Adaptive Robust Learning using Latent Bernoulli Variables
Authors: Aleksandr Karakulev, Dave Zachariah, Prashant Singh
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate our robust learning method and its parameter-free nature on a wide variety of machine learning tasks including online learning and deep learning where it adapts to different levels of noise and maintains high prediction accuracy. The proposed RLVI method is compared to existing approaches in three problem settings: standard parameter estimation, online learning, and deep learning. |
| Researcher Affiliation | Academia | Aleksandr Karakulev 1 Dave Zachariah 1 Prashant Singh 1 2 1Uppsala University, Sweden 2Science for Life Laboratory, Sweden. |
| Pseudocode | Yes | Algorithm 1 RLVI: robust learning from corrupted data and Algorithm 2 RLVI: robust training of neural networks |
| Open Source Code | Yes | Implementation of RLVI and our experiments are available at https://github.com/akarakulev/rlvi. |
| Open Datasets | Yes | We consider the Human Activity Recognition dataset from (Helou, 2023)... The datasets being used are MNIST (Le Cun, 1998)... CIFAR10 and CIFAR100 (Krizhevsky, 2009)... CIFAR80N-O (Yao et al., 2021)... Food101 (Bossard et al., 2014). |
| Dataset Splits | Yes | For CDR and RLVI, 10% of the training data is used as a validation set: in RLVI, we apply regularization (21) after validation accuracy at the current epoch becomes less than the average of its two previous values. |
| Hardware Specification | No | The computations/data handling were enabled by the Berzelius resource provided by the Knut and Alice Wallenberg Foundation at the National Supercomputer Centre and by the National Academic Infrastructure for Supercomputing in Sweden (NAISS) at Chalmers e-Commons at Chalmers, and Uppsala Multidisciplinary Center for Advanced Computational Science (UPPMAX) at Uppsala University, partially funded by the Swedish Research Council through grant agreement nos. 2022-06725 and 2018-05973. This lists computing resources but does not provide specific hardware details like GPU/CPU models or memory. |
| Software Dependencies | No | The paper provides general optimizer names (SGD, Adam) and mentions frameworks implicitly (e.g., PyTorch for ResNet50) but does not list specific version numbers for software dependencies like Python, PyTorch, or CUDA. |
| Experiment Setup | Yes | Table 4: Hyperparameter settings for the deep learning experiments (LR = learning rate, mom. = momentum) lists details such as Model, Optimizer, Epochs, Batch size, Weight decay, LR schedule, and Initial LR for different datasets. |