Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].

Rethinking Teacher-Student Curriculum Learning through the Cooperative Mechanics of Experience

Authors: Manfred Diaz, Liam Paull, Andrea Tacchetti

TMLR 2024 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Through experiments covering supervised learning, reinforcement learning, and classical games, we estimate the cooperative values of experiences and use value-proportional curriculum mechanisms to construct curricula, even in cases where TSCL struggles.
Researcher Affiliation Collaboration Manfred Diaz EMAIL Mila, University of Montreal Liam Paull EMAIL Mila, University of Montreal Canada CIFAR AI Chair Andrea Tacchetti EMAIL Google Deep Mind
Pseudocode Yes Algorithm 1 Generalized Teacher-Student Curriculum Learning.
Open Source Code No Regardless, we plan to release the complete source code of all our experiments.
Open Datasets Yes For instance, in the MNIST dataset (Le Cun & Cortes, 2010)
Dataset Splits No For both MNIST and CIFAR10 (Krizhevsky, 2009), we trained a model on the complete dataset (e.g., for 200 epochs), extracted the confusion matrix on validation, and identify the top-k most confused pairs of classes.
Hardware Specification Yes All models and architectures are implemented with PYTORCH (Paszke et al., 2019), are configured using HYDRA (Yadan, 2019), and fit on a workstation equipped with a 16 GB NVIDIA RTX A4000 GPU, 32 GB of RAM, and 32 CPU cores.
Software Dependencies No All models and architectures are implemented with PYTORCH (Paszke et al., 2019), are configured using HYDRA (Yadan, 2019), and fit on a workstation equipped with a 16 GB NVIDIA RTX A4000 GPU, 32 GB of RAM, and 32 CPU cores.
Experiment Setup Yes Specification of the model architecture and hyper-parameters selection are provided in Table 2.