Awesome Invasive BCI Decoding

A curated list of papers, datasets, benchmarks, and tools for invasive brain–computer interface (iBCI) neural decoding — intracortical microelectrode arrays, ECoG, stereo-EEG, and high-density silicon probes.

Invasive BCIs record neural activity directly from the cortex (or deeper structures) and decode it into control signals, text, speech, or movement. This list focuses on the decoding side: the algorithms, models, datasets, and systems that turn spikes and field potentials into intended actions. Non-invasive modalities (EEG / MEG / fNIRS) are intentionally out of scope here — see Non-Invasive (coming soon).

Legend: 📄 paper · 💻 code · 📊 dataset · 🏆 benchmark · 📚 survey

⚠️ Maintenance note: every link is meant to point at a real paper / repo. If you find a dead link or a paper that belongs in a different bucket, please open an issue or PR — see Contributing.

---

Contents

• How This List Is Organized
• Surveys & Reviews
• By Acquisition Modality
  • Intracortical Microelectrode Arrays
  • ECoG (Electrocorticography)
  • Stereo-EEG / Depth Electrodes
  • Neuropixels / High-Density Probes
• By Task
  • Motor Decoding
  • Speech & Language Decoding
  • Visual Decoding & Prosthesis
  • Affective / Memory / Cognitive State
• By Research Direction
  • Decoding Algorithms & Methods
  • Neural Foundation Models
  • Latent Dynamics Models
  • Long-Term Stability & Recalibration
  • Cross-Subject / Cross-Session Transfer
  • Self-Supervised & Representation Learning
  • Neural Data Generation & Synthesis
  • Closed-Loop & Co-Adaptation
  • LLM / Multimodal / Diffusion Decoding
  • Real-Time / Adaptive Systems
  • China Teams
• Datasets
• Benchmarks & Competitions
• Tools & Libraries
• Recommended Venues
• Non-Invasive BCI (coming soon)
• Contributing

---

How This List Is Organized

A single paper often spans multiple axes (e.g. Willett 2023 is intracortical + speech + a decoding-method contribution). To avoid forcing every paper into one box, this list is organized along three orthogonal axes, and landmark papers are cross-listed where it helps:

1. By Acquisition Modality — how the neural signal is recorded (intracortical array, ECoG, sEEG, Neuropixels).
2. By Task — what is being decoded (movement, speech, vision, cognitive state).
3. By Research Direction — the scientific/engineering problem being tackled (pure decoding, foundation models, long-term stability, data generation, …).

If you only want the milestones, start with Surveys & Reviews then Motor Decoding and Speech & Language Decoding.

---

Surveys & Reviews

• 📚 Human intracortical recording and neural decoding for brain–computer interfaces (Brandman, Cash & Hochberg, 2017, IEEE TNSRE) — paper
• 📚 Computation Through Neural Population Dynamics (Vyas, Golub, Sussillo & Shenoy, 2020, Annual Review of Neuroscience) — paper
• 📚 Neural Decoding for Intracortical Brain–Computer Interfaces (2023, Cyborg and Bionic Systems) — paper
• 📚 The speech neuroprosthesis (Silva, Liu & Chang, 2024, Nature Reviews Neuroscience) — paper
• 📚 Restoring Speech Using Brain–Computer Interfaces (2024, Annual Review of Biomedical Engineering) — paper
• 📚 The expanding repertoire of brain–computer interfaces (2024, Nature Medicine) — paper
• 📚 The speech neuroprosthesis (Luo, Anumanchipalli & Chang, 2024, Nature Reviews Neuroscience) — review of speech-decoding BCIs and articulatory representations. paper
• 📚 Development of visual neuroprostheses: trends and challenges (Fernández et al., 2020, Bioelectronic Medicine) — paper
• 📚 Implantable intracortical microelectrodes: reviewing the present with a focus on the future (2022, Microsystems & Nanoengineering) — electrode tech, biocompatibility, longevity. paper
• 📚 Foundation and large-scale AI models in neuroscience (2025, arXiv) — survey of brain foundation models & large-scale pretraining. paper

---

1. By Acquisition Modality (采集方式)

Intracortical Microelectrode Arrays (Utah / microwire)

Single-/multi-unit spikes and threshold crossings from penetrating arrays (typically M1, PMd, PPC, or speech motor cortex).

• 📄 Neuronal ensemble control of prosthetic devices by a human with tetraplegia (Hochberg / BrainGate, 2006, Nature) — first human Utah-array cursor + prosthetic control. paper
• 📄 Reach and grasp by people with tetraplegia using a neurally controlled robotic arm (Hochberg / BrainGate, 2012, Nature) — paper
• 📄 High-performance neuroprosthetic control by an individual with tetraplegia (Collinger / Schwartz, 2013, The Lancet) — 7-DoF prosthetic limb. paper
• 📄 Decoding motor imagery from the posterior parietal cortex of a tetraplegic human (Aflalo / Andersen, 2015, Science) — opened PPC as an implant site. paper
• 📄 High-performance brain-to-text communication via handwriting (Willett / Shenoy, 2021, Nature) — 90 char/min. paper · 💻 code
• 📄 A high-performance speech neuroprosthesis (Willett / Henderson, 2023, Nature) — 62 words/min, 125k vocab. paper · 💻 code
• 📄 A Year of Telepathy — Neuralink N1 / PRIME study (Neuralink, 2024) — first human flexible-thread implant for cursor control. report
• 📄 Fully implanted brain–computer interface in a locked-in patient with ALS (Vansteensel et al., 2016, NEJM) — first fully-implanted home-use communication BCI. paper
• 📄 The Stentrode endovascular BCI: SWITCH study safety and feasibility (Oxley / Opie, 2023, JAMA Neurology) — first-in-human minimally invasive endovascular (stent-mounted) BCI. paper

ECoG (Electrocorticography)

Subdural / surface high-density grids — broad spatial coverage of speech & sensorimotor cortex.

