Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1]
Aux-Drop: Handling Haphazard Inputs in Online Learning Using Auxiliary Dropouts
Authors: Rohit Agarwal, Deepak Gupta, Alexander Horsch, Dilip K. Prasad
TMLR 2023 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | The efficacy of Aux-Drop has been demonstrated through extensive numerical experiments on SOTA benchmarking datasets that include Italy Power Demand, HIGGS, SUSY and multiple UCI datasets. The code is available at https://github.com/Rohit102497/Aux-Drop. [...] 4 Experiments |
| Researcher Affiliation | Academia | Rohit Agarwal EMAIL Bio-AI Lab, Department of Computer Science Ui T The Arctic University of Norway, Tromsø |
| Pseudocode | Yes | Algorithm 1 Aux-Drop algorithm Require: A deep learning-based online learning model OL, dropout d, z as the Aux Layer |
| Open Source Code | Yes | The code is available at https://github.com/Rohit102497/Aux-Drop. |
| Open Datasets | Yes | We consider the Italy Power Demand dataset1 (Dau et al., 2019), HIGGS2 (Baldi et al., 2014), SUSY3 (Baldi et al., 2014) and 4 different UCI datasets (german4, svmguide35 (Chang & Lin, 2011), magic046, a8a7) (Dua & Graff, 2017). The number of instances and the features of each dataset are listed in Table 2. |
| Dataset Splits | Yes | In all the scenarios, the instances are provided one by one to the model, and the training and testing are performed in a single pass. [...] We considered the first 12 features of the Italy Power Demand dataset as the base features and the last 12 features as the auxiliary features. [...] The trapezoidal streams are simulated by splitting the data into 10 chunks. The number of features in each successive chunk increases with the data stream. The first chunk has the first 10% of the total features, the second chunk has the first 20% features, and so on. [...] For HIGGS and SUSY, we consider the first 5 and 2 features as base features and the next 16 and 6 features as auxiliary features, respectively. |
| Hardware Specification | No | The paper does not explicitly mention specific hardware details such as GPU/CPU models, memory, or type of computing resources used for the experiments. It only generally describes running experiments. |
| Software Dependencies | No | The paper does not provide specific software names with version numbers for libraries, frameworks, or programming languages used (e.g., Python, PyTorch, TensorFlow, CUDA versions). |
| Experiment Setup | Yes | Aux-Drop(ODL) is trained with 11 hidden layers, considering the 3rd hidden layer as Aux Layer. Each layer has 50 nodes and the Aux Layer has 100 nodes. The smoothing rate is set as 0.2, the discount rate is fixed at 0.99 and the dropout is chosen as 0.3. We use cross-entropy loss. The learning rate is 0.3. [...] The dropout value is set as 0.3. The number of nodes in Aux Layer is 100 except for a8a which has 400 nodes in Aux Layer. The smoothing rate for Aux-Drop(ODL) is 0.2 and the discount rate is 0.99. The cross-entropy loss is employed to train the model. The learning rate is 0.1 for the smaller datasets, i.e., german and svmguide3, whereas, for the larger datasets, i.e., magic04 and a8a, it is set as 0.01. |