Neural attractor dynamics in object recognition

Vector Symbolic Finite State Machines in Attractor Neural Networks

Hopfield attractor networks are robust distributed models of human memory, but they lack a general mechanism for effecting state-dependent attractor transitions in response to input. We propose construction rules such that an attractor network may implement an arbitrary finite state machine (FSM), where states and stimuli are represented by high-dimensional random vectors and all state transitions are enacted by the attractor network's dynamics. Numerical simulations show the capacity of the model, in terms of the maximum size of implementable FSM, to be linear in the size of the attractor network for dense bipolar state vectors and approximately quadratic for sparse binary state vectors. We show that the model is robust to imprecise and noisy weights, and so a prime candidate for implementation with high-density but unreliable devices. By endowing attractor networks with the ability to emulate arbitrary FSMs, we propose a plausible path by which FSMs could exist as a distributed computational primitive in biological neural networks.

Multistability in neural systems with random cross-connections

Neural circuits with multiple discrete attractor states could support a variety of cognitive tasks according to both empirical data and model simulations. We assess the conditions for such multistability in neural systems using a firing rate model framework, in which clusters of similarly responsive neurons are represented as single units, which interact with each other through independent random connections. We explore the range of conditions in which multistability arises via recurrent input from other units while individual units, typically with some degree of self-excitation, lack sufficient self-excitation to become bistable on their own. We find many cases of multistability-defined as the system possessing more than one stable fixed point-in which stable states arise via a network effect, allowing subsets of units to maintain each others' activity because their net input to each other when active is sufficiently positive. In terms of the strength of within-unit self-excitation and standard deviation of random cross-connections, the region of multistability depends on the response function of units. Indeed, multistability can arise with zero self-excitation, purely through zero-mean random cross-connections, if the response function rises supralinearly at low inputs from a value near zero at zero input. We simulate and analyze finite systems, showing that the probability of multistability can peak at intermediate system size, and connect with other literature analyzing similar systems in the infinite-size limit. We find regions of multistability with a bimodal distribution for the number of active units in a stable state. Finally, we find evidence for a log-normal distribution of sizes of attractor basins, which produces Zipf's Law when enumerating the proportion of trials within which random initial conditions lead to a particular stable state of the system.

Multistability in neural systems with random cross-connections

Neural circuits with multiple discrete attractor states could support a variety of cognitive tasks according to both empirical data and model simulations. We assess the conditions for such multistability in neural systems, using a firing-rate model framework, in which clusters of neurons with net self-excitation are represented as units, which interact with each other through random connections. We focus on conditions in which individual units lack sufficient self-excitation to become bistable on their own. Rather, multistability can arise via recurrent input from other units as a network effect for subsets of units, whose net input to each other when active is sufficiently positive to maintain such activity. In terms of the strength of within-unit self-excitation and standard-deviation of random cross-connections, the region of multistability depends on the firing-rate curve of units. Indeed, bistability can arise with zero self-excitation, purely through zero-mean random cross-connections, if the firing-rate curve rises supralinearly at low inputs from a value near zero at zero input. We simulate and analyze finite systems, showing that the probability of multistability can peak at intermediate system size, and connect with other literature analyzing similar systems in the infinite-size limit. We find regions of multistability with a bimodal distribution for the number of active units in a stable state. Finally, we find evidence for a log-normal distribution of sizes of attractor basins, which can appear as Zipf's Law when sampled as the proportion of trials within which random initial conditions lead to a particular stable state of the system.

Belief traps: Tackling the inertia of harmful beliefs

Beliefs can be highly resilient in the sense that they are not easily abandoned in the face of counterevidence. This has the advantage of guiding consistent behavior and judgments but may also have destructive consequences for individuals, nature, and society. For instance, pathological beliefs can sustain psychiatric disorders, the belief that rhinoceros horn is an aphrodisiac may drive a species extinct, beliefs about gender or race may fuel discrimination, and belief in conspiracy theories can undermine democracy. Here, we present a unifying framework of how self-amplifying feedbacks shape the inertia of beliefs on levels ranging from neuronal networks to social systems. Sustained exposure to counterevidence can destabilize rigid beliefs but requires organized rational override as in cognitive behavioral therapy for pathological beliefs or institutional control of discrimination to reduce racial biases. Black-and-white thinking is a major risk factor for the formation of resilient beliefs associated with psychiatric disorders as well as prejudices and conspiracy thinking. Such dichotomous thinking is characteristic of a lack of cognitive resources, which may be exacerbated by stress. This could help explain why conspiracy thinking and psychiatric disorders tend to peak during crises. A corollary is that addressing social factors such as poverty, social cleavage, and lack of education may be the most effective way to prevent the emergence of rigid beliefs, and thus of problems ranging from psychiatric disorders to prejudices, conspiracy theories, and posttruth politics.