• 📄 Speech synthesis from neural decoding of spoken sentences (Anumanchipalli / Chartier / Chang, 2019, Nature) — articulatory-intermediate speech synthesis. paper
• 📄 Machine translation of cortical activity to text with an encoder–decoder framework (Makin / Chang, 2020, Nature Neuroscience) — WER as low as 3%. paper
• 📄 Neuroprosthesis for decoding speech in a paralyzed person with anarthria (Moses / Chang, 2021, NEJM) — first real-time sentence decoding in a paralyzed patient. paper
• 📄 A high-performance neuroprosthesis for speech decoding and avatar control (Metzger / Chang, 2023, Nature) — text + synthesized voice + avatar. paper
• 📄 Walking naturally after spinal cord injury using a brain–spine interface (Lorach / Courtine, 2023, Nature) — ECoG-driven digital bridge to spinal stimulation. paper

Stereo-EEG / Depth Electrodes

Sparse depth contacts (often clinical epilepsy implants) used opportunistically for decoding.

• 📄 Real-time synthesis of imagined speech processes from minimally invasive recordings (Angrick / Herff, 2021, Communications Biology) — closed-loop sEEG speech synthesis. paper · 💻 code
• 📄 BrainBERT: self-supervised representation learning for intracranial recordings (Wang et al., 2023, ICLR) — pretrained on unlabeled intracranial (sEEG/ECoG) recordings. paper · 💻 code
• 📄 Brant: foundation model for intracranial neural signal (Zhang et al., 2023, NeurIPS) — large intracranial pretrained model. paper · 💻 code

Neuropixels / High-Density Silicon Probes

Hundreds–thousands of channels; dominant in rodent/NHP systems neuroscience and increasingly in human research.

• 📄 Single-neuron speech sound encoding across the depth of human cortex (Leonard / Chang, 2023, Nature) — Neuropixels in humans reveal single-neuron speech tuning across cortical depth. paper
• 📄 Ultraflexible electrode arrays for months-long high-density recording (Luan / Xie, 2022, Nature Communications) — penetrating ultraflexible arrays stably record thousands of neurons for months. paper
• 📊 Allen Brain Observatory — Visual Coding Neuropixels (Allen Institute, 2019) — see Datasets.
• 📊 International Brain Laboratory (IBL) brain-wide map (IBL, 2021) — see Datasets.
• 💻 Kilosort — GPU spike sorting with drift correction for HD probes; see Tools.

---

2. By Task (任务)

Motor Decoding (运动解码)

Cursor / typing

• 📄 High-performance communication by people with paralysis using an intracortical BCI (Pandarinath / Jarosiewicz, 2017, eLife) — point-and-click typing. paper
• 📄 A high-performance neural prosthesis enabled by control algorithm design (ReFIT-KF) (Gilja / Shenoy, 2012, Nature Neuroscience) — closed-loop intention-retrained Kalman filter. paper
• 📄 Advantages of closed-loop calibration in intracortical BCIs for people with tetraplegia (Jarosiewicz et al., 2013, J. Neural Eng.) — established closed-loop > open-loop calibration. paper
• 📄 Clinical translation of a high-performance neural prosthesis (Gilja et al., 2015, Nature Medicine) — first clinical translation of ReFIT to an ALS participant. paper
• 📄 Rapid calibration of an intracortical BCI for people with tetraplegia (Brandman et al., 2018, J. Neural Eng.) — sub-3-minute decoder calibration. paper
• 📄 Cortical control of a tablet computer by people with paralysis (Nuyujukian et al., 2018, PLOS ONE) — iBCI control of an unmodified commercial tablet. paper
• 📄 Brain control of bimanual movement enabled by recurrent neural networks (Deo / BrainGate, 2024, Scientific Reports) — two cursors via RNN. paper

Robotic arm / prosthesis

• 📄 Direct cortical control of 3D neuroprosthetic devices (Taylor / Tillery / Schwartz, 2002, Science) — foundational 3D cortical device control. paper
• 📄 Cortical control of a prosthetic arm for self-feeding (Velliste / Schwartz, 2008, Nature) — paper
• 📄 Robust decoding of hand kinematics from entire spiking activity using deep learning (Ahmadi / Constandinou, 2021, J. Neural Eng.) — quasi-RNN on ESA for chronically robust kinematics. paper
• 📄 Real-time NHP BCI achieves high-velocity prosthetic finger movements (Nason-Tomaszewski / Chestek, 2022, Nature Communications) — shallow FFN beats ReFIT-KF. paper
• 📄 A high-performance BCI for finger decoding and quadcopter control (Stanford / BrainGate, 2024, Nature Medicine) — independent finger groups. paper
• 📄 Decoding ten-finger movements in human PPC and motor cortex (Caltech / Andersen, 2023, J. Neural Eng.) — paper

