crafted by Gabriel Duarte Arantes
Open research visualization

Neuro-inspired probabilistic deep learning framework

Public HTML interface for exploring calibrated probabilities, predictive uncertainty, neuro-inspired signals, and the topology of the artificial neural network behind the project, making the research navigable directly from the repository.

Responsible use
    Open exploration
    • The charts below use only the repository-hosted static payload.
    • Both 3D scenes run directly in the browser, without a live backend.
    • The viewer can be published as a static page for any repository visitor.

    3D exploration spaces

    Two three-dimensional environments inside the same HTML document: one for neuroscientific dynamics and uncertainty metrics, and one for the artificial neural network used by the deep learning model.

    crafted by Gabriel Duarte Arantes

    Neuro-ML uncertainty dome

    A 3D scene for visualizing signals, risk clusters, and the coupling between confidence, entropy, and predictive instability in the prototype.

    crafted by Gabriel Duarte Arantes

    Deep learning network chamber

    A 3D space dedicated to the classifier architecture: tabular input, temporal branch, multimodal fusion, and the final probabilistic decision layer.

      Mean class probabilities

      Mean calibrated class probabilities across the predictive distribution.

      Mean class variance

      Mean class-wise uncertainty estimated by Monte Carlo Dropout.

      Calibration curves

      Observed reliability compared with predicted confidence.

      Model comparison radar

      Comparison across raw, calibrated, and MC Dropout inference regimes.

      Training loss trajectory

      Compact training history for the research MVP.

      Validation metrics trajectory

      Validation accuracy and macro-F1 across epochs.

      Uncertainty scatter landscape

      Each point represents one sample. The X axis shows confidence, the Y axis shows predictive entropy, and marker size tracks mutual information.

      Confusion matrix: raw

      Deterministic output before post-hoc calibration.

      Confusion matrix: calibrated

      Output after temperature scaling.

      Confusion matrix: MC Dropout

      Mean output under stochastic inference.

      Architecture snapshot

      Text summary of the topology used by the neural network 3D scene.

      This visualization is interpretive and exploratory. It represents the logical multimodal skeleton of the architecture rather than a raw dump of the internal PyTorch graph.

      Repository-ready viewer

      Static viewer structure designed for public repository navigation.

      Static HTML Canvas 3D Chart.js No live backend Device: unknown

      To refresh the data, regenerate the local payload. The frontend remains accessible even for visitors who only want to explore the repository and understand the model visually.

      Most uncertain samples

      The most ambiguous cases under stochastic inference. Interpretation should always combine probability, entropy, and mutual information.

      Sample Predicted True Confidence Entropy MI Mean probabilities