Regularising Knowledge Transfer by Meta Functional Learning
Authors: Pan Li, Yanwei Fu, Shaogang Gong
IJCAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experiments on three Few Shot Learning (FSL) benchmarks (mini Image Net, CIFAR-FS and CUB) show that meta functional learning for regularisation knowledge transfer can benefit improving FSL classifiers. |
| Researcher Affiliation | Collaboration | Pan Li1 , Yanwei Fu2 and Shaogang Gong1 1Queen Mary University of London 2School of Data Science, and MOE Frontiers Center for Brain Science, Fudan University |
| Pseudocode | Yes | Algorithm 1 Meta Functional Learning (MFL). Algorithm 2 MFL with Iterative Updates (MFL-IU). |
| Open Source Code | No | The paper mentions re-running code from prior works ([Chen et al., 2019] and [Wang et al., 2017]) for comparison, but does not provide a statement or link for the open-sourcing of their own method's code. |
| Open Datasets | Yes | Datasets. We employed three FSL datasets: 1) mini Image Net is a subset of the ILSVRC-12 dataset and contains 100 classes with 600 images per class. We followed the split in [Ravi and Larochelle, 2016]... 2) CIFAR-FS is a dataset... Following the split in [Bertinetto et al., 2019]... 3) CUB is a fine-grained dataset... We used... with the previous setting in [Hilliard et al., 2018], and we conducted all experiments with the cropped images provided in [Triantafillou et al., 2017]. |
| Dataset Splits | Yes | mini Image Net... used 64, 16 and 20 classes as base, validation and novel set. CIFAR-FS... used 64 classes to construct the base set, 16 and 20 for validation and novel set. CUB... We used 100, 50 and 50 classes for base, validation and novel set... For training the network, we randomly split the images from base classes into (90%, 10%) partitions as (train, validation) sets. |
| Hardware Specification | No | The paper does not provide specific details about the hardware used for running experiments, such as GPU models, CPU specifications, or cloud computing instance types. |
| Software Dependencies | No | The paper mentions using 'Batch Norm', 'dropout', 'Leaky Re LU', and 'Logistic Regression (LR)' but does not specify version numbers for any software libraries, frameworks, or programming languages used. |
| Experiment Setup | Yes | For training the network, we randomly split the images from base classes into (90%, 10%) partitions as (train, validation) sets. We trained the backbone over 120 epochs. The batch size and learning rate are set as 64 and 0.01. For training MFL/MFL-IU, we employed Batch Norm (0.1), dropout (0.9) and Leaky Re LU (0.01), and the parameters for the first and second fully connected layers are 6000 and 1601 respectively. Moreover, we trained MFL/MFL-IU over 50 epochs with batch size (256) and learning rate (0.01). We adopted the Logistic Regression (LR) function as the base binary classifier and the parameters for computing functional set are Ml = 5, Ms = 100, k = {1, 2, 3, 4} and H = 1e{ 2, 1, 0, 1, 2}. Specifically, we set s = {1, 2, 3, 4, 5} to construct functional tuple sets for s-shot learning scenarios in FSL. |