Pre-Training and Fine-Tuning Generative Flow Networks
Authors: Ling Pan, Moksh Jain, Kanika Madan, Yoshua Bengio
ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Extensive experimental results validate the efficacy of our approach, demonstrating the effectiveness of pre-training the OC-GFN, and its ability to swiftly adapt to downstream tasks and discover modes more efficiently. |
| Researcher Affiliation | Academia | 1Hong Kong University of Science and Technology 2Mila Qu ebec AI Institute 3Universit e de Montr eal 4CIFAR AI Chair |
| Pseudocode | Yes | Algorithm 1 Reward-free Pre-training of Unsupervised GFlow Nets. Algorithm 2 Supervised Fine-Tuning of Outcome-Conditioned GFlow Nets. |
| Open Source Code | No | All baseline methods are implemented based on the open-source implementation,1 where we follow the default hyperparameters and setup as in (Bengio et al., 2021). The code will be released upon publication of the paper. |
| Open Datasets | Yes | We employ the same standard reward function for Grid World from (Bengio et al., 2021)... We study the bit sequence generation task (Malkin et al., 2022)... For the TFBind-8 generation task, we follow the same setup as in (Jain et al., 2022)... We consider 30 different downstream tasks studied in (Barrera et al., 2016)... We now study a larger task of generating RNA sequences that bind to a given target introduced in (Lorenz et al., 2011). |
| Dataset Splits | No | The paper does not explicitly provide training, validation, or test dataset splits. The experimental setup describes generative tasks where models learn to sample from distributions rather than being trained on fixed dataset splits in the traditional supervised learning sense. |
| Hardware Specification | No | The paper mentions "GPU memory (MiB)" in Figure 18 comparing computation costs, but it does not specify any particular GPU models, CPU models, or other detailed hardware specifications used for running the experiments. |
| Software Dependencies | No | The paper mentions using the "Adam (Kingma & Ba, 2015) optimizer" but does not provide specific software dependencies with version numbers (e.g., Python, PyTorch, CUDA versions). |
| Experiment Setup | Yes | A detailed description of the setup and hyperparameters can be found in Appendix C.1. The GFlow Net model is a feedforward network consisting of two hidden layers with 256 hidden units per layer using Leaky Re LU activation. We train all models with the Adam (Kingma & Ba, 2015) optimizer (learning rate is 0.001) based on samples from a parallel of 16 rollouts in the environment. |