snn-sequence-timing-replay-speed

name	snn-sequence-timing-replay-speed
description	Spiking Temporal Memory (sTM) model for learning sequence timing and flexible replay speed control - biologically plausible timing encoding via oscillatory modulation
trigger_words	["spiking neural network","sequence timing","replay speed","spTM temporal memory","oscillatory control","sequential processing","spatiotemporal patterns"]
activation_keywords	["sequence timing learning","replay speed control","oscillatory background","spatiotemporal encoding","element-specific timing","sparse representation"]
version	1.0.0
last_updated	2026-06-19T00:00:00.000Z
paper_source	arXiv:2605.22523
authors	Melissa Lober, Younes Bouhadjar, Markus Diesmann, Tom Tetzlaff
submitted	2026-05-21T00:00:00.000Z

Learning Sequence Timing and Control of Replay Speed in Networks of Spiking Neurons

Background

Processing sequential inputs is a fundamental brain function, underlying:

• Sensory perception
• Language processing
• Motor control

Challenge: Represent not only event order, but also their precise timing.

Existing models can learn sequential structure but lack:

• Biologically plausible mechanisms for element-specific timing
• Flexible control of replay speed

Core Innovation: Spiking Temporal Memory (sTM) Model

Previous sTM Capabilities

• Each sequence element represented by small set of neurons firing synchronously
• Element identity encoded in set of active neurons (sequential context)
• Could learn order but not timing

New Contributions

1. Duration Encoding: Element-specific timing via sequential activation of neuronal populations
2. Wide Timescale Coverage: Encode sequences across diverse durations
3. Replay Speed Control: Oscillatory background inputs serve as clock signal
4. Biological Plausibility: Sparse spatiotemporal patterns encode elapsed time

Methodology

Timing Mechanism

Key insight: Duration of sequence elements represented by sequential activation of element-specific neuronal populations

Element A → Population A neurons fire in sequence
- Neuron 1 (time t0)
- Neuron 2 (time t0+δ)
- Neuron 3 (time t0+2δ)
...
Duration = N * δ (where N = number of neurons in population)

Replay Speed Control

Oscillatory background inputs act as:

• Clock signal for sequence replay
• Flexible speed modulation mechanism
• Correlated with EEG/LFP characteristics

Fast oscillations → Fast replay
Slow oscillations → Slow replay

Encoding Principles

• Elapsed time encoded by unique sparse spatiotemporal patterns
• Each moment has distinct neural signature
• No dedicated "time neurons" - timing emerges from population dynamics

Key Findings