Generalizablity of Memorization Neural Network
Authors: Lijia Yu, Xiao-Shan Gao, Lijun Zhang, Yibo Miao
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | I Experiments We try to verify Theorem 7.3 on MNIST and CIFAR10 [33]. For MNIST, we tested all binary classification problems with different label compositions. For each pair of labels, we use 500 corresponding samples with each label in the original dataset to form a new dataset Dtr, and then construct memorization network for Dtr by Theorem 7.3. For each binary classification problem, Table 1 shows the accuracy on the samples with such two labels in testset. |
| Researcher Affiliation | Academia | Lijia Yu1, Xiao-Shan Gao2,3, , Lijun Zhang1,3, Yibo Miao2,3 1 Key Laboratory of System Software (Chinese Academy of Sciences) and State Key Laboratory of Computer Science, Institute of Software, Chinese Academy of Sciences 2Academy of Mathematics and Systems Science, Chinese Academy of Sciences Beijing 100190, China 3University of Chinese Academy of Sciences, Beijing 100049, China |
| Pseudocode | No | Our paper does not contain explicit pseudocode blocks or algorithms labeled as such. The methods are described using mathematical formulations and descriptive text. |
| Open Source Code | Yes | We have provided our codes in the supplemental matrial. |
| Open Datasets | Yes | I Experiments We try to verify Theorem 7.3 on MNIST and CIFAR10 [33]. |
| Dataset Splits | No | Our paper specifies the creation of a dataset `Dtr` for training and evaluation on a `testset`, but does not explicitly mention a separate `validation` split for hyperparameter tuning or early stopping. |
| Hardware Specification | Yes | the device is GPU NVIDIA Ge Force RTX 3090 |
| Software Dependencies | No | Our paper mentions using Resnet18 and crossentropy as the loss function, but does not provide specific version numbers for software libraries or dependencies like PyTorch, TensorFlow, or Python. |
| Experiment Setup | Yes | train Resnet18 [28] on Dtr (with 20 epochs, learning rate 0.1, use crossentropy as loss function, device is GPU NVIDIA Ge Force RTX 3090) |