Handwriting

• 📄 High-performance brain-to-text communication via handwriting (Willett / Shenoy, 2021, Nature) — paper · 💻 code

FES / walking / somatosensory feedback

• 📄 Restoring cortical control of functional movement in a human with quadriplegia (Bouton, 2016, Nature) — cortex→FES neural bypass. paper
• 📄 Restoration of reaching and grasping through brain-controlled muscle stimulation (Ajiboye / Hochberg, 2017, The Lancet) — paper
• 📄 Intracortical microstimulation of human somatosensory cortex (Flesher / Gaunt, 2016, Science Translational Medicine) — bidirectional BCI tactile feedback. paper
• 📄 Stable, precise tactile sensations via multi-electrode ICMS of S1 (2024, Nature Biomedical Engineering) — paper
• 📄 Walking naturally after spinal cord injury using a brain–spine interface (Lorach / Courtine, 2023, Nature) — ECoG-driven digital bridge to spinal stimulation. paper

Speech & Language Decoding (语音与语言解码)

Speech-to-text

2015–2020

• 📄 Brain-to-text: decoding spoken phrases from phone representations (Herff / Schultz, 2015, Frontiers in Neuroscience) — origin of the "Brain-to-Text" paradigm. paper
• 📄 Real-time decoding of question-and-answer speech dialogue using human cortical activity (Moses / Chang, 2019, Nature Communications) — interactive perceived-question + produced-answer decoding. paper
• 📄 Neural ensemble dynamics in dorsal motor cortex during speech in people with paralysis (Stavisky et al., 2019, eLife) — dorsal M1 also encodes articulator movement. paper
• 📄 Decoding spoken English from intracortical arrays in dorsal precentral gyrus (Wilson / Stavisky, 2020, J. Neural Eng.) — phoneme decoding from a non-canonical speech site. paper

2022–2023

• 📄 Generalizable spelling using a speech neuroprosthesis (Metzger / Chang, 2022, Nature Communications) — 1152-word silent spelling. paper
• 📄 A high-performance speech neuroprosthesis (Willett / Henderson, 2023, Nature) — paper · 💻 code
• 📄 High-resolution neural recordings improve the accuracy of speech decoding (Metzger / Chang, 2023, Nature Communications) — denser ECoG grids → better speech decoding. paper

2024–2025

• 📄 An accurate and rapidly calibrating speech neuroprosthesis (Card / Stavisky, 2024, NEJM) — 99.6% on 50 words, 30-min calibration. paper
• 📄 A bilingual speech neuroprosthesis via shared cortical articulatory representations (Silva / Chang, 2024, Nature Biomedical Engineering) — paper
• 📄 Inner speech in motor cortex and implications for speech neuroprostheses (Kunz / Willett / Henderson, 2025, Cell) — paper
• 📄 Brain-to-text with context-aware neural representations and LLMs (2025, J. Neural Eng.) — Brain-to-Text '24 winning approach. paper
• 📄 Transfer learning via distributed brain recordings enables reliable speech BCI (2025, Nature Communications) — distributed recordings + LLMs + transfer learning. paper

Speech synthesis

• 📄 Reconstructing speech from human auditory cortex (Pasley et al., 2012, PLoS Biology) — origin of neural speech reconstruction. paper
• 📄 Speech synthesis from ECoG using densely connected 3D CNNs (Angrick / Herff, 2019, J. Neural Eng.) — paper
• 📄 A neural speech decoding framework leveraging deep learning and speech synthesis (Chen / Wang, NYU, 2024, Nature Machine Intelligence) — paper · 💻 code
• 📄 A streaming brain-to-voice neuroprosthesis to restore naturalistic communication (Littlejohn / Cho / Anumanchipalli / Chang, 2025, Nature Neuroscience) — near-real-time streaming synthesis. paper
• 📄 An instantaneous voice-synthesis neuroprosthesis (Wairagkar et al., 2025, Nature) — closed-loop instantaneous brain-to-voice with audio feedback (UC Davis/BrainGate). paper · 💻 code

Imagined / internal speech

• 📄 Representation of internal speech by single neurons in human supramarginal gyrus (Wandelt / Andersen, 2024, Nature Human Behaviour) — paper · 💻 code

Tonal language (Mandarin)

• 📄 Decoding and synthesizing tonal language speech from brain activity (2023, Science Advances) — paper
• 📄 A brain-to-text framework for decoding natural tonal sentences (2024, Cell Reports) — paper
• 📄 Real-time decoding of full-spectrum spoken Mandarin syllables (2025, Science Advances) — paper

Visual Decoding & Prosthesis (视觉解码)

Pure image-reconstruction-from-cortex with invasive recordings is still sparse; most work here is cortical-stimulation visual prosthesis (phosphene generation).

