Research map / Probabilistic and scientific modeling / Dynamical systems
Recurrent circuit models of inference
Explains Bayesian inference and learning with recurrent circuits and synaptic mechanisms.
70 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in Dynamical systems, keyword cluster “brain · activity · neurons”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
19 papers in 2023–24 and 21 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic shrank, from 8.4% to 5.5%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2016 · A Bayesian method for reducing bias in neural representational similarity analysis (NeurIPS 2016 · 32 citations)
- 2017 · InfoGAIL: Interpretable Imitation Learning from Visual Demonstrations (NeurIPS 2017 · 136 citations)
- 2018 · Manifold-tiling Localized Receptive Fields are Optimal in Similarity-preserving Neural Networks (NeurIPS 2018 · 39 citations)
- 2019 · Better Transfer Learning with Inferred Successor Maps (NeurIPS 2019 · 20 citations)
- 2020 · Reverse-engineering recurrent neural network solutions to a hierarchical inference task for mice (NeurIPS 2020 · 45 citations)
- 2021 · Latent Equilibrium: A unified learning theory for arbitrarily fast computation with arbitrarily slow neurons (NeurIPS 2021 · 10 citations)
- 2022 · Distinguishing discrete and continuous behavioral variability using warped autoregressive HMMs (NeurIPS 2022 · 10 citations)
- 2023 · Disentanglement with Biological Constraints: A Theory of Functional Cell Types (ICLR 2023 · 13 citations)
- 2024 · Model Based Inference of Synaptic Plasticity Rules (NeurIPS 2024 · 2 citations)
- 2025 · From Synapses to Dynamics: Obtaining Function from Structure in a Connectome Constrained Model of the Head Direction Circuit (NeurIPS 2025 · 2 citations)
- 2026 · Structure Abstraction and Generalization in a Hippocampal-Entorhinal Inspired World Model (ICML 2026 · 0 citations)
Earliest papers here
- A Bayesian method for reducing bias in neural representational similarity analysis (NeurIPS 2016 · 32 citations)
- Neurons Equipped with Intrinsic Plasticity Learn Stimulus Intensity Statistics (NeurIPS 2016 · 3 citations)
- InfoGAIL: Interpretable Imitation Learning from Visual Demonstrations (NeurIPS 2017 · 136 citations)
What has been tried
49 papers here report 365 measured changes (a component added, removed or swapped, compared with a baseline): helped 58%, no clear effect 19%, hurt 17%, mixed 6%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Add a synaptic weight-dependent term (wij) to the fly plasticity rule | Drosophila two-alternative choice task · percent deviance explained | p = 5×10−5 | Helped | Model Based Inference of Synaptic Plasticity Rules (NeurIPS 2024 · 2 citations) |
| use a low-rank AR-k model with rank r=35 instead of a full-rank AR-k model | Mouse motor cortex photostimulation datasets · prediction performance | No clear effect | Active learning of neural population dynamics using two-photon holographic optogenetics (NeurIPS 2024 · 1 citation) | |
| use math-agent features rather than one-hot encoded raw math numbers | Math reasoning · MAPE | 0.3813 -> 0.3652 | Helped | CogReact: A Reinforced Framework to Model Human Cognitive Reaction Modulated by Dynamic Intervention (ICML 2025 · 0 citations) |
| Use sparse connectivity rather than random dense connectivity | Sister-cell response similarity · Pearson correlation | Helped | Inference with correlated priors using sisters cells (NeurIPS 2025 · 0 citations) | |
| replace population-conditioned MLP forecasting with direct POYO forecasting | Zebrafish (Deisseroth), 512 PCs · prediction score | -0.971 -> 0.525 | Helped | POCO: Scalable Neural Forecasting through Population Conditioning (NeurIPS 2025 · 0 citations) |
| Train on biased behavioral trajectories | Spatial position estimation · active–quiescent decoded-output distribution similarity | Helped | Sufficient conditions for offline reactivation in recurrent neural networks (ICLR 2024 · 0 citations) | |
| Vary dendritic tree depth and width while keeping parameter counts roughly constant | Swiss-roll manifold · Learning speed | Mixed | Complex priors and flexible inference in recurrent circuits with dendritic nonlinearities (ICLR 2024 · 0 citations) | |
| Vary the number of tensor components, selecting R=15 | Tensor reconstruction · reconstruction-error variability | Mixed | Learning dynamic representations of the functional connectome in neurobiological networks (ICLR 2024 · 0 citations) | |
| Add tuning-dependent SOM inhibition to E neurons | Stimulus posterior · sampling speed | Helped | The Bayesian sampling in a canonical recurrent circuit with a diversity of inhibitory interneurons (NeurIPS 2024 · 0 citations) | |
| Increase interneurons from K=2 to K=3 | Kodak filter-response pairs, spatial offsets less than d=32 · mutual information | Helped | Shaping the distribution of neural responses with interneurons in a recurrent circuit model (NeurIPS 2024 · 0 citations) | |
| Replace SVD with K-means for selecting initial motion parameters | EED parameter initialization · similarity of selected parameter points | No clear effect | Continuous Spatiotemporal Events Decoupling through Spike-based Bayesian Computation (NeurIPS 2024 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 15 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Baselines
- MLP · 2 of 15 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 40 papers; from OpenAlex and the papers' affiliations).
- Stanford University · 4 papers
- University College London · 3 papers
- New York University · 3 papers
- Massachusetts Institute of Technology · 3 papers
- Harvard University · 2 papers
- Allen Institute · 2 papers
- University of Washington · 2 papers
- University of Texas Southwestern Medical Center · 2 papers
- Flatiron Institute · 2 papers
- Peking University · 2 papers
Use it
Download BibTeX of 40 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Latent models of neural activity (70 papers)
- Low-rank and diffusion neural dynamics (53 papers)
- Functional connectivity modeling (50 papers)
- Synaptic plasticity learning rules (72 papers)
Elsewhere in the topic
- Physics-informed dynamics learning (127 papers)
- Latent dynamics models (111 papers)
- State-space inference for stochastic systems (104 papers)
- Switching and law-discovery dynamics (81 papers)
- Latent-space control and planning (78 papers)
- Data assimilation from sparse observations (58 papers)
- Stochastic population dynamics (53 papers)
- Stable contractive dynamics (42 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).