Dendritic Localized Learning: Toward Biologically Plausible Algorithm
Authors: Changze Lv, Jingwen Xu, Yiyang Lu, Xiaohua Wang, Zhenghua Wang, Zhibo Xu, Di Yu, Xin Du, Xiaoqing Zheng, Xuanjing Huang
ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive empirical experiments demonstrate that DLL satisfies all three criteria of biological plausibility while achieving state-of-the-art performance among algorithms that meet these requirements. Furthermore, DLL exhibits strong generalization across a range of architectures, including MLPs, CNNs, and RNNs. These results, benchmarked against existing biologically plausible learning algorithms, offer valuable empirical insights for future research. We conduct extensive experiments on leveraging the DLL algorithm to train MLPs, CNNs, and RNNs across various tasks, including image recognition, text character prediction, and time-series forecasting, showing comparable performance to backpropagation. |
| Researcher Affiliation | Academia | 1School of Computer Science, Fudan University 2School of Software Technology, Zhejiang University. Correspondence to: Xiaoqing Zheng <EMAIL>. |
| Pseudocode | Yes | A. Global Algorithm of Dendritic Localized Learning In this section, we show the global algorithm of dendritic localized learning in a pseudo-code style. Algorithm 1 Algorithm of Dendritic Localized Learning |
| Open Source Code | Yes | Our code is available at https://github.com/Lvchangze/ Dendritic-Localized-Learning. |
| Open Datasets | Yes | We utilize several widely used benchmark datasets for image recognition tasks: MNIST, Fashion MNIST, SVHN, and CIFAR-10. ... For the text character prediction task, we utilize the Harry Potter Series to evaluate our model s performance.. Harry Potter Series. This dataset includes the entire text of the Harry Potter book series... For the time-series forecasting task, we utilize the Electricity, Metr-la, and Pems-bay. |
| Dataset Splits | Yes | The MNIST dataset consists of... with 60,000 training images and 10,000 test images. ... Fashion MNIST... contains 60,000 training images and 10,000 test images... CIFAR-10 is a dataset comprising... The dataset is split into 50,000 training images and 10,000 test images... |
| Hardware Specification | No | No specific hardware details (such as exact GPU/CPU models, processor types with speeds, or memory amounts) were provided for the experiments. The text only mentions "To fairly compare time consumption across architectures, we used the CPU instead of the GPU." |
| Software Dependencies | No | The paper mentions software components like "Adam optimizer" and "hyperbolic tangent activation function" but does not provide specific version numbers for any programming languages, libraries, or solvers used in the implementation. |
| Experiment Setup | Yes | For the MNIST dataset, we set the learning rate to 1 10 3 and used a batch size of 128. For Fashion MNIST, considering its increased complexity relative to MNIST, we adjusted the learning rate to 5 10 4 and used a batch size of 64. For the SVHN dataset, which presents a higher level of complexity, we set the learning rate to 5 10 5 and maintained a batch size of 64. Finally, for CIFAR-10, the most challenging dataset among the four, we set the learning rate to 8 10 5 and also used a batch size of 64. For consistency across all datasets, we standardized the hyperparameters to a learning rate of 5 10 5 and a batch size of 64. Ultimately, we configured the hidden layer sizes for different datasets as follows: for the Harry Potter Series, the hidden layer size was set to 324; for the Electricity and Metr-la dataset, it was set to 300; and for the Pems-bay dataset, the hidden layer size was set to 384. |