• 📄 Shape perception via a high-channel-count neuroprosthesis in monkey visual cortex (Chen / Roelfsema, 2020, Science) — 1024-channel V1/V4 phosphene patterns. paper
• 📄 Dynamic stimulation of visual cortex produces form vision in sighted and blind humans (Beauchamp / Yoshor, 2020, Cell) — paper
• 📄 MEIcoder: decoding visual stimuli from neural activity (2025, arXiv) — reconstruction from single-cell V1 activity. paper

Affective / Memory / Cognitive State (情绪/记忆/认知)

• 📄 Mood variations decoded from multi-site intracranial human brain activity (Sani / Shanechi, 2018, Nature Biotechnology) — paper
• 📄 Closed-loop enhancement and neural decoding of cognitive control in humans (Widge et al., 2022, Nature Biomedical Engineering) — paper
• 📄 Developing a hippocampal neural prosthetic to facilitate human memory encoding and recall (Hampson / Song / Berger, 2018, J. Neural Eng.) — MIMO memory prosthesis. paper
• 📄 Closed-loop modulation of remote hippocampal representations (2024, Neuron) — paper

---

3. By Research Direction (研究方向)

Decoding Algorithms & Methods (纯解码)

Classic + modern decoders mapping neural activity → behavior.

Classic (2002–2017)

• 📄 Instant neural control of a movement signal (Serruya / Hatsopoulos / Donoghue, 2002, Nature) — foundational closed-loop primate cursor control. paper
• 📄 Learning to control a brain–machine interface for reaching and grasping by primates (Carmena / Nicolelis, 2003, PLoS Biology) — paper
• 📄 Bayesian population decoding of motor cortical activity using a Kalman filter (Wu et al., 2006, Neural Computation) — KF as the workhorse BCI decoder. paper
• 📄 A recurrent neural network for closed-loop intracortical BMI decoders (Sussillo et al., 2012, J. Neural Eng.) — earliest RNN (echo-state) closed-loop BMI decoder. paper
• 📄 A high-performing BMI driven by low-frequency LFPs alone and with spikes (Stavisky et al., 2015, J. Neural Eng.) — LFP as a stable complementary control signal. paper
• 📄 Making brain–machine interfaces robust to future neural variability (Sussillo et al., 2016, Nature Communications) — multiplicative RNN decoder robust to drift. paper
• 📄 Augmenting iBMI with neurally driven error detectors (Even-Chen et al., 2017, J. Neural Eng.) — closed-loop neural error correction. paper

Deep-learning era (2020–2025)

• 📄 Machine learning for neural decoding (Glaser / Kording, 2020, eNeuro) — tutorial + benchmark of modern ML decoders vs classical. paper · 💻 code
• 📄 Representation learning for neural population activity (Neural Data Transformer, NDT) (Ye / Pandarinath, 2021) — first Transformer for spiking activity. paper · 💻 code
• 📄 STNDT: spatiotemporal Transformer for neural population activity (Le & Shlizerman, 2022, NeurIPS) — paper · 💻 code
• 📄 seegnificant: neural decoding from sEEG accounting for electrode variability across subjects (2024, NeurIPS) — cross-subject training over heterogeneous sEEG layouts. paper
• 📄 LFADS (Pandarinath 2018) — see Latent Dynamics Models.

Neural Foundation Models (神经基础模型)

Large, pretrained, transferable models for neural data.

2023

• 📄 A unified, scalable framework for neural population decoding (POYO) (Azabou et al., 2023, NeurIPS) — spike tokenization + PerceiverIO. paper · 💻 code
• 📄 Neural Data Transformer 2 (NDT2): multi-context pretraining (Ye et al., 2023, NeurIPS) — paper · 💻 code
• 📄 BrainBERT: self-supervised representation learning for intracranial recordings (Wang et al., 2023, ICLR) — paper · 💻 code
• 📄 Brant: foundation model for intracranial neural signal (Zhang et al., 2023, NeurIPS) — paper · 💻 code

2024

• 📄 Towards a "universal translator" for neural dynamics (IBL-MtM) (Zhang, Wang et al., 2024, NeurIPS) — multi-task masking foundation model. paper · 💻 code
• 📄 Neuroformer: multimodal & multitask generative pretraining for brain data (Antoniades et al., 2024, ICLR) — paper · 💻 code
• 📄 Brant-2: foundation model for brain signals (ZJU-BrainNet, 2024) — extends Brant to multi-modal (sEEG + EEG) with broader cross-subject coverage. paper

2025

• 📄 POYO+: multi-session, multi-task decoding across cell-types & regions (Azabou et al., 2025, ICLR) — paper · 💻 code
• 📄 Neural Data Transformer 3 (NDT3): a generalist intracortical motor decoder (Ye et al., 2025, bioRxiv) — pretrained on ~2000 h. paper · 💻 code
• 📄 Population Transformer (PopT): population-level representations of neural activity (Chau et al., 2025, ICLR) — paper · 💻 code
• 📄 POSSM: generalizable, real-time neural decoding with hybrid state-space models (Ryoo, Azabou et al., 2025) — paper
• 📄 NEDS: neural encoding and decoding at scale (Zhang et al. / IBL, 2025, ICML, Spotlight) — multimodal multi-task joint encoding+decoding pretrained across animals. paper
• 📄 NeurIPT: foundation model for neural interfaces (2025, NeurIPS) — general pretraining spanning clinical diagnosis to BCI decoding. paper
• 📄 NuCLR: learning neuron identity from population context (2025, NeurIPS) — self-supervised neuron-level representations from raw population activity. paper

Latent Dynamics Models (神经流形与动力学)

Foundations (2012–2018)

• 📄 Neural population dynamics during reaching (Churchland / Cunningham / Shenoy, 2012, Nature) — dynamical-systems view of motor cortex that underpins modern decoders. paper
• 📄 LFADS: latent factor analysis via dynamical systems (original) (Sussillo / Pandarinath, 2016) — paper · 💻 code
• 📄 Inferring single-trial neural population dynamics using sequential autoencoders (LFADS) (Pandarinath et al., 2018, Nature Methods) — paper · 💻 code

