Reproduce “Large Distant Gradients Need Not Be Reliable: reliability-weighted credit assignment for long-horizon autoregressive forecasting” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Junhao Zhao, David Michael Simberg, Jacob Kang, Colin Connor Kurniawan, Nan Xu
Internal-DW reweights gradient routes inside residual blocks during forecasting, while preserving forward dynamics and losses at every rollout horizon. It calibrates gains for identity and nonlinear routes using calibration probes. In a controlled system, it lowers route risk versus full BPTT. Forecast errors fall on four history-dominated, weak-drive testbeds, but results can be neutral or worse elsewhere.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.