arXiv:2606. 30376v1 Announce Type: new Abstract: Aligning generative flow models on continuous spaces via online reinforcement learning is constrained by intractable trajectory likelihoods.
By Zheming Fu, Ruizhe He, Wei Shang, Xiaoxiao Ma, Lei Wang, Chang Liu, Siming Fu
arXiv:2607. 27372v1 Announce Type: new Abstract: The deep learning revolution, kicked off by AlexNet, taught us that end-to-end training beats decomposing a problem into hand-designed stages.
By Alexi Gladstone, Heng Ji, Yilun Du
arXiv:2608.24855v1 Announce Type: new
Abstract: Latent world models are inherently strong encoders that transform image pixel to latent embedding, yet existing world models still rely on online traje...
By Hsiang-Wei Huang, Jianxu Shangguan, Junbin Lu, Jenq-Neng Hwang
The paper introduces Bridge Graphical Models (BGMs), a framework that decomposes continuous‑time generative models into independent design choices: endpoint coupling, bridge law, Markovian projection, and current‑preserving dynamics. It defines the Markovization gap as the time‑integrated conditional variance of bridge velocity given the Markov state, quantifying an irreducible loss before training. Experiments on synthetic, latent, and pixel‑space tasks (CIFAR‑10 and Fashion‑MNIST) show that a feature‑space proxy of this gap, estimated quickly before training, predicts downstream training loss and FID in the same direction as full training results.
By Tiantian Zhang
arXiv:2605. 29920v2 Announce Type: replace Abstract: We introduce Midpoint Generative Models (MGM), a principled framework for training one-step generative models.
By Daniil Shlenskii, Nikita Gushchin, Lev Novitskiy, Dmitry V. Dylov, Alexander Korotin
Generative models have undergone many generations of evolution, from VAEs/GANs to diffusion/flow matching. Along the way, the underlying techniques have become more complicated and various beliefs about what drives strong empirical performance have taken hold.