2020–2022

• 📄 pi-VAE: identifiable & interpretable latent models of neural activity (Zhou & Wei, 2020, NeurIPS) — paper · 💻 code
• 📄 Recurrent switching dynamical systems for multiple populations (mp-rSLDS) (Glaser / Linderman, 2020, NeurIPS) — joint switching dynamics across interacting populations. paper
• 📄 A general recurrent state-space framework for decision-making neural dynamics (Zoltowski / Pillow / Linderman, 2020, ICML) — recurrent SSM subsuming drift-diffusion models. paper
• 📄 Inferring latent dynamics via Poisson Latent Neural ODEs (PLNDE) (Kim / Pillow / Brody, 2021, ICML) — nonlinear latent dynamics of spike trains via neural ODEs. paper
• 📄 AutoLFADS: large-scale auto-tuned single-trial dynamics estimation (Keshtkaran / Sedler / Pandarinath, 2022, Nature Methods) — paper · 💻 code

2023–2025

• 📄 iLDS: infinite recurrent switching linear dynamical systems (Lee et al., 2023, ICLR) — paper
• 📄 gpSLDS: Gaussian-process switching linear dynamical systems (Hu et al., 2024, NeurIPS) — paper
• 📄 irSLDS: parsing neural dynamics with infinite recurrent switching LDS (Geadah / IBL / Pillow, 2024, ICLR) — auto-discovers number of dynamical regimes. paper
• 📄 cc-GPFA: conditionally-conjugate GPFA for spike-count data (Nadew / Quinn, 2024, ICML) — Pólya-Gamma augmented GPFA with closed-form inference. paper
• 📄 FINDR: flow-field inference from neural data using deep recurrent networks (Kim et al., 2025, ICML) — unsupervised nonlinear stochastic flow fields. paper
• 📄 iSSM: identifying neural dynamics using interventional state-space models (Macke lab, 2025, ICML) — leverages perturbations to recover causal dynamics. paper

Long-Term Stability & Recalibration (长期稳定性)

Keeping a decoder working over months/years despite neural drift and electrode degradation. This is the deepest section of the list — organized into the manifold-stability foundations, manifold-alignment stabilizers, domain-adaptation / adversarial alignment, clinical self-calibration, and the instability-measurement / benchmark works that ground them.

Neural-manifold foundations of stability

• 📄 Neural manifolds for the control of movement (Gallego / Perich / Solla / Miller, 2017, Neuron) — defines "neural modes" / the manifold framework; conceptual basis for treating the manifold as the stable substrate. paper
• 📄 Cortical population activity within a preserved neural manifold underlies multiple motor behaviors (Gallego / Perich / Miller, 2018, Nature Communications) — one preserved manifold supports many behaviors. paper
• 📄 A neural population mechanism for rapid learning (Perich / Gallego / Miller, 2018, Neuron) — new output without changing within-manifold structure → manifold as stable scaffold. paper
• 📄 Learning by neural reassociation (Golub / Yu / Chase / Batista, 2018, Nature Neuroscience) — short-term BCI learning recombines fixed activity patterns. paper
• 📄 Long-term stability of cortical population dynamics underlying consistent behavior (Gallego / Perich / Miller, 2020, Nature Neuroscience) — cornerstone: latent dynamics stay stable across years, enabling manifold-based decoders that survive while unit-based decoders fail. paper
• 📄 Preserved neural dynamics across animals performing similar behaviour (Safaie / Gallego / Miller, 2023, Nature) — CCA-aligned latent dynamics transfer across individuals. paper · 💻 code
• 📄 Aligning latent representations of neural activity (Gallego group, 2022, Nature Reviews Neuroscience) — review: low-dim latent alignment as the principled way to compare neural activity across time/neurons/individuals. paper

Manifold-alignment stabilizers & recalibration

• 📄 Stabilization of a BCI via alignment of low-dimensional neural spaces (Degenhart / Bishop / Yu, 2020, Nature Biomedical Engineering) — unsupervised manifold-alignment stabilizer. paper · 💻 code
• 📄 Stabilizing BCIs through alignment of latent dynamics (NoMAD) (Karpowicz / Pandarinath, 2025, Nature Communications) — LFADS + unsupervised distribution alignment. paper · 💻 code
• 📄 Plug-and-play stability for intracortical BCIs (2024, IEEE) — paper
• 📄 Plug-and-play stability: a one-year demonstration of seamless brain-to-text (Fan et al., 2023, NeurIPS) — unsupervised self-recalibration, 93.84% over a year. paper · 📊 data
• 📄 Long-term unsupervised recalibration of cursor-based intracortical BCIs (HMM) (Wilson et al. / Stanford, 2025, Nature Biomedical Engineering) — HMM infers intended targets to retrain without calibration blocks. paper
• 📄 Closed-loop decoder adaptation shapes neural plasticity for skillful neuroprosthetic control (Orsborn / Carmena, 2014, Neuron) — classic CLDA co-adaptation paradigm. paper

