Optimal Brain Apoptosis
Mingyuan Sun, Zheng Fang, Jiaxu Wang, Junjie Jiang, Delei Kong, Chenming Hu, Yuetong Fang, Renjing Xu
2025-03-03
Summary
This paper talks about a new method called Optimal Brain Apoptosis (OBA) for making artificial intelligence systems more efficient by removing unnecessary parts without losing much performance.
What's the problem?
As AI systems like Convolutional Neural Networks (CNNs) and Transformers get bigger and more complex, they need more computing power and resources to run. This makes them slower and more expensive to use.
What's the solution?
The researchers created OBA, which is a smarter way to 'prune' or remove unnecessary parts of these AI systems. OBA uses advanced math (the Hessian matrix) to figure out which parts of the AI are most important. Unlike older methods, OBA calculates this importance directly for each part of the AI, making it more accurate. They also found a way to break down this complex math problem to make it easier to solve.
Why it matters?
This matters because it could make AI systems run faster and use less energy without becoming less accurate. This could lead to more efficient AI that can be used in more places, like on smartphones or in applications where speed is crucial. The researchers tested their method on several well-known AI models and different datasets, showing that it works well in real-world situations.
Abstract
The increasing complexity and parameter count of Convolutional Neural Networks (CNNs) and Transformers pose challenges in terms of computational efficiency and resource demands. Pruning has been identified as an effective strategy to address these challenges by removing redundant elements such as neurons, channels, or connections, thereby enhancing computational efficiency without heavily compromising performance. This paper builds on the foundational work of Optimal Brain Damage (OBD) by advancing the methodology of parameter importance estimation using the Hessian matrix. Unlike previous approaches that rely on approximations, we introduce Optimal Brain Apoptosis (OBA), a novel pruning method that calculates the Hessian-vector product value directly for each parameter. By decomposing the Hessian matrix across network layers and identifying conditions under which inter-layer Hessian submatrices are non-zero, we propose a highly efficient technique for computing the second-order Taylor expansion of parameters. This approach allows for a more precise pruning process, particularly in the context of CNNs and Transformers, as validated in our experiments including VGG19, ResNet32, ResNet50, and ViT-B/16 on CIFAR10, <PRE_TAG>CIFAR100</POST_TAG> and Imagenet datasets. Our code is available at https://github.com/NEU-REAL/OBA.