Towards Reusable Network Components by Learning Compatible Representations

Authors: Michael Gygli, Jasper Uijlings, Vittorio Ferrari7620-7629

AAAI 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We systematically analyse these approaches on the task of image classification on standard datasets. We demonstrate that we can produce components which are directly compatible without any fine-tuning or compromising accuracy on the original tasks. Afterwards, we demonstrate the use of compatible components on three applications: Unsupervised domain adaptation, transferring classifiers across feature extractors with different architectures, and increasing the computational efficiency of transfer learning.
Researcher Affiliation Industry Michael Gygli, Jasper Uijlings, Vittorio Ferrari Google Research gyglim@google.com, jrru@google.com, vittoferrari@google.com
Pseudocode No The paper describes its methods and training schemes in prose and mathematical formulations but does not include any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not contain any explicit statements about releasing source code or providing links to a code repository for the methodology described.
Open Datasets Yes We systematically analyse these approaches on the task of image classification on standard datasets. ... We train one network on the CIFAR-10 (Krizhevsky 2009) train set and one on the STL10 (Coates, Ng, and Lee 2011) train set. ... It is common to start from a model pre-trained for ILSVRC-12 classification (Donahue et al. 2013; Ren et al. 2015; He et al. 2017).
Dataset Splits No The paper mentions training on 'CIFAR-10 train set' and 'STL10 train set', and evaluating on 'CIFAR-10 test' and 'STL-10 test'. While these are standard datasets with defined train/test splits, the paper does not explicitly provide details about a separate validation split, its size, or how it was used.
Hardware Specification No The paper does not provide specific details about the hardware (e.g., GPU/CPU models, memory) used to run the experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers, such as programming language versions or library versions (e.g., 'Python 3.8, PyTorch 1.9').
Experiment Setup No While the paper discusses network architectures (ResNet-56, WRN-28, MobileNet V2), and mentions 'fine-tuning for 1000 steps' or 'up to 5 epochs' in specific contexts, it lacks comprehensive details on the experimental setup such as specific learning rates, batch sizes, optimizer choices, or full training schedules for the main experiments.