A model of naturalistic decision making in preference tests

Decisions as to whether to continue with an ongoing activity or to switch to an alternative are a constant in an animal’s natural world, and in particular underlie foraging behavior and performance in food preference tests. Stimuli experienced by the animal both impact the choice and are themselves impacted by the choice, in a dynamic back and forth. Here, we present model neural circuits, based on spiking neurons, in which the choice to switch away from ongoing behavior instantiates this back and forth, arising as a state transition in neural activity. We analyze two classes of circuit, which differ in whether state transitions result from a loss of hedonic input from the stimulus (an “entice to stay” model) or from aversive stimulus-input (a “repel to leave” model). In both classes of model, we find that the mean time spent sampling a stimulus decreases with increasing value of the alternative stimulus, a fact that we linked to the inclusion of depressing synapses in our model. The competitive interaction is much greater in “entice to stay” model networks, which has qualitative features of the marginal value theorem, and thereby provides a framework for optimal foraging behavior. We offer suggestions as to how our models could be discriminatively tested through the analysis of electrophysiological and behavioral data.

Many decisions are of the ilk of whether to continue sampling a stimulus or to switch to an alternative, a key feature of foraging behavior. We produce two classes of model for such stay-switch decisions, which differ in how decisions to switch stimuli can arise. In an “entice-to-stay” model, a reduction in the necessary positive stimulus input causes switching decisions. In a “repel-to-leave” model, a rise in aversive stimulus input produces a switch decision. We find that in tasks where the sampling of one stimulus follows another, adaptive biological processes arising from a highly hedonic stimulus can reduce the time spent at the following stimulus, by up to ten-fold in the “entice-to-stay” models. Along with potentially observable behavioral differences that could distinguish the classes of networks, we also found signatures in neural activity, such as oscillation of neural firing rates and a rapid change in rates preceding the time of choice to leave a stimulus. In summary, our model findings lead to testable predictions and suggest a neural circuit-based framework for explaining foraging choices.

Precision multidimensional neural population code recovered from single intracellular recordings

Neurons in sensory cortices are more naturally and deeply integrated than any current neural population recording tools (e.g. electrode arrays, fluorescence imaging). Two concepts facilitate efforts to observe population neural code with single-cell recordings. First, even the highest quality single-cell recording studies find a fraction of the stimulus information in high-dimensional population recordings. Finding any of this missing information provides proof of principle. Second, neurons and neural populations are understood as coupled nonlinear differential equations. Therefore, fitted ordinary differential equations provide a basis for single-trial single-cell stimulus decoding. We obtained intracellular recordings of fluctuating transmembrane current and potential in mouse visual cortex during stimulation with drifting gratings. We use mean deflection from baseline when comparing to prior single-cell studies because action potentials are too sparse and the deflection response to drifting grating stimuli (e.g. tuning curves) are well studied. Equation-based decoders allowed more precise single-trial stimulus discrimination than tuning-curve-base decoders. Performance varied across recorded signal types in a manner consistent with population recording studies and both classification bases evinced distinct stimulus-evoked phases of population dynamics, providing further corroboration. Naturally and deeply integrated observations of population dynamics would be invaluable. We offer proof of principle and a versatile framework.

Attractor-state itinerancy in neural circuits with synaptic depression

Neural populations with strong excitatory recurrent connections can support bistable states in their mean firing rates. Multiple fixed points in a network of such bistable units can be used to model memory retrieval and pattern separation. The stability of fixed points may change on a slower timescale than that of the dynamics due to short-term synaptic depression, leading to transitions between quasi-stable point attractor states in a sequence that depends on the history of stimuli. To better understand these behaviors, we study a minimal model, which characterizes multiple fixed points and transitions between them in response to stimuli with diverse time- and amplitude-dependencies. The interplay between the fast dynamics of firing rate and synaptic responses and the slower timescale of synaptic depression makes the neural activity sensitive to the amplitude and duration of square-pulse stimuli in a nontrivial, history-dependent manner. Weak cross-couplings further deform the basins of attraction for different fixed points into intricate shapes. We find that while short-term synaptic depression can reduce the total number of stable fixed points in a network, it tends to strongly increase the number of fixed points visited upon repetitions of fixed stimuli. Our analysis provides a natural explanation for the system's rich responses to stimuli of different durations and amplitudes while demonstrating the encoding capability of bistable neural populations for dynamical features of incoming stimuli.

