Small Sample Algorithm Optimization for Large Language Models

ChatGPT has sparked significant interest as the latest large language model due to its impressive logical coherence and robust interactive capabilities, allowing it to engage in natural conversations and deliver top-notch services. However, traditional self-attention mechanisms struggle with contextual information processing, relying on prior or posterior context. XLN et addresses this limitation by randomizing input sequences, enabling the model to capture contextual information comprehensively. As a result, XLNet, which leverages a Transformer decoder with an efficient global information-capturing attention mechanism, is the preferred model architecture for this purpose. Furthermore, this study has developed a novel incremental data dimensionality reduction module called IDPCA, taking advantage of PCA benefits, to optimize data dimensionality reduction tasks. Meta-learning is also utilized to fine-tune the small sample algorithm optimization. In experiments, the massive data anomaly detection scenario is considered, with Fl-Score as the evaluation metric. The results show that the proposed optimized model has achieved a better performance.