Domain adaptation & adversarial alignment

• 📄 Adversarial domain adaptation for stable brain-machine interfaces (ADAN) (Farshchian / Gallego / Bengio / Miller, 2019, ICLR) — seminal adversarial latent-distribution alignment for BMI stability. paper · 💻 code
• 📄 Using adversarial networks to extend BCI decoding accuracy over time (Cycle-GAN aligner) (Ma et al. / Miller Lab, 2023, eLife) — cycle-consistent full-dimensional alignment across days; outperforms ADAN. paper · 💻 code
• 📄 Robust alignment of cross-session recordings by behaviour via unsupervised domain adaptation (Jude / Hennig, 2022, ICML) — seqVAE + UDA for stable cross-session latents. paper
• 📄 ERDiff: latent dynamics alignment with diffusion models (Wang et al. / Georgia Tech, 2023, NeurIPS) — source-free distribution alignment using a diffusion prior. paper · 💻 code
• 📄 A cryptography-based approach for movement decoding (Dyer et al., 2017, Nature Biomedical Engineering) — label-free cross-time/subject alignment, early recalibration paradigm. paper
• 📄 SeSA: speed-enhanced subdomain adaptation for long-term stable neural decoding (Wang / Wang et al. / ZJU, 2024) — subdomain-alignment regression for long-term stability. paper
• 📄 Partial domain adaptation for stable neural decoding in intracortical BCIs (Wang / Qi et al. / ZJU, 2025) — partial DA for stable cross-day representations. paper
• 📄 T-TIME: test-time information-maximization ensemble for plug-and-play BCIs (2024) — fully-online label-free test-time adaptation. paper

Clinical self-calibration (intent-inference)

• 📄 Virtual typing by people with tetraplegia using a self-calibrating intracortical BCI (Jarosiewicz / BrainGate, 2015, Science Translational Medicine) — retrospective target inference (RTI) auto-compensates nonstationarity. paper
• 📄 Retrospectively supervised click-decoder calibration for self-calibrating point-and-click BCIs (BrainGate, 2017) — recalibrate clicks from real usage data. paper

Measuring instability, longevity & benchmarks

• 📄 Intra-day signal instabilities affect decoding performance in an intracortical neural interface (Perge / BrainGate, 2013, J. Neural Eng.) — foundational characterization of why recalibration is needed. paper
• 📄 MINDFUL: measuring instability in chronic human intracortical recordings (2024, Communications Biology) — label-free instability metric to decide when to recalibrate. paper
• 📄 FALCON: few-shot algorithms for consistent neural decoding (SNEL / Pandarinath, 2024, NeurIPS D&B) — first unified benchmark for stable long-term decoding. paper
• 📄 LINK: long-term intracortical neural activity & kinematics (Temmar / Willsey, 2025, NeurIPS D&B) — 303 sessions / 117k trials of chronic finger movement for drift & stability research. paper
• 📄 Longevity and reliability of chronic unit recordings using Utah arrays (2021, J. Neural Eng.) — paper

Cross-Subject / Cross-Session Transfer (跨被试/跨会话迁移)

• 📄 MYOW: mine your own view — self-supervised representation learning (Azabou / Dyer, 2021) — generalizes across V1/CA1/motor cortex. paper · 💻 code
• 📄 Robust alignment of cross-session recordings via behaviour (UDA + seqVAE) (2022) — unsupervised domain adaptation for stable cross-session decoding. paper
• 📄 Neural Latent Aligner (NLA): cross-trial alignment of neural representations (Cho et al., 2023, ICML) — unsupervised cross-trial latent alignment for speech/neural data. paper
• 📄 Leveraging generative models for unsupervised alignment of neural time series (2024, ICLR) — source-free seqVAE reuse to stabilize decoders across sessions. paper
• 📄 FDA: flow matching for few-trial neural adaptation with stable latent dynamics (Wang et al., 2025, ICML) — flow-matching distribution alignment for fast decoder adaptation. paper · 💻 code
• 📄 Neural Embeddings Rank (NER): aligning 3D latent dynamics with movements (2024, NeurIPS) — rank-preserving embedding for stable long-term M1/PMd decoding. paper
• 📄 MINT: simple decoding of behavior from a complicated neural manifold (Perkins / Cunningham / Churchland, 2024, eLife) — interpretable manifold-based decoder, robust in stereotyped BCI settings. paper
• 📄 Transfer learning in BCIs with adversarial variational autoencoders (Özdenizci et al. / MERL, 2018) — adversarial VAE for invariant cross-subject/session features. paper
• 📄 (see also POYO, NDT2, PopT under Foundation Models; and the Long-Term Stability section, whose manifold-alignment / domain-adaptation methods directly target cross-session transfer)

Self-Supervised & Representation Learning (自监督表征学习)

• 📄 CEBRA: learnable latent embeddings for joint behavioural & neural analysis (Schneider / Lee / Mathis, 2023, Nature) — paper · 💻 code
• 📄 Swap-VAE: drop, swap, and generate — disentangling neural activity (Liu / Azabou / Dyer, 2021, NeurIPS) — paper

Neural Data Generation & Synthesis (数据生成)

