SPARK · Cognitive AI

Cognitive AI: modelling human cognition

Collecting signals is not the goal. SPARK builds a family of AI models trained on synchronised, documented cognitive data, to recognise cognitive states, anticipate how they change, relate them to performance and learn how each person works.

02 · What it measures

What the models work on

Cognitive states

EEG signatures of vigilance, sustained attention, fatigue or overload.

Change over time

What comes before a drop in focus, a slower reaction or a loss of regularity.

Brain and performance

Which cognitive features go with lines, braking, decisions and results.

03 · What the models learn

Five things the models learn

  1. 01

    Recognising cognitive states

    Identifying EEG signatures of vigilance, sustained attention, fatigue or cognitive overload.

  2. 02

    Anticipating cognitive decline

    Learning the changes that come before a drop in focus, a longer reaction time or a loss of regularity.

  3. 03

    Modelling brain and performance

    Learning which cognitive features go with lines, braking, decisions and performance on track.

  4. 04

    An individual cognitive model

    Learning how each person works: usual signatures, variations and room to progress.

  5. 05

    Adaptive training

    Adapting exercises, difficulty and training conditions to each person's capacities and progress.

04 · Cognitive Digital Twin

The person, and their cognitive twin

The real person

What they do, their results, their brain signals.

Their cognitive twin

What the model has learnt of how they work, and what it anticipates.

What the twin learns over time

  • Their reference brain function.
  • How they react to different situations.
  • How well they hold attention.
  • How sensitive they are to fatigue and pressure.
  • The conditions in which their performance degrades.
  • The conditions in which they learn best.

The way to AI cognitive coaching.

05 · Rigour

Models that can be trusted

Correlation is not causation: every model is tested on what it really predicts.

Models are validated on people they never saw during training.

Raw data, extracted features, experimental annotations and model predictions are kept strictly apart.

Consent, data governance and pseudonymisation come first, with clear conditions for reusing data to train models.

06 · Products

Three products, one platform

01

SPARK Cognitive Monitoring

Measuring and following cognitive states during the activity.

02

SPARK Cognitive Intelligence

AI models that analyse the links between cognition, behaviour and performance.

03

SPARK Cognitive Training

An intelligent system that personalises training and measures progress.

07 · Architecture

One architecture under every domain

Four layers, one platform.

  1. Raw
    Data

    • EEG
    • Eye tracking
    • ECG
    • Behaviour
    • Performance
  2. Cognitive
    Features

    • Attention
    • Mental workload
    • Fatigue
    • Stress
    • Anticipation
  3. Cognitive
    Intelligence

    • AI models
    • Correlations
    • Predictions
    • Adaptation
  4. Cognitive
    Performance

    • Understand
    • Predict
    • Improve
    • Train

Talk to us about your field

Tell us what you do. Someone on the team answers you personally.

Talk to us