A statistical method for analyzing and comparing spatiotemporal cortical activation patterns

Information in the cortex is encoded in spatiotemporal patterns of neuronal activity, but the exact nature of that code still remains elusive. While onset responses to simple stimuli are associated with specific loci in cortical sensory maps, it is completely unclear how the information about a sustained stimulus is encoded that is perceived for minutes or even longer, when discharge rates have decayed back to spontaneous levels. Using a newly developed statistical approach (multidimensional cluster statistics (MCS)) that allows for a comparison of clusters of data points in n-dimensional space, we here demonstrate that the information about long-lasting stimuli is encoded in the ongoing spatiotemporal activity patterns in sensory cortex. We successfully apply MCS to multichannel local field potential recordings in different rodent models and sensory modalities, as well as to human MEG and EEG data, demonstrating its universal applicability. MCS thus indicates novel ways for the development of powerful read-out algorithms of spatiotemporal brain activity that may be implemented in innovative brain-computer interfaces (BCI).

Itinerancy between attractor states in neural systems

Converging evidence from neural, perceptual and simulated data suggests that discrete attractor states form within neural circuits through learning and development. External stimuli may bias neural activity to one attractor state or cause activity to transition between several discrete states. Evidence for such transitions, whose timing can vary across trials, is best accrued through analyses that avoid any trial-averaging of data. One such method, hidden Markov modeling, has been effective in this context, revealing state transitions in many neural circuits during many tasks. Concurrently, modeling efforts have revealed computational benefits of stimulus processing via transitions between attractor states. This review describes the current state of the field, with comments on how its perceived limitations have been addressed.

Multisensory integration and cross-modal learning in synaesthesia: A unifying model

Recent research into synaesthesia has highlighted the role of learning, yet synaesthesia is clearly a genetic condition. Here we ask how can the idea that synaesthesia reflects innate, genetic differences be reconciled with models that suggest it is driven by learning. A number of lines of evidence suggest that synaesthesia relies on, or at least interacts with, processes of multisensory integration that are common across all people. These include multisensory activations that arise in early regions of the brain as well as feedback from longer-term cross-modal associations generated in memory. These cognitive processes may interact independently to influence the phenomenology of the synaesthetic experience, as well as the individual differences within particular types of synaesthesia. The theoretical framework presented here is consistent with both an innate difference as the fundamental driver of the condition of synaesthesia, and with experiential and semantic influences on the eventual phenotype that emerges. In particular, it proposes that the internally generated synaesthetic percepts are treated similarly to other sensory information as the brain is learning the multisensory attributes of objects and developing cross-modal associations that merge in the concept of the object.

Recollection and Familiarity Exhibit Dissociable Similarity Gradients: A Test of the Complementary Learning Systems Model

Memory can often be triggered by retrieval cues that are quite different from the originally encoded events, but how different memory processes respond to variations in cue–target similarity is poorly understood. We begin by presenting simulations using a neurocomputational model of recognition memory (i.e., the complementary learning systems model), which proposes that the hippocampus supports recollection of associative information whereas the surrounding cortex supports assessments of item familiarity. The simulations showed that increases in the similarity between retrieval cues and learned items led to relatively linear increases in a cortex-based memory signal but led to steeper and more thresholded increases in the hippocampal signal. We then tested the predictions of the model by examining the effects of varying cue–target similarity in two recognition memory experiments in which participants studied a list of computer-generated faces and then, at test, gave confidence and remember/know responses to morphed faces. In both experiments, as cue–target similarity was increased, familiarity-based recognition increased in a gradual and relatively linear fashion, whereas recollection showed significantly steeper gradients. The results show that recollection and familiarity exhibit distinct similarity functions in recognition memory that correspond with predicted retrieval dynamics of the hippocampus and cortex, respectively.