• 📄 LDNS: latent diffusion for neural spiking data (Kapoor / Macke, 2024, NeurIPS) — paper · 💻 code
• 📄 Spike-GAN: synthesizing realistic neural population activity with GANs (Molano-Mazón / Panzeri, 2018, ICLR) — paper
• 📄 Spiking GANs with a neural network discriminator (Rosenfeld / Simeone, 2022, IEEE TC) — paper
• 📄 Diffusion-based generation of neural activity from disentangled latent codes (McCart / Sedler / Pandarinath, 2024) — conditional diffusion for unsupervised disentangled spiking codes. paper
• 📄 BeNeDiff: behavior-relevant & disentangled neural dynamics via diffusion (Wu et al., 2024, NeurIPS) — isolates and generates behavior-relevant neural dynamics. paper
• 📄 Generative modeling of neural dynamics via latent stochastic differential equations (2024) — latent SDE with state/input-dependent drift & diffusion. paper
• 📄 LangevinFlow: Langevin flows for modeling neural latent dynamics (Song et al., 2025) — physics-inspired seqVAE with underdamped Langevin latents. paper

Closed-Loop & Co-Adaptation (闭环与协同自适应)

• 📄 Pain control by co-adaptive learning in a brain–machine interface (Zhang / Hu, 2020, Current Biology) — paper
• 📄 BRAND: backend for realtime asynchronous neural decoding (Ali et al. / SNEL, 2024) — real-time closed-loop platform; see Tools. paper
• 📄 Closed-loop control of theta oscillations enhances human episodic memory (2025, Nature Communications) — paper
• 📄 Machine-learning-based brain-signal decoding for adaptive DBS (2022, Experimental Neurology) — ML decoding of neural biomarkers driving adaptive DBS (PD/ET/dystonia). paper
• 📄 Closed-loop deep brain stimulation with reinforcement learning and neural simulation (2024, IEEE TNSRE) — RL-optimized stimulation parameters for Parkinson's aDBS. paper

LLM / Multimodal / Diffusion Decoding (大模型与多模态)

Emerging direction coupling neural decoders with large language models, multimodal fusion, and diffusion generative models.

• 📄 Towards an end-to-end framework for invasive brain signal decoding with LLMs (2024, Interspeech) — couples LLMs with end-to-end speech neuroprosthesis decoding. paper
• 📄 LLMs help alleviate cross-subject variability in brain signal decoding (2025) — LLM embeddings to align neural signals with language representations across subjects. paper
• 📄 Progress, challenges and future of linguistic neural decoding with LLMs (2025, Communications Biology) — survey of LLM-driven brain-to-text decoding. paper
• 📄 Transformer-based neural speech decoding from surface and depth electrodes (2025, J. Neural Eng.) — Transformer over non-grid surface + depth intracranial electrodes. paper
• 📄 (see also BrainBERT, Brant, Neuroformer, NEDS under Foundation Models)

Real-Time / Adaptive Systems (实时与自适应)

• 📄 Tailoring deep learning for real-time BCIs (RAP) (2025) — real-time adaptive pooling retrofits offline deep models for online decoding. paper
• 📄 EDAPT: towards calibration-free BCIs with continual online adaptation (2025, J. Neural Eng.) — task/model-agnostic continual online learning. paper
• 📄 Low-latency neural inference on an edge device for real-time imagined handwriting (2025, Scientific Reports) — edge-device low-latency imagined-handwriting recognition. paper
• 📄 (see also BRAND real-time platform under Tools)

China Teams (国内团队)

Selected invasive-BCI decoding work from Chinese groups (Tsinghua, NeuCyber/SIMIT, Fudan, etc.). Entries without a peer-reviewed paper are marked as preprint/report.

• 📄 Fine-grained 2D cursor control with an epidural minimally invasive BCI (NEO) (Tsinghua / Hong Bo, 2025, medRxiv preprint) — 8-contact epidural ECoG (NEO) for fine 2D cursor control in tetraplegia. paper
• 📄 Real-time decoding of full-spectrum spoken Mandarin (NeuCyber / SIMIT, 2025, Science Advances) — see Speech → Mandarin. paper
• 📄 SEEG-audio contrastive matching for Chinese speech decoding (2025) — Mandarin sEEG speech-decoding paradigm with contrastive audio alignment. paper
• 📄 Decoding handwriting trajectories from intracortical brain signals (2025, Advanced Science) — continuous handwriting-trajectory decoding from cortex. paper
• 📄 Brant: foundation model for intracranial neural signal (Zhejiang Univ. / ZJU-BrainNet, 2023, NeurIPS) — see Foundation Models. paper

---

Datasets (数据集)

