Model-Aware Contrastive Learning: Towards Escaping the Dilemmas

Authors: Zizheng Huang, Haoxing Chen, Ziqi Wen, Chao Zhang, Huaxiong Li, Bo Wang, Chunlin Chen

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive remarkable empirical results in vision, sentence, and graph modality validate our approach s general improvement for representation learning and downstream tasks.
Researcher Affiliation Collaboration 1Nanjing University 2Ant Group 3China Telecom.
Pseudocode Yes A simple example pseudocode of Eqn.(11) is shown as Algorithm 1.
Open Source Code No The paper states 'Codes of models are implemented on mmselfsup (Contributors, 2021)' which refers to a third-party toolbox, but does not provide an explicit statement or link for the open-source code of their own described methodology (MACL).
Open Datasets Yes We mainly experiment on the Image Net ILSVRC-2012 (i.e., Image Net-1K) (Deng et al., 2009) and use standard Res Net-50 (He et al., 2016) as image encoders. CIFAR10 (Krizhevsky et al., 2009) and the subset Image Net-100 (Tian et al., 2020a) are also considered. ... We conduct experiments on six commonly used benchmarks (Morris et al., 2020).
Dataset Splits Yes We mainly experiment on the Image Net ILSVRC-2012 (i.e., Image Net-1K) (Deng et al., 2009) and use standard Res Net-50 (He et al., 2016) as image encoders. CIFAR10 (Krizhevsky et al., 2009) and the subset Image Net-100 (Tian et al., 2020a) are also considered. ... For linear evaluation, the trained CL models are evaluated by fine-tuning a linear classifier for 100 epochs with 128-batch size on top of frozen backbones.
Hardware Specification Yes Codes of models are implemented on mmselfsup (Contributors, 2021) with several Tesla A100 80G GPUs.
Software Dependencies Yes algorithms are performed based on Huggingface s transformers package1 and evaluated with Sent Eval toolkit2. ... 1https://github.com/huggingface/transformers,version 4.2.1.
Experiment Setup Yes Models optimizations are completed by LARS with a base learning rate of 0.3 (0.3 Batch Size/256) and weight decay of 1e-6. We also use the cosine decay learning rate schedule with 10 epochs warmup. Parameters {τ0, α, A0} are set to {0.1, 0.5, 0}.