DecNefLab: A Modular and Interpretable Simulation Framework for Decoded Neurofeedback

TL;DR

DecNefLab addresses the robustness and interpretability challenges of decoded neurofeedback by offering a modular in silico framework that replaces human participants with latent-variable generative models. It enables direct observation of internal cognitive states $Z$ while streaming observable data $X$ through a trainable classifier, thereby allowing causal analysis of how feedback shapes learning. The paper demonstrates a concrete instantiation using a Variational Autoencoder and Fashion-MNIST, showing that alternative-class choices, initial conditions, and stochastic regulation critically influence learning trajectories and apparent non-response. By exposing the full cognitive trajectory and feedback topology, DecNefLab supports principled protocol design, reproducibility, and rapid in silico experimentation prior to human studies.

Abstract

Decoded Neurofeedback (DecNef) is a flourishing non-invasive approach to brain modulation with wide-ranging applications in neuromedicine and cognitive neuroscience. However, progress in DecNef research remains constrained by subject-dependent learning variability, reliance on indirect measures to quantify progress, and the high cost and time demands of experimentation. We present DecNefLab, a modular and interpretable simulation framework that formalizes DecNef as a machine learning problem. Beyond providing a virtual laboratory, DecNefLab enables researchers to model, analyze and understand neurofeedback dynamics. Using latent variable generative models as simulated participants, DecNefLab allows direct observation of internal cognitive states and systematic evaluation of how different protocol designs and subject characteristics influence learning. We demonstrate how this approach can (i) reproduce empirical phenomena of DecNef learning, (ii) identify conditions under which DecNef feedback fails to induce learning, and (iii) guide the design of more robust and reliable DecNef protocols in silico before human implementation. In summary, DecNefLab bridges computational modeling and cognitive neuroscience, offering a principled foundation for methodological innovation, robust protocol design, and ultimately, a deeper understanding of DecNef-based brain modulation.

Figures (4)

• Figure 1: Pipeline of a generic DecNef simulation.
• Figure 2: Probabilities given by each binary discriminator to the images generated by the VAE via latent space sampling. Coordinates: $z_i$. Background Color: $p(y=y^\star \mid x_i)$, where $x_i$ is generated from $z_i$ using the VAE's decoder $D_\mathcal{G}$. Red markers: Location of the latent class prototypes. Warm-colored images: Class prototypes $x^\star$ and $x^{\mathrm{alt}}$ for the target and alternative classes. Grayscale images: Projections to the native space of the data of a random sample of latent coordinates, annotated by red numbers in the main panel and displaying $p(y=y^\star \mid x_i)$ above each grayscale image.
• Figure 3: Evolution of $p_t$ during the 1000 simulations for each experiment, grouped and averaged by initial point $z_0$ (thin lines). The color indicates the category of the Gaussian latent prototype from which each $z_0$ was sampled. The bold lines represent the average $p_t$ across all trajectories for which the initial point was sampled around the same latent prototype.
• Figure 4: Trajectories in the latent space for the DecNef simulation with target-class $y^\star=\text{"T-shirt/top"}$ using either a discriminator with the alternative-class "Trouser" (a) or "Dress" (b). Colored background: $p(y=y^\star \mid x)$ where $x=D_\mathcal{G}(z)$ for each coordinate $z$ and the VAE decoder denoted as $D_\mathcal{G}$. Thin white lines: all the individual trajectories. Bold lines: Average position of the 10 trajectories sharing the same $z_0$, with vertical color bar indicating time progression.