Tracking functional changes in nonstationary signals with evolutionary ensemble bayesian model for robust neural decoding
Abstract
Neural signals are typical nonstationary data where the functional mapping between neural activities and the intentions (such as the velocity of movements) can occasionally change. Existing studies mostly use a fixed neural decoder, thus suffering from an unstable performance given neural functional changes. We propose a novel evolutionary ensemble framework (EvoEnsemble) to dynamically cope with changes in neural signals by evolving the decoder model accordingly. EvoEnsemble integrates evolutionary computation algorithms in a Bayesian framework where the fitness of models can be sequentially computed with their likelihoods according to the incoming data at each time slot, which enables online tracking of time-varying functions. Two strategies of evolve-at-changes and history-model-archive are designed to further improve efficiency and stability. Experiments with simulations and neural signals demonstrate that EvoEnsemble can track the changes in functions effectively thus improving the accuracy and robustness of neural decoding. The improvement is most significant in neural signals with functional changes.
Type
Publication
Advances in Neural Information Processing Systems
🔶 Coping with Neural Signal Nonstationarity in Brain-Computer Interfaces: The EvoEnsemble Approach 🔶
🔄 Background: Nonstationary Neural Signals
- Nonstationarity is a common characteristic in real-world signals.
- Neural signals are inherently nonstationary due to:
- Internal neural dynamics
- Synaptic plasticity driven by learning and adaptation
- This variability poses a major challenge to brain-computer interfaces (BCIs):
- A fixed decoder may fail as neural encoding functions evolve
- Leads to degradation in decoding accuracy over time
📘 Neural Encoding as a Time-Varying Function
The neural encoding process can be formulated as:
\[ y_t = h_t(x_t) + q_t \]Where:
- \( y_t \): observed neural signal at time \( t \)
- \( x_t \): the state to be encoded (e.g., cursor velocity)
- \( h_t(\cdot) \): time-varying neural encoding function
- \( q_t \): noise term (biological or external)
Most traditional decoders (e.g., OLE, Kalman Filter) assume a stationary encoding function:
\[ h_t(\cdot) \equiv h_0(\cdot), \quad \forall t \in \{1, 2, 3, \ldots\} \]However, in online BCI systems, studies have shown that:
- \( h_t(\cdot) \) changes significantly over time
- These changes are influenced by real-time feedback (e.g., cursor velocity, control errors)
- Encoding function variability can occur even within a single trial
❓ Challenge: How to Handle Functional Changes?
🛠️ 1. Re-calibration-Based Solutions
- Periodic/manual recalibration of decoders
- Brandman et al.: Fast Bayesian recalibration (2–5s)
- Eliminates open-loop phase
- Still interrupts user experience
- Not responsive to short-term signal changes
⚙️ 2. Adaptive/Dynamic Model-Based Solutions
a. Dual State-Space Models (Wang et al.)
- Estimate both kinematics and tuning functions
- Does not directly model neural variability
b. DyEnsemble (Qi et al.)
- Dynamic ensemble Bayesian filter
- Maintains a static model pool
- Assembles decoder online via Bayesian filtering
- Handles limited nonstationarity
- Limited under long-term continuous changes
🌱 Proposed Solution: EvoEnsemble
We propose the Evolutionary Ensemble Bayesian Filter (EvoEnsemble), treating neural functional changes as a functional tracking problem.
🧠 Key Ideas
- Maintains a population of candidate functions
- Uses evolutionary computation to adapt over time
- Computes fitness via signal likelihood
- Tracks time-varying functions using Bayesian model averaging
📌 Main Contributions
✅ Novel Bayesian-Evolutionary Framework
- Extends evolutionary computation for robust state estimation with neural functional drift
🔁 Online Functional and State Tracking
- Particle-based method to estimate both neural functions and latent states (e.g., cursor velocity)
⚡ Efficiency and Stability Strategies
- Evolve-at-changes: trigger evolution only when changes are detected
- History-model-archive: maintain stable models for reuse
🧪 Experimental Results
- Conducted on simulations and real neural signals
- EvoEnsemble outperforms traditional decoders (e.g., Kalman Filter)
- Shows significant improvement especially under nonstationary conditions
