How Lateral Connections and Spiking Dynamics May Separate Multiple Objects Moving Together

Abstract

Over successive stages, the ventral visual system of the primate brain develops neurons that respond selectively to particular objects or faces with translation, size and view invariance. The powerful neural representations found in Inferotemporal cortex form a remarkably rapid and robust basis for object recognition which belies the difficulties faced by the system when learning in natural visual environments. A central issue in understanding the process of biological object recognition is how these neurons learn to form separate representations of objects from complex visual scenes composed of multiple objects. We show how a one-layer competitive network comprised of ‘spiking’ neurons is able to learn separate transformation-invariant representations (exemplified by one-dimensional translations) of visual objects that are always seen together moving in lock-step, but separated in space. This is achieved by combining ‘Mexican hat’ functional lateral connectivity with cell firing-rate adaptation to temporally segment input representations of competing stimuli through anti-phase oscillations (perceptual cycles). These spiking dynamics are quickly and reliably generated, enabling selective modification of the feed-forward connections to neurons in the next layer through Spike-Time-Dependent Plasticity (STDP), resulting in separate translation-invariant representations of each stimulus. Variations in key properties of the model are investigated with respect to the network’s ability to develop appropriate input representations and subsequently output representations through STDP. Contrary to earlier rate-coded models of this learning process, this work shows how spiking neural networks may learn about more than one stimulus together without suffering from the ‘superposition catastrophe’. We take these results to suggest that spiking dynamics are key to understanding biological visual object recognition.

The emergence of polychronization and feature binding in a spiking neural network model of the primate ventral visual system.

“…• In some simulations, network performance is enhanced by incorporating multiple synaptic connections between each pair of pre-and postsynaptic neurons, where these connections have different axonal transmission delays. This permits STDP to strengthen just one (or a subset) of these connections in order to effectively select the functional transmission delay between the two neurons (Deger, Helias, Rotter, & Diesmann, 2012;Fares & Stepanyants, 2009). Using this underlying model architecture, the current study investigates the following hypotheses: emergence of polychronization, emergence of binding neurons, and "holographic principle" in the brain.…”

STDP in lateral connections creates category-based perceptual cycles for invariance learning with multiple stimuli

“…3.1 , the network was a single layer of 512 conductance-based leaky integrate-and-fire (gLIF) excitatory pyramidal neurons interconnected with plastic lateral excitatory connections initialised to zero strength. Unlike the simulations of Evans and Stringer ( 2013 ), these lateral connections were modified through learning rather than imposing a fixed ‘Mexican Hat’ profile upon them. There was also a separate pool of 128 inhibitory interneurons with fixed strength lateral connections to and from each of the excitatory cells.…”

“…Previously, we demonstrated how biologically plausible properties of a spiking neural network may provide additional ways to overcome the superposition catastrophe in this way. This involved desynchronising the volleys of spikes representing each stimulus with respect to each other through hard-wired lateral connections and cell firing-rate adaptation in a competitive network (Evans and Stringer 2013 ). This ‘Mexican hat’ lateral connectivity profile (consisting of short-range excitation with long-range/global inhibition) proved to be an effective mechanism for desynchronising spatially separate representations.…”

“…This paper explores the effects of early exposure to stimuli in building psychological categories of objects to aid the process of scene segmentation. The novel step in this paper, compared to the model of Evans and Stringer ( 2013 ), is to introduce plasticity into the lateral excitatory–excitatory ( ) connections of the first neuronal layer. Rather than associating the common features of a category of stimuli in the feed-forward synapses between layers, the prototypical features may instead be grouped through a strengthening of the lateral connections between them.…”

“…Conversely, the spike volley of the other stimulus should fall outside of this LTP time window, instead coming after the output spikes and subjecting the corresponding synapses to LTD. This should allow transformation-invariant representations to form individually for the translating stimuli through the CT learning mechanism, as explored in earlier work (Evans and Stringer 2012 , 2013 ).…”