Cyclic Sparse Training: Is it Enough?
This addresses the computational inefficiency of iterative pruning for sparse networks, offering a more efficient method for researchers and practitioners in deep learning.
The paper tackles the problem of achieving high-performance sparse neural networks by challenging the hypothesis that iterative pruning's success is due to mask identification and regularization, instead attributing it to cyclic training schedules. It shows that pruning at initialization with cyclic training outperforms standard iterative pruning, and proposes SCULPT-ing to match state-of-the-art methods at high sparsity with reduced computational cost.
The success of iterative pruning methods in achieving state-of-the-art sparse networks has largely been attributed to improved mask identification and an implicit regularization induced by pruning. We challenge this hypothesis and instead posit that their repeated cyclic training schedules enable improved optimization. To verify this, we show that pruning at initialization is significantly boosted by repeated cyclic training, even outperforming standard iterative pruning methods. The dominant mechanism how this is achieved, as we conjecture, can be attributed to a better exploration of the loss landscape leading to a lower training loss. However, at high sparsity, repeated cyclic training alone is not enough for competitive performance. A strong coupling between learnt parameter initialization and mask seems to be required. Standard methods obtain this coupling via expensive pruning-training iterations, starting from a dense network. To achieve this with sparse training instead, we propose SCULPT-ing, i.e., repeated cyclic training of any sparse mask followed by a single pruning step to couple the parameters and the mask, which is able to match the performance of state-of-the-art iterative pruning methods in the high sparsity regime at reduced computational cost.