Micro model is our own naming convention for micro models that are re learned using live videos. Conduct research using correct, negative, and uncertain samples. Improve the accuracy of AI vision.
In the field of AI, "micromodels" is a gradually mentioned concept, mainly in contrast to traditional macro models and small models. Micro models typically refer to small machine learning models focused on specific tasks, which specialize in processing specific types of data or specific problems, such as part of speech recognition of individual words in text, recognition of specific entities, etc. They require less data compared to large models and can provide higher accuracy within a narrower range.
| Level | Parameter Magnitude | Hardware Deployment | Functional Positioning | Core Values |
|---|---|---|---|---|
| MacroModels | >10B | Cloud Computing Cluster | General Ability | Acquiring general cognitive abilities |
| Small Models | 100M–10B | Single Server/Workstation | Vertical Scene Optimization | High precision vertical scene+internalization of domain knowledge |
| MicroModels | <100M | Embedded devices | Single Task Reasoning | Ultra low latency+hardware level optimization |
In some practical applications, such as medical image processing and annotation, smart city video analysis, etc., micro models have been used to improve efficiency and reduce development costs. This method also emphasizes the modularity and combinatorial ability of the model, completing complex tasks through the collaboration of multiple micro models rather than relying on a single large model.
If you are interested in this direction, you can explore using micro models to replace or supplement traditional deep learning methods in specific scenarios.
| We are applying for a micro model patent. Please refer to: | 《An efficient micro model generation method and system based on on-site video data》 |
| 《An adaptive micro model and dynamic optimization method for closed-loop feedback of false positive data》 |
