Blog
Insights on BCI, neurotechnology, and probabilistic AI.
Reactive Message Passing for BCI: How RxInfer.jl Brings Active Inference to Real Time
Most Active Inference tutorials stop at the math. This post goes further — explaining how reactive message passing turns the Free Energy Principle into a runtime algorithm, and why RxInfer.jl makes it practical for real-time BCI pipelines.
Weekly Neurotech & BCI Digest — April 13, 2026
This week's curated digest for ML/BCI engineers: EEG foundation models, calibration-free decoding, Columbia's single-chip implant, g.tec Spring School, and the expanding BCI clinical ecosystem.
EEG Foundation Models in Practice: What REVE Brings to BCI Preprocessing
Classical EEG preprocessing relies on hand-tuned filters and spatial methods that break across sessions and subjects. This post explains how foundation models like REVE work, how they fit into the Nimbus Studio preprocessing stack alongside CSP and causal filtering, and why a learned preprocessing layer is a natural complement to probabilistic downstream models.
Active Inference vs. Deep Learning for BCI: Why Uncertainty Quantification Changes Everything
Deep learning dominates ML benchmarks, but BCI pipelines face a different challenge: uncertainty is not noise to suppress — it is signal to leverage. This post compares deep learning and Active Inference for neural decoding, and shows how Nimbus Studio and NimbusSDK make the probabilistic approach practical.
Choosing the Right Bayesian Classifier for Your BCI Pipeline
NimbusSDK ships four Bayesian classifiers, but which one belongs in your pipeline? This guide breaks down NimbusLDA, NimbusQDA, NimbusSoftmax, and NimbusSTS — what they assume, where they excel, and when to switch — so you can make the right call before training a single model.
Weekly Neurotech & BCI Digest — Week of April 6, 2026
This week: Neuralink's automated surgery push, Columbia's single-chip implant, China's BCI clinical trials, a new visual imagery EEG dataset, MOABB benchmarks, WEF regulation report, and a $400M+ March funding wave.
March at Nimbus Studio: EEG calibration and stronger research workflows
March updates: live EEG calibration, multi-subject benchmarking, semantic channel mapping, and reliability improvements.
NimbusSTS in Practice: Handling EEG Drift (Without Recalibration)
A practical, non-repetitive guide to NimbusSTS: when to use adaptive state-space decoding, how the predict/update loop maps to EEG drift, how to tune adaptation speed, and how to deploy in Nimbus Studio.
Weekly Neurotech & BCI Digest — March 30, 2026
This week: China lands the world's first commercial BCI approval, BrainGate hits a new typing speed record in Nature Neuroscience, foundation models for EEG decoding gain traction, and Beijing's five-year plan redraws the competitive map for neural interfaces.
Implementing Closed-Loop Active Inference BCI Control in Nimbus Studio
Hands-on guide to building a real-time closed-loop Active Inference control pipeline in Nimbus Studio: loop topology, update signals, and deployment tips.
What Is Active Inference? A Practical Primer for BCI Engineers
Active Inference is the theoretical engine behind Nimbus BCI — but for engineers coming from ML or classical neuroscience, the concepts can feel abstract. This primer breaks down generative models, free energy minimization, and real-time belief updating in practical terms, with direct connections to RxInfer, NimbusSDK, and Nimbus Studio.
Cross-Session BCI Transfer with Bayesian Priors: Reuse, Adapt, and Personalize
Focuses on cross-session transfer: how to carry priors across days/users, update with minimal calibration data, and avoid full retraining in practice.
Weekly Neurotech & BCI Digest — March 23, 2026
This week's top stories in neurotech: China's landmark first commercial BCI approval, BrainGate's high-speed iBCI typing results in Nature Neuroscience, Neuralink's Blindsight visual cortex push, a $400M+ funding surge, and a look at hardware-agnostic SDK tooling.
Beyond Binary: Multi-Class BCI Decoding with Bayesian Softmax and NimbusSoftmax
Most BCI tutorials assume two classes — left hand vs. right hand, P300 vs. non-P300. But real-world applications demand more. This post explains how Bayesian Multinomial Logistic Regression works, why it outperforms classical softmax for neural data, and how to build a multi-class decoder with NimbusSoftmax in Nimbus Studio.
Decoding P300 ERPs with Bayesian QDA: A Practical Guide for BCI Engineers
P300-based BCIs are among the most clinically validated paradigms — but classical LDA classifiers struggle when class distributions overlap and signal-to-noise ratios are low. This post explains how Bayesian Quadratic Discriminant Analysis (NimbusQDA) improves P300 detection through uncertainty quantification and flexible covariance modeling, and shows how to scaffold a complete speller pipeline in Nimbus Studio.
Bayesian CSP and Motor Imagery: Building a Robust EEG Decoder with NimbusLDA
Motor imagery BCI relies on spatial filtering and classification — but classical pipelines leave uncertainty on the table. This post walks ML and BCI engineers through combining Common Spatial Patterns (CSP) with Bayesian LDA using NimbusLDA to build decoders that are accurate, calibrated, and production-ready.
