Research map / Time series / Causal and generative time series
Complex Gated Recurrent Neural Networks
Moritz Wolter, Angela Yao
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add complex gated recurrent cells with unitary state transitions | Human motion prediction · Human3.6M | Euler-angle error: 0.42 -> 0.41 (80ms); 0.74 -> 0.73 (160ms); 1.12 -> 1.12 (320ms); 1.27 -> 1.26 (400ms) | Mixed |
| Replace the full-width GRU with cgRNN using a smaller hidden state | Human motion prediction · Human3.6M | parameters: 3.4M -> 1.8M | Improved |
| Add gating to a complex RNN with unitary transitions (ablation) | cgRNN · Adding problem | task performance | Improved |
| Add gating to a complex RNN with unitary transitions (ablation) | cgRNN · Memory problem | task performance | No clear effect |
| Use unitary state transition matrices instead of unconstrained transitions (ablation) | cgRNN · Memory and adding problems | convergence | Improved |
| Replace modReLU with bounded Hirose non-linearity while removing unitary transitions (ablation) | cgRNN · Memory and adding problems | task performance | Improved |
| Use modReLU instead of Hirose non-linearity with unitary transitions (ablation) | cgRNN · Memory and adding problems | task performance | Improved |
| Replace unitary complex transitions with real orthogonal transitions in the free-real variant (ablation) | cgRNN · Memory problem | task performance | Hurt |
| Replace the modSigmoid gate with alternative tested gate functions (ablation) | cgRNN · Synthetic tasks | convergence | Mixed |
Closest papers by meaning
- Gating Revisited: Deep Multi-layer RNNs That Can Be Trained (TPAMI 2021 · similarity 0.87)
- A recurrent neural network without chaos (ICLR 2017 · similarity 0.86)
- Hierarchically Gated Recurrent Neural Network for Sequence Modeling (NeurIPS 2023 · similarity 0.85)
- Dynamical Isometry and a Mean Field Theory of RNNs: Gating Enables Signal Propagation in Recurrent Neural Networks (ICML 2018 · similarity 0.85)
- Efficient recurrent architectures through activity sparsity and sparse back-propagation through time (ICLR 2023 · similarity 0.85)
- Unconditional stability of a recurrent neural circuit implementing divisive normalization (NeurIPS 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
