Pulse Modulated Neural Nets with Adaptive Temporal Windows for Enhanced Video Perception

Abstract

Greg Passmore

A key challenge in modern artificial intelligence is the limited awareness of temporal dynamics. This is especially evident in AI's struggles with video animation and interpreting sequential frames or time-dependent data. This paper presents a new method for modulating neuron spikes to synchronize with incoming frames, addressing the temporal limitations of voltage AI models.

This approach involves a specialized class of neuron, similar to Wolfgang Maass's liquid state machine concept. However, our model differs significantly in its synchronization strategy. Neurons are synchronized with frame import, incorporating a flexible time window that enables spike modulation ranging from 1 to 1024 spikes per frame. This adaptability is crucial for accurately processing and interpreting time-based data.

The main focus of this paper is to explore the operational principles of these specialized neurons. While the wiring and architectural design of these neurons will be discussed in a future article, this paper aims to establish the foundational understanding of how these neurons function individually and in response to temporal data inputs.

By synchronizing neuron spikes with frame inputs and allowing for variable spike modulation, this method aims to enhance AI's ability to handle time-based data. The proposed neuron model represents a step towards more advanced and temporally aware AI systems, potentially unlocking new capabilities in video animation and other time-sensitive applications.

PDF