• 📊 Neural Latents Benchmark '21 — MC_Maze / MC_RTT / Area2_Bump / DMFC_RSG (Pei et al., 2021) — curated intracortical spiking datasets. data · 💻 tools
• 📊 Sabes Lab — NHP reaching with sensorimotor arrays (O'Doherty / Sabes, 2017) — ~20k macaque reaches with M1/S1 spikes + LFP. data
• 📊 AJILE12 — long-term naturalistic human ECoG (Peterson et al., 2022) — 55 days of ECoG + pose across 12 subjects. paper · data
• 📊 Willett et al. speech neuroprosthesis data (Stanford / BrainGate, 2023) — intracortical spiking during attempted speech (~12k sentences). data
• 📊 Allen Brain Observatory — Visual Coding Neuropixels (Allen Institute, 2019) — data · 💻 SDK
• 📊 International Brain Laboratory (IBL) brain-wide map (IBL, 2021) — standardized Neuropixels across many labs. paper · data
• 📊 DANDI Archive (BRAIN Initiative) — NWB-formatted neurophysiology archive incl. human single-unit & Neuropixels. data
• 📊 Willett handwriting BCI dataset (Stanford / Dryad, 2021) — intracortical spiking during attempted handwriting. data
• 📊 Flint center-out reaching (CRCNS pmd-1) (Flint / Slutzky, Northwestern) — macaque M1/PMd spikes + LFP; common kinematic-decoding benchmark. data
• 📊 O'Doherty/Makin/Sabes NHP reaching (Indy/Loco) (2020) — multichannel M1/S1 self-paced grid reaching; used in decoder-stability studies. data
• 📊 Kai Miller human ECoG library (Miller, Stanford, 2016/2019) — curated human ECoG across 16 behavioral experiments. data
• 📊 Neurotycho macaque ECoG (Fujii / Nagasaka, RIKEN) — long-duration multichannel macaque ECoG with synchronized behavior. data
• 📊 IEEG.org — International Epilepsy Electrophysiology Portal (Litt Lab, UPenn) — large-scale human intracranial EEG/ECoG with tools + API. portal
• 📊 Epilepsy-iEEG multicenter dataset (OpenNeuro ds003029) — standardized BIDS-iEEG human intracranial recordings. data
• 📊 Human single-neuron Sternberg working-memory dataset (Rutishauser group, 2024, Scientific Data) — rare human single-unit recordings with behavior. data

---

Benchmarks & Competitions (基准与竞赛)

• 🏆 Neural Latents Benchmark '21 (Pei et al., 2021, NeurIPS D&B) — latent-variable models of spiking activity. paper · 💻 code
• 🏆 FALCON — few-shot algorithms for consistent neural decoding (Karpowicz et al. / SNEL, 2024, NeurIPS D&B) — decoder stability across sessions. paper · site
• 🏆 Brain-to-Text Benchmark '24 (Willett et al. / BrainGate, 2024) — attempted-speech → text. paper · site
• 🏆 Brain-to-Text '25 (Kaggle) (Willett et al. / BrainGate, 2025) — site

---

Tools & Libraries (工具与库)

• 💻 CEBRA — consistent latent embeddings from neural + behavioral data. code
• 💻 POYO / torch_brain — unified transformer framework for population decoding. code
• 💻 SpikeInterface — unified Python framework wrapping many spike sorters. code
• 💻 Kilosort — GPU template-matching spike sorter with drift correction. code
• 💻 Neo — common object model + readers/writers for electrophysiology formats. code
• 💻 Neurodata Without Borders (NWB) — standardized neurophysiology data format & ecosystem. paper
• 💻 BRAND — graph-of-nodes platform for low-latency real-time closed-loop BCI. paper
• 💻 nlb_tools — loading & evaluation for the Neural Latents Benchmark. code
• 💻 falcon-challenge — submit/evaluate decoders on the FALCON few-shot benchmark. code
• 💻 KordingLab Neural_Decoding — classic + ML/deep decoders (Wiener, Kalman, LSTM, XGBoost). code
• 💻 lfads-torch — PyTorch reimplementation of LFADS. code
• 💻 autolfads-tf2 — TF2 AutoLFADS with population-based hyperparameter tuning. code
• 💻 MountainSort5 — density-based spike sorter (Flatiron Institute). code
• 💻 Tridesclous — offline + real-time spike sorter. code
• 💻 Elephant — electrophysiology analysis toolkit (spikes, LFP) on Neo. code
• 💻 Pynapple — lightweight time-series analysis for systems neuroscience. code
• 💻 MNE-Python — human neurophysiology toolkit incl. iEEG/ECoG analysis. code

---

Recommended Venues (推荐会议与期刊)

A submission / tracking guide for this cross-disciplinary field — top conferences and journals across AI/ML, neuroscience, biomedical & neural engineering, clinical medicine, and the Nature/Science sub-journal family, annotated with CCF and Tsinghua (THU) ranks where they apply.

➡️ See the full list in VENUES.md.

Quick orientation:

• Method / algorithm work → NeurIPS · ICML · ICLR · AAAI (CCF-A); speech → ACL · ICASSP · Interspeech.
• Decoding systems / devices / clinical → J. Neural Eng. · IEEE TNSRE · IEEE TBME; conferences IEEE EMBC · NER · BCI Meeting.
• Landmark demonstrations → Nature · Science · NEJM · Nat. Biomed. Eng. · Nat. Neurosci. · Science Robotics.

---

Non-Invasive BCI (coming soon)

🚧 Coming soon. Non-invasive modalities (EEG, MEG, fNIRS, fMRI) are out of scope for this list today. A companion section / sibling list is planned. Contributions welcome — see Contributing.

---

Contributing

Contributions are very welcome! Please:

1. Keep the focus invasive (intracortical / ECoG / sEEG / depth / HD probes). Non-invasive work belongs in the future companion list.
2. Use the entry format: <emoji> **Title** (First-author / Lab, Year, Venue) — one-line contribution. [paper](url) · [💻 code](url) · [📊 data](url)
3. Place the paper under the axis that best fits; cross-list a landmark paper only when it genuinely helps discovery.
4. Verify every link resolves before submitting.
5. Prefer official DOIs / publisher pages / lab pages over aggregators when available.

Open an issue for discussion or a pull request to add entries.

---

License

To the extent possible under law, the maintainers have waived all copyright and related rights to this work (CC0 1.0). The linked papers and resources remain under their respective licenses.