H-ensemble: An Information Theoretic Approach to Reliable Few-Shot Multi-Source-Free Transfer
Authors: Yanru Wu, Jianning Wang, Weida Wang, Yang Li
AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our method is empirically validated by ablation studies, along with extensive comparative analysis with other task ensemble and transfer learning methods. We show that the Hensemble can successfully learn the optimal task ensemble, as well as outperform prior arts. |
| Researcher Affiliation | Academia | Yanru Wu1, Jianning Wang2, Weida Wang1, Yang Li1* 1Tsinghua Shenzhen International Graduation School, Tsinghua University 2School of Computer Science and Engineering, Harbin Institute of Technology, Shenzhen |
| Pseudocode | Yes | Algorithm 1: H-ensemble: Training |
| Open Source Code | No | The paper does not include an unambiguous statement of open-source code release or a link to a code repository for the methodology described. |
| Open Datasets | Yes | We conduct extensive experiments on four benchmark datasets. Vis DA-2017 (Peng et al. 2017) is a visual domain transfer challenge dataset [...] Office-31 (Saenko et al. 2010) is a standard transfer learning dataset [...] Office-Caltech (Gong et al. 2012) contains 2,533 images [...] Office-Home (Venkateswara et al. 2017) is a more challenging dataset |
| Dataset Splits | No | Following the standard protocol of few-shot learning, the training data for k-shot is k samples per class randomly selected from the target task. This describes the training data size but not explicit train/validation/test splits of the whole dataset. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running experiments (e.g., GPU/CPU models, memory specifications). |
| Software Dependencies | No | The paper does not provide specific software dependencies with version numbers, such as Python libraries or deep learning frameworks. |
| Experiment Setup | No | The paper states 'Source models are set to Resnet18 with hidden dim 256 trained on full training sets' and mentions 'k samples per class' for few-shot learning. It also lists 'Learning rate λ' as a parameter in Algorithm 1. However, specific values for hyperparameters like learning rate, batch size, or number of epochs are not provided in the main text. |