Combinatorial Routing for Neural Trees
Authors: Jiahao Li, Ruichu Cai, Yuguang Yan
IJCAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We compare Comb Ro with the existing algorithms on 3 public image datasets, demonstrating its superior performance in terms of accuracy. Visualization results further validate the effectiveness of the multicast routing schema. |
| Researcher Affiliation | Academia | Jiahao Li , Ruichu Cai , Yuguang Yan School of Computer Science, Guangdong University of Technology, Guangzhou, China {jiahaoli.gdut, cairuichu}@gmail.com, ygyan@gdut.edu.cn |
| Pseudocode | Yes | Now, with the conception laid before, we can design our algorithm as follows: 1. To begin with, all the predecessors of L are deduplicated as V(L ) = {v(l) = Null | l L }, in which v(l) is the predecessor of l... 4. Finally, ϱ is equal to ϱ(L ) = {(v, ϱ(v)) | v T \ L } and denotes the probability expression of T . |
| Open Source Code | Yes | Code is available at https://github.com/Jiahao Li-gdut/Comb Ro. |
| Open Datasets | Yes | To evaluate the effectiveness of Comb Ro, three classification benchmarking datasets, including CIFAR10 [Krizhevsky et al., 2009], CIFAR100 [Krizhevsky et al., 2009] and tiny Image Net [Le and Yang, 2015], are adopted. |
| Dataset Splits | No | The paper mentions training on 'training sets' but does not explicitly provide details about training/validation/test dataset splits, specific percentages, or counts for validation. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU/CPU models or processor types used for running the experiments. |
| Software Dependencies | No | The paper mentions optimizers (SGD) and techniques (Gumbel softmax) but does not provide specific version numbers for any software dependencies or libraries. |
| Experiment Setup | Yes | We train Comb Ro using SGD optimizer with the initial learning rate of 0.1. After 30 epochs, the learning rate is decayed by half every 20 epochs until the training stops at the 100th epoch. In addition, we set the batch size to 128, the weight decay to 10 4, the Nesterov momentum to 0.9, and the Gumbel softmax temperature τ to 0.5. The hyperparameter γ for regularization term is set to 0 for the first 80 iterations, and then set to 0.1 for the subsequent 20 iterations. |