, an experience playback buffer area is created, and the experience playback buffer area is used for storing sample data trained each time and weight parameters of the correspondingly obtained model; performing task incremental learning training on the model by using modulation signal data sets under different tasks and sample data stored in the experience playback buffer area to obtain a trained model; communication signals in a dynamic spectrum environment are obtained in real time and input into the trained model, a corresponding communication signal modulation mode recognition result is obtained, and accurate recognition of the modulation mode is achieved.

The application discloses an indoor robot vision navigation method based on an optimal strategy playback, which is divided into an on-policy algorithm and an off-policy algorithm, current navigation targets are trained in the on-policy, and old navigation targets are reviewed in the off-policy to reduce catastrophic forgetting. The method comprises a single feature extraction network and a strategy network. The input of the strategy network is the output of the state feature extraction network, namely, the advanced feature map of a current image. The output of the whole network is two functions: a strategy function pi (a|s) and a value function Q (s|a). The whole training process is divided into two stages: an on-policy stage and an off-policy stage. In the on-policy stage, an agent learns the strategy of a new target through interaction with an environment, and in the off-policy stage, the agent reviews the learned strategy by using the optimal experience of old targets stored in a memory to prevent catastrophic forgetting. The application utilizes the experience in the memory to learn the experience of old navigation targets, so that forgetting can be reduced to the maximum.

The invention provides a unified framework incremental learning remote sensing scene classification method and system, and belongs to the technical field of image processing. Comprising the steps of unified database construction, unified backbone network, classification header network model construction, unified hyper-parameter construction, linear programming incremental learning supervised training normal form network training, classification prediction result output and unified model verification. According to the method, a unified remote sensing image classification model framework normal form can be constructed, the framework supports embedding of various models, so that the models are accurate, universal and easy to verify, various modal data are supported to serve as model input, and the universality of the framework is remarkably enhanced; and meanwhile, a supervised learning training normal form based on linear programming incremental learning is designed, so that the model can be subjected to continuous fully supervised training on a large-scale heterogeneous data set, and the generalization of the model is enhanced.

The present invention discloses an image classification method based on multi-level adaptive feature fusion incremental learning, including step 1, the first training, obtaining the first image sample set and the first model after training; step 2, when a new type needs to be identified, incremental update is performed, and after the Mth incremental update, the M+1th model and the fusion coefficient of the Mth model and the M+1th model are obtained; M represents the number of incremental updates, which is a natural number greater than or equal to 1; step 3, during testing, the M+1th model and the Mth model are used to perform feature fusion according to the fusion coefficient, and the prediction result is output. This method solves the problem that the current object recognition algorithm cannot continuously learn new categories and new knowledge, and the incremental learning algorithm cannot effectively utilize the information of the old model and forgets the old knowledge to a large extent.

The present invention relates to a low-latency data sharing method for intelligent connected vehicles based on distributed learning, and belongs to the field of mobile communication technology. The method includes obtaining a global model sent from a roadside unit; the global model is obtained by aggregating a student model and a global aggregation weight; collecting vehicle data, and training a student model and a teacher model corresponding to the global model locally; uploading the student model to the roadside unit, and the global aggregation weight determined based on the teacher model; if the degree of deviation between the global model and the student model exceeds a preset threshold, uploading part of the vehicle data to the data buffer of the roadside unit; obtaining shared data sent from the roadside unit, and locally correcting the student model and the teacher model corresponding to the global model; the shared data is determined by the new and old ratios of the uploaded part of the vehicle data and the change rate of important model parameters. In the technical solution provided in the embodiments of the present application, the accuracy of the model can be significantly improved.

The invention discloses a super network-based continuous learning relation extraction method, and relates to the field of natural language processing and machine learning. The model comprises a sample encoder, a network generator based on a super network and a task-specific projection head; the sample encoder is used for acquiring an embedded representation of each instance; the network generator generates a specific projection head for the current task, and the projection head is used for executing the classification of the specific task. To reduce conflicts between similar relationships in a model representation space, a particular projection head is created. For a new task, the embedding of a series of specific tasks is initialized. The embedded input network generators create a dedicated projection head for current task relationship classification. Each task has its specific representation space. And conflicts of similar relationships in a model representation space are effectively reduced. In the model training process, new task training, sample selection and multi-task replay are adopted, and the problem of disastrous forgetting is effectively relieved through strategies such as cross-task loss and comparative learning.

The utility model discloses a USB charger protection box, including first box body, second box body, connecting piece and pendant, first box body and second box body are detachably connected through screw thread structure, and the inner wall of first box body and second box body all is equipped with second protective pad, the outside wall of second box body is fixedly connected with connecting piece, and the both ends of pendant are linked with connecting piece respectively, and the connecting portion of first box body and second box body is equipped with sealing element, and the outer surface of first box body and second box body all is equipped with antiskid line, in the utility model, through adopting first box body and second box body of screw thread connection, cooperate the sealing element of connecting portion, can effectively enhance the sealing property of box body, realize good dustproof waterproof effect, and the second protective pad of box body inner wall, utilize the elastic material buffer impact force, form the all -round anti -drop protection to charger, solved the problem that the existing charger is easy to damage because of collision, dust and water.

The application relates to the technical field of context management, in particular to a method for optimizing a context window of a large model based on cognition. In the application, a PageRank algorithm is used to perform iterative updating on a directional correlation weight and output an entry importance distribution, the ratio of importance to length is entered into sorting and reserved items are selected under capacity constraints, multiple conclusion dependent evidence is entered into a forced reserved path, contradictory evidence is entered into an exclusive selection and replacement numbering registration path, a finite state machine is used to filter windowable entries according to a context stage identifier, and a life cycle marker and version number are linked to exclude expired conflict content, conclusion numbers are aggregated to form aggregated units and an index placeholder is written for excessive units, non-referenced units are executed to reserve index mapping, the probability of forgetting and repeatedly asking in cross-round interaction is reduced, window occupancy fluctuation converges, invalid reasoning overhead and calculation cost are reduced, and the proportion of task target related information is increased, thereby improving output consistency.