Neural Drift and Why It Breaks Your BCI Classifier (And How Adaptive Bayesian Models Fix It)
EEG signals shift within and across sessions, silently breaking classifiers trained on static data. This post explains why neural drift happens, how Bayesian adaptive models like NimbusSTS track and correct for it in real time, and what this means for building BCI systems that work outside the lab.
Weekly Neurotech & BCI Digest — March 16, 2026
This week's curated neurotech digest covers China's landmark first-ever commercial BCI approval, Neuralink's high-volume production push, FDA regulatory friction, and the state of open-source BCI tooling heading into mid-2026.
Explainable by Design: Why Probabilistic BCI Models Are Built for FDA Approval
Most BCI classifiers are accurate in the lab but opaque to regulators. This post explains how probabilistic AI and Active Inference produce transparent, auditable models that align with FDA's Bayesian guidance and accelerate the path to clinical deployment.
From Classifier to Agent: How Generative Models Are Redefining BCI Decoding
Most BCI systems treat neural decoding as a classification problem — but generative models and Active Inference reframe it as inference over a world model. This post walks ML and BCI engineers through the conceptual shift from discriminative classifiers to generative agents, and explains why it produces more robust, adaptive decoders.
Active Inference for Closed-Loop BCI: The Self-Correcting Architecture
A conceptual architecture guide to closed-loop BCI with Active Inference: the perception–action loop, expected free energy, and how self-correction emerges when actions are chosen to reduce uncertainty and prediction error.
Weekly Neurotech & BCI Digest — Mar 9, 2026
This week's digest covers China's national BCI policy push and NeuroXess's surface-mesh human trials, the emerging non-invasive ultrasound wave led by Merge Labs and Nudge's $100M raise, a systematic review on LLM-BCI integration, a new visual imagery flickering-pattern paradigm, and the FDA's draft Bayesian guidance and what it means for BCI trial design.
From EEG to Action: Building Your First Real-Time BCI Pipeline with Nimbus Studio
Building a real-time BCI pipeline from scratch takes weeks of boilerplate. This guide walks ML and BCI engineers through assembling a complete, production-ready pipeline using Nimbus Studio's visual interface and Bayesian SDK — from raw EEG to live predictions with confidence scores.
The Free Energy Principle, Demystified: A Practical Guide for BCI Engineers
The Free Energy Principle is the theoretical backbone of Active Inference — and increasingly, of state-of-the-art BCI systems. This post strips away the philosophy and explains what it actually means for engineers building real-time neural decoders.
Factor Graphs and Message Passing: The Engine Behind Real-Time Bayesian BCI
Most Bayesian inference algorithms are too slow for real-time BCI — unless you structure your model as a factor graph and let message passing do the heavy lifting. This post explains how belief propagation works, why it's ideal for neural decoding pipelines, and how RxInfer and Nimbus Studio make it practical.
Uncertainty Quantification in BCI: Why Confidence Scores Matter as Much as Accuracy
Most BCI classifiers report a single prediction — but for real-world deployment, knowing how confident that prediction is can be just as critical as getting it right. This post explains uncertainty quantification from first principles and shows how Bayesian models expose calibrated confidence scores for safer, more adaptive BCI systems.
Weekly Neurotech & BCI Digest — Mar 2, 2026
This week's digest covers MIT's transcranial focused ultrasound roadmap for deep-brain neuromodulation, Rice University's dye-free molecular atlas of an Alzheimer's brain, NeuroXess reaching 54 human implants with China's first battery-integrated BCI chip, CorTec's second human implantation, Meta's Neural Band accessibility research with University of Utah, the geopolitical acceleration of China's BCI ecosystem, and the 2026 Queen Elizabeth Prize awarded to neural interface pioneers.
Nimbus Studio: February 2026 Product Updates
February was a major release month for Nimbus Studio: the new desktop app, cloud GPU offloading, and the ZUNA foundation model preprocessing node.
Within-Session Non-Stationarity in EEG BCIs: An Adaptive Bayesian Approach (NimbusSTS)
Deep technical dive into within-session non-stationarity (drift) in EEG BCIs and how online state-space Bayesian adaptation (NimbusSTS) maintains performance.
Weekly Neurotech & BCI Digest — Mar 1, 2026
This week's digest covers inner speech decoding from motor cortex at 125k-word vocabulary, new electrode-scaling insights challenging cortical dimensionality theory, Columbia's single-chip neural implant, Neuralink's GB-PRIME trial reaching 7 participants and its push toward high-volume production, Paradromics entering clinical trials, and Morgan Stanley's $80B market outlook.
Introduction to Active Inference for BCI
A technical but accessible introduction to Active Inference for brain-computer interfaces (BCI), explaining the perception-action loop, variational free energy, and how probabilistic modeling enables robust real-time decoding and adaptive control.