Research map / Time series / Causal and generative time series
Combining Recurrent, Convolutional, and Continuous-time Models with Linear State Space Layers
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| use structured HiPPO state matrix A instead of random A (ablation) | LSSL · Permuted MNIST | accuracy: 62\% with random A | Helped |
| train the HiPPO-structured A (LSSL vs LSSL-f) (ablation) | LSSL · Speech Commands raw (length 16000) | accuracy: 90.64 -> 95.87 | Helped |
| well-specified (learnable) timescale Delta t instead of a poorly specified one (ablation) | LSSL-f · sequential CIFAR | accuracy: 49.3\% with poorly specified Delta t | Helped |
| change test-time Delta t to adapt to halved sampling rate (LSSL vs CKConv) | LSSL · Speech Commands raw, sampling rate 1 -> 1/2 | accuracy: 88.66 vs 65.96 (CKConv) | Helped |
| train on raw length-16000 signal instead of MFCC features | LSSL · Speech Commands | accuracy: 93.58 -> 95.87 | Helped |
| LSSL-f versus specialized ResNet-18 trained directly on images | LSSL-f · Sequential CelebA | accuracy: 78.89 vs 81.35 (Attractive) | Hurt |
| LSSL-f vs CKConv/UnICORNN on convergence speed | LSSL-fixed · Permuted MNIST | epochs to 98% accuracy: 16 ep. vs 118 ep. (CKConv) | Helped |
Closest papers by meaning
- Kernel-Based Approaches for Sequence Modeling: Connections to Neural Methods (NeurIPS 2019 · similarity 0.90)
- Shallow RNN: Accurate Time-series Classification on Resource Constrained Devices (NeurIPS 2019 · similarity 0.89)
- Recurrent neural networks: vanishing and exploding gradients are not the end of the story (NeurIPS 2024 · similarity 0.88)
- Learning dynamics in linear recurrent neural networks (ICML 2025 · similarity 0.88)
- Understanding the Differences in Foundation Models: Attention, State Space Models, and Recurrent Neural Networks (NeurIPS 2024 · similarity 0.87)
- Recurrent neural network dynamical systems for biological vision (NeurIPS 2024 · similarity 0.87)
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.