Generative Flow Networks or GFlowNets is a new paradigm of neural net training, developed at MILA since 2021.
GFlowNets are related to Monte-Carlo Markov chain methods (as they sample from a distribution specified by an energy function), reinforcement learning (as they learn a policy to sample composed objects through a sequence of steps), generative models (as they learn to represent and sample from a distribution) and amortized variational methods (as they can be used to learn to approximate and sample from an otherwise intractable posterior, given a prior and a likelihood). GFlowNet are trained to generate an object through a sequence of steps with probability proportional to some reward function (or with denoting the energy function), given at the end of the generative trajectory.[1]
Through generative models and variational inference, GFlowNets are also related to Active Inference.
GFlowNets promise better interpretability and more robust reasoning than the current auto-regressive LLMs[2].
Pan, L., Malkin, N., Zhang, D., & Bengio, Y. (2023). Better Training of GFlowNets with Local Credit and Incomplete Trajectories (arXiv:2302.01687). arXiv. https://doi.org/10.48550/arXiv.2302.01687
Bengio, Y., & Hu, E. (2023, March 21). Scaling in the service of reasoning & model-based ML. Yoshua Bengio. https://yoshuabengio.org/2023/03/21/scaling-in-the-service-of-reasoning-model-based-ml/