Powering Finetuning in Few-Shot Learning: Domain-Agnostic Bias Reduction with Selected Sampling
Authors: Ran Tao, Han Zhang, Yutong Zheng, Marios Savvides8467-8475
AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | In this section, we first conduct comprehensive ablation experiments to verify the effectiveness of both DCM and SS and analyze how our method boosts performance under shot analysis. Then we compare our results with the other latest techniques. We follow the same setting and evaluation metrics in meta-Baseline (Chen et al. 2020). |
| Researcher Affiliation | Academia | Carnegie Mellon University taoran1@cmu.edu, Hanz3, yutongzh, marioss@andrew.cmu.edu |
| Pseudocode | No | The paper does not contain structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any concrete access information or explicit statements about the availability of source code. |
| Open Datasets | Yes | We evaluate our method on Meta-Dataset(Triantafillou et al. 2019), which is so far the most comprehensive benchmark for few-shot learning composed of multiple existing datasets in different domains. ... The ILSVRC-2012 (Russakovsky et al. 2015) in Meta Dataset is splitted into 712 training, 158 validation and 130 test classes. |
| Dataset Splits | Yes | The ILSVRC-2012 (Russakovsky et al. 2015) in Meta Dataset is splitted into 712 training, 158 validation and 130 test classes. |
| Hardware Specification | No | The paper mentions using Res Net18 and Res Net34 backbones but does not specify the hardware (e.g., GPU/CPU models, memory) used for running the experiments. |
| Software Dependencies | No | The paper mentions optimizers (SGD, Adam) and loss functions (softmax cross-entropy) but does not provide specific version numbers for any software libraries or dependencies. |
| Experiment Setup | Yes | The initial learning rate is set to 0.1 with 0.0001 weight decay and decreases by a factor of 0.1 every 30 epochs with total 90 epochs. Both models are trained using the SGD optimizer with batch size 256. ... learning rate of 0.00005, Adam optimizer and 25 total epochs. σ in the proposal distribution for sampling is set to 0.1. |