Prototypical Contrastive Predictive Coding

Authors: Kyungmin Lee

ICLR 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Through extensive experiments, we validate the effectiveness of our method compared to various supervised and self-supervised knowledge distillation baselines.
Researcher Affiliation Academia Kyungmin Lee Agency for Defense Development kyungmnlee@gmail.com
Pseudocode Yes The Py Torch style pseudo-code for our Proto CPC is demonstrated in Algorithm 1.
Open Source Code No The paper does not provide an explicit statement or link for the open-source code of the described methodology.
Open Datasets Yes We experiment on CIFAR-100 (Krizhevsky et al., 2009) and Image Net (Deng et al., 2009)
Dataset Splits Yes Table 3: Top-1 and Top-5 error rates (%) of student network Res Net-18 on Image Net validation set.
Hardware Specification No The paper does not specify the hardware (e.g., GPU models, CPU models, or cloud computing instances) used for running the experiments.
Software Dependencies No The paper mentions 'Py Torch style pseudocode' but does not specify version numbers for PyTorch or any other software dependencies.
Experiment Setup Yes For CIFAR-100, we initialize the learning rate as 0.05, and decay it by 0.1 every 30 epochs after the first 150 epochs until the last 240 epoch. [...] Batch size is 64 for CIFAR-100 or 256 for Image Net. [...] For probability of teacher, we use SK operator with 3 steps of iteration and τt = 0.04. For probability of student, we set τs = 0.1. The prior momentum for Proto CPC loss is 0.9. We use SGD optimizer with batch size 512 and weight decay is 1e-4. The learning rate is 0.6 and is decayed by cosine learning rate schedule to 1e-6.