Research map / Architectures and efficiency / RNNs and normalization
Brain-inspired network comparison
Compares artificial networks with brain systems and derives biologically plausible architectures and objectives.
57 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2016–2026) · an approach in RNNs and normalization, keyword cluster “brain · biological · feedback”
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.
11 papers in 2023–24 and 20 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 1.7% to 2.1%.
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 · Optimal Architectures in a Solvable Model of Deep Networks (NeurIPS 2016 · 25 citations)
- 2017 · AdaNet: Adaptive Structural Learning of Artificial Neural Networks (ICML 2017 · 107 citations)
- 2018 · Rapid Adaptation with Conditionally Shifted Neurons (ICML 2018 · 121 citations)
- 2019 · Brain-Like Object Recognition with High-Performing Shallow Recurrent ANNs (NeurIPS 2019 · 107 citations)
- 2020 · Neuron Shapley: Discovering the Responsible Neurons (NeurIPS 2020 · 61 citations)
- 2021 · Understanding Adaptive, Multiscale Temporal Integration In Deep Speech Recognition Systems (NeurIPS 2021 · 10 citations)
- 2022 · How to Train Your Wide Neural Network Without Backprop: An Input-Weight Alignment Perspective (ICML 2022 · 2 citations)
- 2023 · A Unified, Scalable Framework for Neural Population Decoding (NeurIPS 2023 · 22 citations)
- 2024 · Epitopological learning and Cannistraci-Hebb network shape intelligence brain-inspired theory for ultra-sparse advantage in deep learning (ICLR 2024 · 4 citations)
- 2025 · STINR: Deciphering Spatial Transcriptomics via Implicit Neural Representation (CVPR 2025 · 5 citations)
- 2026 · BioToken and BioFM – Biologically-Informed Tokenization Enables Accurate and Efficient Genomic Foundation Models (ICML 2026 · 3 citations)
Earliest papers here
- Optimal Architectures in a Solvable Model of Deep Networks (NeurIPS 2016 · 25 citations)
- Using Fast Weights to Attend to the Recent Past (NeurIPS 2016)
- AdaNet: Adaptive Structural Learning of Artificial Neural Networks (ICML 2017 · 107 citations)
What has been tried
45 papers here report 434 measured changes (a component added, removed or swapped, compared with a baseline): helped 59%, no clear effect 16%, hurt 18%, 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 |
|---|---|---|---|---|
| Use a suitable network configuration such as width 200 and depth 4 | Mouse embryos, sampling rate 0.7 · MSE | 0.223; 0.222; 0.220; 0.221; 0.225; 0.231; 0.234 | No clear effect | STINR: Deciphering Spatial Transcriptomics via Implicit Neural Representation (CVPR 2025 · 5 citations) |
| Use sparse rather than all-to-all lateral connectivity | MNIST, larger hidden layers · test accuracy | 97.5 %; 98.0 % | Helped | What should a neuron aim for? Designing local objective functions based on information theory (ICLR 2025 · 2 citations) |
| Prune core feature channels after training | Sensorium 2022 · ARI of embedding clusters | Helped | Reproducibility of predictive networks for mouse visual cortex (NeurIPS 2024 · 2 citations) | |
| add Oli and Ast as multiple Glia units instead of using one Glia unit | CIFAR10 · accuracy (%) | 96.68 -> 96.90 | Helped | GliaNet: Adaptive Neural Network Structure Learning with Glia-Driven (CVPR 2025 · 1 citation) |
| train the model rather than use random initialization | Le Petit Prince English, 25% most reliable voxels · left-right brain-correlation difference | Helped | fMRI predictors based on language models of increasing complexity recover brain left lateralization (NeurIPS 2024 · 1 citation) | |
| Increase initial weight gain from 0.0 to 1.0 | Sussillo 2015 · Procrustes distance | 0.461~±~0.006 -> 0.398~±~0.007 | Helped | Can Biologically Plausible Temporal Credit Assignment Rules Match BPTT for Neural Similarity? E-prop as an Example (ICML 2025 · 0 citations) |
| Replace path-based CH3-L3p regrowth with node-based CH2-L3n | MNIST, Fashion MNIST, and EMNIST · accuracy | 98.81 ± 0.04* -> 98.76 ± 0.05; 90.93 ± 0.03* -> 90.67 ± 0.05; 87.61 ± 0.07* -> 87.82 ± 0.04* | Mixed | Brain network science modelling of sparse neural networks enables Transformers and LLMs to perform as fully connected (NeurIPS 2025 · 0 citations) |
| Increase the sample size used to estimate JNE | JNE sample-size simulation · Pearson correlation with |sin(x)| | 0.87 -> 1.00 | Helped | Jacobian-Based Interpretation of Nonlinear Neural Encoding Model (NeurIPS 2025 · 0 citations) |
| Use integrated gradients instead of other attribution methods for edge weights | ResNet50 and GoogLeNet, last 10 layers · correlation with loss | Helped | NeurFlow: Interpreting Neural Networks through Neuron Groups and Functional Interactions (ICLR 2025 · 0 citations) | |
| Use the forward-backward intersection criterion instead of denoising all channels | CIFAR-100 · Top-1 accuracy (%) | 65.8 -> 70.5 | Helped | Dual-Perspective Activation: Efficient Channel Denoising via Joint Forward-Backward Criterion for Artificial Neural Networks (NeurIPS 2024 · 0 citations) |
| Use learned GCM structures instead of fixed-density structures | Across reported tasks and resolutions · accuracy | Helped | Beyond Pairwise Connections: Extracting High-Order Functional Brain Network Structures under Global Constraints (NeurIPS 2025) |
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.
Datasets
- CIFAR-10 · 5 of 15 papers
- MNIST · 3 of 15 papers
- Fashion-MNIST · 2 of 15 papers
Baselines
- ResNet · 2 of 15 papers
- ViT · 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 31 papers; from OpenAlex and the papers' affiliations).
- Harvard University · 3 papers
- Stanford University · 3 papers
- Southwest University · 2 papers
- Tsinghua University · 2 papers
- Massachusetts Institute of Technology · 2 papers
- University of California, Berkeley · 2 papers
- EPFL · 2 papers
Use it
Download BibTeX of 39 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
- Hebbian and predictive coding (52 papers)
- Backprop-free credit assignment (44 papers)
Elsewhere in the topic
- Efficient network compression and training (229 papers)
- Theory of depth and width (223 papers)
- Representation geometry and collapse (169 papers)
- Dataset distillation and data-efficient training (158 papers)
- Neural scaling laws (155 papers)
- Distributed LLM inference systems (147 papers)
- Layer-wise analysis of language models (145 papers)
- Dataset condensation and flow estimation (126 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).