3results about How to "Reduce generalization" patented technology

This invention relates to the field of image processing technology, specifically disclosing a zero-shot text attribute graph learning method based on prototype mutual prompting enhancement. The method calculates instance-aware contrast loss and constrains the similarity of the same node's embedding in two views. It performs PCA dimensionality reduction on the token embedding of the LLM, extracts the first P principal components to form an aligned coordinate system C, maps the components to this coordinate system, and calculates the informative-aware contrast loss. It freezes the pre-trained GNN and the LLM backbone network, extracts node structural embeddings Z from the pre-trained GNN, generates K prototype embeddings through K linear projectors, and injects them as soft prompts into the LLM instructions. The LLM outputs K expert prompts. It calculates the weights of each expert prompt through a router model. It freezes the LLM training of related projectors and routers, freezes the GNN training of the LLM prompt adaptation layer, and iterates and optimizes 1-3 times. This method can cover complex cross-modal information in text and images while reducing the accumulation of zero-shot transfer bias.

The application discloses a targeted adversarial attack enhancement method based on channel feature selection, belongs to the technical field of artificial intelligence security, and comprises the following steps: a heat map is calculated through a Grad-CAM algorithm, and a significant region is obtained; the significant region is supplemented to a significant map; a local image is obtained by randomly cutting the significant map; the local image is scaled to the same size as the original image; the original image and the local image are input into a CNN with the same adversarial perturbation; and a channel feature selection method is applied to optimize the adversarial perturbation. The application can improve the transferability of adversarial samples to a greater extent in the case of targeted attack; the application creatively applies model attention to targeted adversarial attack, so that the perturbation learns how to better transfer the significant features of the original image to the target category. The application focuses on improving the transferability of adversarial samples, thereby improving the success rate of black-box attack.

This invention discloses a multi-model fusion method and system for permanent magnet motor fault diagnosis, belonging to the field of permanent magnet motor fault diagnosis technology. It aims to improve the flexibility and adaptability of models in multi-task learning and hierarchical classification by introducing multiple Softmax-layer CNNs and fusing them with multiple models, while effectively solving the problems of weak noise resistance and insufficient generalization ability of single models. Specifically, the technical solution of this invention first collects the vibration and current signals of the faulty permanent magnet motor, then uses Markov transfer fields to adaptively enhance the time-domain signals into images. Using a multi-Softmax-layer CNN and XGboost as base learners and an IWOA-SVM model as the meta-learner, a high-precision fault diagnosis result is finally obtained. This invention combines the advantages of multiple Softmax layers and multi-model fusion, effectively solving the multi-task fault diagnosis problem under complex working conditions, and improving the accuracy, reliability, and adaptability of fault diagnosis.