Component-Aware Pruning Framework for Neural Network Controllers via Gradient-Based Importance Estimation
This work addresses the problem of efficient model compression for advanced neural controllers, offering a more informed pruning approach, though it appears incremental as it builds on existing gradient-based methods.
The paper tackles the challenge of pruning multi-component neural network controllers by introducing a gradient-based importance estimation framework, which outperforms static heuristics in identifying structural dependencies and dynamic importance shifts, as demonstrated with autoencoder and TD-MPC agent experiments.
The transition from monolithic to multi-component neural architectures in advanced neural network controllers poses substantial challenges due to the high computational complexity of the latter. Conventional model compression techniques for complexity reduction, such as structured pruning based on norm-based metrics to estimate the relative importance of distinct parameter groups, often fail to capture functional significance. This paper introduces a component-aware pruning framework that utilizes gradient information to compute three distinct importance metrics during training: Gradient Accumulation, Fisher Information, and Bayesian Uncertainty. Experimental results with an autoencoder and a TD-MPC agent demonstrate that the proposed framework reveals critical structural dependencies and dynamic shifts in importance that static heuristics often miss, supporting more informed compression decisions.