HAM: A Hyperbolic Step to Regulate Implicit Bias
This work addresses a bottleneck in deep learning optimization for researchers and practitioners, offering an incremental improvement that enhances convergence and sparsity in various applications.
The paper tackles the issue of slow convergence due to hyperbolic implicit bias in optimization algorithms by proposing HAM, which alternates optimizer steps with hyperbolic mirror steps, leading to improved performance across tasks like vision and language model fine-tuning, with minimal overhead.
Understanding the implicit bias of optimization algorithms has become central to explaining the generalization behavior of deep learning models. For instance, the hyperbolic implicit bias induced by the overparameterization $m \odot w$--though effective in promoting sparsity--can result in a small effective learning rate, which slows down convergence. To overcome this obstacle, we propose HAM (Hyperbolic Aware Minimization), which alternates between an optimizer step and a new hyperbolic mirror step. We derive the Riemannian gradient flow for its combination with gradient descent, leading to improved convergence and a similar beneficial hyperbolic geometry as $m \odot w$ for feature learning. We provide an interpretation of the the algorithm by relating it to natural gradient descent, and an exact characterization of its implicit bias for underdetermined linear regression. HAM's implicit bias consistently boosts performance--even of dense training, as we demonstrate in experiments across diverse tasks, including vision, graph and node classification, and large language model fine-tuning. HAM is especially effective in combination with different sparsification methods, improving upon the state of the art. The hyperbolic step requires minimal computational and memory overhead, it succeeds even with small batch sizes, and its implementation integrates smoothly with existing optimizers.