Optimization of Artificial Intelligence Inference Engine in Computer Networks Based on Big Data Analysis

Abstract

To address the issue of the degradation of the performance of artificial intelligence inference engines under dynamic constraints of computer networks, starting from the coupling mechanism of computation and communication, a joint optimization model for inference based on network state perception was constructed. By modeling network delay and bandwidth fluctuations, and introducing operator-level scheduling and adaptive execution mechanisms, the dynamic regulation of inference execution paths was achieved. Experimental results show that under dynamic network enhancement conditions, the proposed method can control the average communication delay within 263 ms, reducing path jitter by approximately $40 %$ compared to the baseline strategy, effectively alleviating the amplification effect of network uncertainty on the stability of inference.