Research map / Time series / Causal and generative time series
Seeing the forest and the tree: Building representations of both individual and collective dynamics with transformers
Ran Liu, Mehdi Azabou, Max Dabagia, Jingyun Xiao, Eva L Dyer
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 |
|---|---|---|---|
| separate individual module before population module (EIT vs baseline BM) | Transformer · Synthetic 3-body, decoding interaction | accuracy (%): 19.99 -> 91.70 | Improved |
| separate individual module before population module | Transformer · Synthetic 3-body, decoding all combinations | accuracy (%): 2.03 -> 41.99 | Improved |
| separate individual module before population module | Transformer · Synthetic 3-body, decoding type | accuracy (%): 4.03 -> 37.76 | Improved |
| separate individual module before population module | Transformer · Synthetic 2-body, decoding all combinations | accuracy (%): 87.90 -> 94.72 | Improved |
| add interaction (population) module on top of individual temporal transformer (ablation) | EIT · Mihi-Chewie T=2 M-2 | accuracy (%): 71.71 -> 81.61 | Improved |
| add interaction (population) module on top of individual temporal transformer (ablation) | EIT · Jenkins Maze velocity | R2 x100: 71.18 -> 82.15 | Improved |
| add interaction (population) module on top of individual temporal transformer (ablation) | EIT · Jenkins Maze position | R2 x100: 87.09 -> 94.77 | Improved |
| EIT vs supervised NDT in-domain decoding | EIT · Jenkins Maze position | R2 x100: 94.88 -> 94.77 | No clear effect |
| EIT vs supervised NDT in-domain decoding | EIT · Mihi-Chewie T=6 C-2 | accuracy (%): 94.29 -> 92.59 | Hurt |
| EIT vs NDT-Sup in-domain decoding | EIT · Mihi-Chewie T=2 M-1 | accuracy (%): 83.93 -> 86.51 | Improved |
| frozen EIT with linear readout vs retrained NDT on a new population | EIT · Mihi-Chewie T=2 M-2 (trained on C-2) | accuracy (%): 73.06 -> 75.24 | Improved |
| frozen EIT with linear readout vs retrained NDT on a new population | EIT · Mihi-Chewie T=6 C-1 | accuracy (%): 85.24 -> 81.42 | Hurt |
Closest papers by meaning
- Identifying interactions across brain areas while accounting for individual-neuron dynamics with a Transformer-based variational autoencoder (NeurIPS 2025 · similarity 0.89)
- Learning Interpretable Hierarchical Dynamical Systems Models from Time Series Data (ICLR 2025 · similarity 0.89)
- Population Transformer: Learning Population-level Representations of Neural Activity (ICLR 2025 · similarity 0.88)
- Dynamical modeling of nonlinear latent factors in multiscale neural activity with real-time inference (NeurIPS 2025 · similarity 0.87)
- Meta-Dynamical State Space Models for Integrative Neural Data Analysis (ICLR 2025 · similarity 0.87)
- A scalable generative model for dynamical system reconstruction from neuroimaging data (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.
