MLAAN: Scaling Supervised Local Learning with Multilaminar Leap Augmented Auxiliary Network

Authors: Yuming Zhang, Shouxin Zhang, Peizhe Wang, Feiyu Zhu, Dongzhi Guan, Junhao Su, Jiabin Liu, Changpeng Cai

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our experiments on the CIFAR-10, STL10, SVHN, and Image Net datasets show that MLAAN can be seamlessly integrated into existing local learning frameworks, significantly enhancing their performance and even surpassing end-to-end (E2E) training methods, while also reducing GPU memory consumption. We conduct experiments with Res Net (He et al. 2016) architectures of various depths on CIFAR-10 (Krizhevsky, Hinton et al. 2009), SVHN (Netzer et al. 2011), STL-10 (Coates, Ng, and Lee 2011), and Image Net (Deng et al. 2009). MLAAN is integrated with DGL (Belilovsky, Eickenberg, and Oyallon 2019) and Info Pro (Wang et al. 2021b) to assess performance and compare with BP (Rumelhart et al. 1985) and original supervised local learning methods.
Researcher Affiliation Academia 1Southeast University, Nanjing 211189, China 2University of Shanghai for Science and Technology, Shanghai 200093, China
Pseudocode No The paper describes methods using mathematical equations and architectural diagrams but does not contain a clearly labeled 'Pseudocode' or 'Algorithm' block with structured steps.
Open Source Code Yes Code https://github.com/Wendy231015/MLAAN
Open Datasets Yes Our experiments on the CIFAR-10, STL10, SVHN, and Image Net datasets show that MLAAN can be seamlessly integrated into existing local learning frameworks, significantly enhancing their performance and even surpassing end-to-end (E2E) training methods, while also reducing GPU memory consumption. The efficacy of MLAAN is validated on a suite of benchmark image classification datasets, including CIFAR-10 (Krizhevsky, Hinton et al. 2009), STL-10 (Coates, Ng, and Lee 2011), SVHN (Netzer et al. 2011), and Image Net (Deng et al. 2009).
Dataset Splits No The paper mentions using well-known datasets like CIFAR-10, STL-10, SVHN, and ImageNet but does not explicitly state the train/test/validation split percentages, sample counts, or refer to a specific predefined standard split for these datasets within the text.
Hardware Specification No The paper discusses 'GPU Memory' in Table 3 and mentions 'reducing GPU memory consumption' but does not specify the particular models of GPUs or other hardware components used for the experiments (e.g., NVIDIA A100, Intel CPU model).
Software Dependencies No The paper does not provide specific version numbers for any software libraries, frameworks (like PyTorch or TensorFlow), or programming languages used in the implementation.
Experiment Setup Yes Experimental Setup: Performance is compared under consistent K conditions, typically K = 3, ensuring rigorous and fair evaluation. Comparative Analysis of Unfair Epochs: DGL(Epochs=400), DGL*(Epochs=300).