Skip to content

Suite No. 3: Sensory Omission & Oddball Detection Paradigm

A tutorial demonstrating expected sensory stimuli, unexpected deviants, and sensory omission detection under package-native paradigm configuration.

Open in Colab

Learning Objectives

After completing this tutorial, you will understand:

  1. Sensory Omission Config — how to declare stimulus schedules where expected sensory events are dynamically omitted.
  2. Oddball (Deviant) Paradigm — how to configure rare deviant stimulus streams (expected vs unexpected).
  3. Event Window Analysis — how to segment and aggregate population responses across baseline, stimulus, and post-stimulus windows.
  4. Independent Readouts — how to inspect multi-area column outputs under distinct stimulus conditions.
  5. Coverage Limits — why omission responses are diagnosed as uncalibrated proxy readouts.

Biological/Computational Question

Question: How does a sensory omission (unexpected silence) modulate population and field readouts in a feedforward-recurrent column compared to standard and deviant tones?

Context: Spike adaptation and recurrent inhibition can cause a transient suppression of responses to repeated standard tones. In contrast, rare unexpected deviant tones or the abrupt omission of an expected tone can trigger distinctive transient rebound or novelty-proxy responses. Projecting these under clear paradigm conditions ensures exact validation of mismatch and omission diagnostics.

Mathematical Glossary Flow

Here, we define the events and conditions for the sensory omission paradigm:

1. expected Condition (Standard Tone)

  • Sensory sequence definition: A repeated standard tone (80% probability) presented after a baseline buffer. $\(\text{sequence} = (\text{pre}, \text{standard}, \text{post})\)$
  • Onset and Duration: $\(\tau_{\text{onset}} = 200.0\text{ ms}, \quad \tau_{\text{dur}} = 100.0\text{ ms}\)$
  • Worded description: The expected standard condition models repeated sensory inputs with standard gain.
  • Implementation location: jtfne.omission_oddball_paradigm()

2. unexpected Condition (Deviant Tone)

  • Sensory sequence definition: A rare deviant tone (10% probability) replacing the standard stimulus. $\(\text{sequence} = (\text{pre}, \text{deviant}, \text{post})\)$
  • Worded description: The unexpected deviant condition models sensory novelty with deviant gains.
  • Implementation location: jtfne.omission_oddball_paradigm()

3. omitted Condition (Sensory Omission)

  • Sensory sequence definition: The expected tone is omitted (10% probability), resulting in complete silence. $\(\text{sequence} = (\text{pre}, \text{silence}, \text{post\_omission})\)$
  • Worded description: The omitted condition verifies whether network transients or sensory rebounds occur at the expected stimulus window in the absence of external drive.
  • Implementation location: jtfne.omission_oddball_paradigm()

Canonical Import

All notebooks and scripts leverage the canonical packagefacade import:

import jaxfne as jtfne

Configuration & Simulation

# Create the omission/oddball paradigm configuration
paradigm = jtfne.omission_oddball_paradigm(
    standard_onset_ms=200.0,
    standard_duration_ms=100.0,
    deviant_duration_ms=100.0,
    pre_stimulus_buffer_ms=200.0,
    post_stimulus_buffer_ms=500.0,
)

# Set up column and register probes
cfg = (jtfne.Configuration()
    .runtime(seed=42, dtype="float32", duration_ms=1000.0, dt_ms=0.1)
    .column("V1_column", layers=["L2/3", "L4", "L5"], n=60)
    .cell_type_drives({"E": 6.5, "PV": 3.0})
    .set_emitter("izhikevich", "cortical_eig")
    .probes(["spikes", "LFP-proxy", "CSD-proxy"]))

# Construct model and apply paradigm conditions
model = jtfne.construct(cfg)

Figures

  • Figure 1: Sensory Event Timeline — Overview of expected tone, unexpected deviant tone, and omitted tone event windows.
  • Figure 2: Expected raster/rate — Spiking activity of regular standard conditions.
  • Figure 3: Unexpected deviant raster/rate — Spiking activity under rare deviant stimulus presentation.
  • Figure 4: Omission raster/rate — Network activity during expected tone window silence and post-omission window.
  • Figure 5: LFP/CSD-proxy Contrast — Extracellular-like profile comparison showing sensory vs omission profiles.

Coverage Boundary

This tutorial covers: - Declarative omission/oddball sensory configuration. - Stimulus-locked time segmentation and event window metadata. - Layer-specific standard, deviant, and omission comparisons.

This tutorial does NOT cover: - Physiological active inference validation. - Real biological metabolism modeling. - Maxwell/Poisson-based physical field solve equations. - Calibrated extracellular amplitude mapping.