Early recurrent feedback facilitates visual object recognition under challenging conditions

Standard models of the visual object recognition pathway hold that a largely feedforward process from the retina through inferotemporal cortex leads to object identification. A subsequent feedback process originating in frontoparietal areas through reciprocal connections to striate cortex provides attentional support to salient or behaviorally-relevant features. Here, we review mounting evidence that feedback signals also originate within extrastriate regions and begin during the initial feedforward process. This feedback process is temporally dissociable from attention and provides important functions such as grouping, associational reinforcement, and filling-in of features. Local feedback signals operating concurrently with feedforward processing are important for object identification in noisy real-world situations, particularly when objects are partially occluded, unclear, or otherwise ambiguous. Altogether, the dissociation of early and late feedback processes presented here expands on current models of object identification, and suggests a dual role for descending feedback projections.

Recurrent Processing during Object Recognition

How does the brain learn to recognize objects visually, and perform this difficult feat robustly in the face of many sources of ambiguity and variability? We present a computational model based on the biology of the relevant visual pathways that learns to reliably recognize 100 different object categories in the face of naturally occurring variability in location, rotation, size, and lighting. The model exhibits robustness to highly ambiguous, partially occluded inputs. Both the unified, biologically plausible learning mechanism and the robustness to occlusion derive from the role that recurrent connectivity and recurrent processing mechanisms play in the model. Furthermore, this interaction of recurrent connectivity and learning predicts that high-level visual representations should be shaped by error signals from nearby, associated brain areas over the course of visual learning. Consistent with this prediction, we show how semantic knowledge about object categories changes the nature of their learned visual representations, as well as how this representational shift supports the mapping between perceptual and conceptual knowledge. Altogether, these findings support the potential importance of ongoing recurrent processing throughout the brain's visual system and suggest ways in which object recognition can be understood in terms of interactions within and between processes over time.

Deformation of attractor landscape via cholinergic presynaptic modulations: a computational study using a phase neuron model

Corticopetal acetylcholine (ACh) is released transiently from the nucleus basalis of Meynert (NBM) into the cortical layers and is associated with top-down attention. Recent experimental data suggest that this release of ACh disinhibits layer 2/3 pyramidal neurons (PYRs) via muscarinic presynaptic effects on inhibitory synapses. Together with other possible presynaptic cholinergic effects on excitatory synapses, this may result in dynamic and temporal modifications of synapses associated with top-down attention. However, the system-level consequences and cognitive relevance of such disinhibitions are poorly understood. Herein, we propose a theoretical possibility that such transient modifications of connectivity associated with ACh release, in addition to top-down glutamatergic input, may provide a neural mechanism for the temporal reactivation of attractors as neural correlates of memories. With baseline levels of ACh, the brain returns to quasi-attractor states, exhibiting transitive dynamics between several intrinsic internal states. This suggests that top-down attention may cause the attention-induced deformations between two types of attractor landscapes: the quasi-attractor landscape (Q-landscape, present under low-ACh, non-attentional conditions) and the attractor landscape (A-landscape, present under high-ACh, top-down attentional conditions). We present a conceptual computational model based on experimental knowledge of the structure of PYRs and interneurons (INs) in cortical layers 1 and 2/3 and discuss the possible physiological implications of our results.

Emergence in the central nervous system

"Emergence" is an idea that has received much attention in consciousness literature, but it is difficult to find characterizations of that concept which are both specific and useful. I will precisely define and characterize a type of epistemic ("weak") emergence and show that it is a property of some neural circuits throughout the CNS, on micro-, meso- and macroscopic levels. I will argue that possession of this property can result in profoundly altered neural dynamics on multiple levels in cortex and other systems. I will first describe emergent neural entities (ENEs) abstractly. I will then show how ENEs function specifically and concretely, and demonstrate some implications of this type of emergence for the CNS.

Self-organization of multiple spatial and context memories in the hippocampus

One obstacle to understanding the exact processes unfolding inside the hippocampus is that it is still difficult to clearly define what the hippocampus actually does, at the system level. Associated for a long time with the formation of episodic and semantic memories, and with their temporary storage, the hippocampus is also regarded as a structure involved in spatial navigation. These two independent perspectives on the hippocampus are not necessarily exclusive: proposals have been put forward to make them fit into the same conceptual frame. We review both approaches and argue that three critical developments need consideration: (a) recordings of neuronal activity in rodents, revealing beautiful spatial codes expressed in entorhinal cortex, upstream of the hippocampus; (b) comparative behavioral results suggesting, in an evolutionary perspective, qualitative similarity of function across homologous structures with a distinct internal organization; (c) quantitative measures of information, shifting the focus from who does what to how much each neuronal population expresses each code. These developments take the hippocampus away from philosophical discussions of all-or-none cause-effect relations, and into the quantitative mainstream of modern neural science.