PulseAugur
LIVE 13:06:31
research · [2 sources] ·
0
research

New HyCNNs architecture offers improved convex function learning and optimal transport

Researchers have developed Hyper Input Convex Neural Networks (HyCNNs), a new architecture designed to learn convex functions more efficiently than existing Input Convex Neural Networks (ICNNs). HyCNNs integrate Maxout networks with ICNN principles, offering theoretical advantages in depth utilization and scalability. Experiments show HyCNNs outperform ICNNs and MLPs in convex regression and interpolation tasks, and are effective in learning high-dimensional optimal transport maps for synthetic data and single-cell RNA sequencing. AI

Summary written by gemini-2.5-flash-lite from 2 sources. How we write summaries →

IMPACT Introduces a more parameter-efficient architecture for convex function learning, potentially improving performance on tasks like optimal transport.

RANK_REASON Academic paper introducing a novel neural network architecture.

Read on arXiv stat.ML →

COVERAGE [2]

  1. arXiv stat.ML TIER_1 · Shayan Hundrieser, Insung Kong, Johannes Schmidt-Hieber ·

    Hyper Input Convex Neural Networks for Shape Constrained Learning and Optimal Transport

    arXiv:2604.26942v1 Announce Type: cross Abstract: We introduce Hyper Input Convex Neural Networks (HyCNNs), a novel neural network architecture designed for learning convex functions. HyCNNs combine the principles of Maxout networks with input convex neural networks (ICNNs) to cr…

  2. arXiv stat.ML TIER_1 · Johannes Schmidt-Hieber ·

    Hyper Input Convex Neural Networks for Shape Constrained Learning and Optimal Transport

    We introduce Hyper Input Convex Neural Networks (HyCNNs), a novel neural network architecture designed for learning convex functions. HyCNNs combine the principles of Maxout networks with input convex neural networks (ICNNs) to create a neural network that is always convex in the…