Open neural data. Build brain-machine intelligence.
SiClink builds infrastructure for invasive brain-computer interfaces, with a focus on high-throughput neural data, neural I/O, and decoding systems that connect cortical signals with silicon intelligence.
Overview · Stack · First Release · Roadmap · Contributing
SiClink is a BCI infrastructure company building the information I/O layer for invasive neural interfaces. Our open-source work starts with visual decoding: neural-to-image retrieval from invasive mouse visual cortex recordings.
We are gradually releasing tools, benchmarks, baselines, and evaluation workflows that make invasive neural data easier to process, compare, and build on.
| 🧬 Neural data I/O | 👁️ Visual decoding | 🧪 Open benchmarks |
|---|---|---|
| Structured pipelines for invasive neural recordings and experiment data flow. | Retrieval, matching, and future reconstruction from cortical visual signals. | Reproducible splits, metrics, baselines, leaderboards, and evaluation workflows. |
| Layer | What it enables | Direction |
|---|---|---|
| 🧠 Invasive neural interface | High-bandwidth access to cortical signals | Record |
| 🔌 Neural data I/O | Structured ingestion, preprocessing, and experiment data flow | Stream |
| 🧬 Neural representation | Feature learning and signal representations for downstream tasks | Learn |
| 📊 Decoding benchmarks | Standardized tasks, splits, metrics, and reproducible evaluation | Compare |
| 🖼️ Visual reconstruction | Retrieval, image matching, and future reconstruction workflows | Decode |
| ⚡ Closed-loop validation | Latency-aware systems for online neural decoding research | Validate |
| Repository | Focus | Status |
|---|---|---|
| SiClink/neural-to-image-retrieval | Neural-to-image retrieval for visual decoding from invasive mouse visual cortex recordings. |  |
This repository is the first public entry point in SiClink's visual decoding roadmap. More modules will follow as the stack matures.
| Release | Scope | Status |
|---|---|---|
| Visual Stimulus Retrieval from Neural Signals | Match invasive mouse visual cortex recordings to candidate visual stimuli. | |
| Spike Sorting and Online Sorting Research Preview | Experimental sorting pipelines for spikes, MUA, and feature confidence estimates from broadband invasive recordings. |  |
| Mouse Visual Decoding Benchmark, MouseVDB v1 | Offline benchmark with standardized splits, metrics, and evaluation scripts. | |
| MouseVDB Leaderboard Beta | Reference evaluation workflow and beta leaderboard for comparing visual decoding methods. | |
| Non-Human Primate Offline Visual Decoding Benchmark v1 | Offline visual decoding benchmark for invasive non-human primate recordings. |  |
| Mouse Visual Reconstruction Benchmark v1 | Benchmark for image and video reconstruction from mouse visual cortex recordings. | |
| Mouse Real-Time Visual Decoding Benchmark v1 | Latency-aware benchmark for online decoding from streaming mouse recordings. | |
| Non-Human Primate Visual Reconstruction Benchmark v1 | Benchmark for image and video reconstruction from invasive non-human primate recordings. | |
| Non-Human Primate Real-Time Visual Decoding Benchmark v1 | Latency-aware benchmark for online decoding from streaming non-human primate recordings. | |
| 🧠 More on the Way 💡 | New benchmarks, datasets, and tools coming. Stay tuned for the next release. 🚀 |  |
We welcome developers and researchers working on invasive neural data, visual decoding, retrieval systems, neural representation learning, and BCI infrastructure.
The best starting point is SiClink/neural-to-image-retrieval. Current areas where contributions are especially useful:
- Reproducing neural-to-image retrieval experiments and sharing reproducibility notes.
- Improving data loading, preprocessing, and benchmark scripts for public datasets.
- Adding baselines for neural representation learning, retrieval, and evaluation.
- Improving documentation, tutorials, examples, and experiment notes.
- Discussing benchmark design, leaderboard tools, train/test splits, and future dataset releases through issues.
For algorithm collaboration, dataset inquiries, or benchmark discussions, contact us at algorithms@siclink.com.