We demonstrate that a randomly connected attractor network with dynamic synapses can discriminate between similar sequences containing multiple stimuli suggesting such networks provide a general basis for neural computations in the brain. The network contains units representing assemblies of pools of neurons, with preferentially strong recurrent excitatory connections rendering each unit bi-stable. Weak interactions between units leads to a multiplicity of attractor states, within which information can persist beyond stimulus offset. When a new stimulus arrives, the prior state of the network impacts the encoding of the incoming information, with short-term synaptic depression ensuring an itinerancy between sets of active units. We assess the ability of such a network to encode the identity of sequences of stimuli, so as to provide a template for sequence recall, or decisions based on accumulation of evidence. Across a range of parameters, such networks produce the primacy (better final encoding of the earliest stimuli) and recency (better final encoding of the latest stimuli) observed in human recall data and can retain the information needed to make a binary choice based on total number of presentations of a specific stimulus. Similarities and differences in the final states of the network produced by different sequences lead to predictions of specific errors that could arise when an animal or human subject generalizes from training data, when the training data comprises a subset of the entire stimulus repertoire. We suggest that such networks can provide the general purpose computational engines needed for us to solve many cognitive tasks. Electronic supplementary material The online version of this article (10.1007/s10827-019-00717-5) contains supplementary material, which is available to authorized users.
In the classical view of economic choices, subjects make rational decisions evaluating the costs and benefits of options in order to maximize their overall income. Nonetheless, subjects often fail to reach optimal outcomes. The overt value of an option drives the direction of decisions, but covert factors such as emotion and sensitivity to sunk cost are thought to drive the observed deviations from optimality. Many questions remain to be answered as to (1) which contexts contribute the most to deviation from an optimal solution; and (2) the extent of these effects. In order to tackle these questions, we devised a decision-making task for mice, in which cost and benefit parameters could be independently and flexibly adjusted and for which a tractable optimal solution was known. Comparing mouse behavior with this optimal solution across parameter settings revealed that the factor most strongly contributing to suboptimal performance was the cost parameter. The quantification of sensitivity to sunk cost, a covert factor implicated in our task design, revealed it as another contributor to reduced optimality. In one condition where the large reward option was particularly unattractive and the small reward cost was low, the sensitivity to sunk cost and the cost-led suboptimality almost vanished. In this regime and this regime only, mice could be viewed as close to rational (here, ‘rational’ refers to a state in which an animal makes decisions basing on objective valuation, not covert factors). Taken together, our results suggest that “rationality” is a task-specific construct even in mice.
The decision of whether to continue with a current action or to stop and consider alternatives is ever present in the life of an animal. Such continuous-time decision making lies at the heart of food preference tests whose outcomes are typically quantified by a single variable, the total amount consumed. However, the dynamics that give rise to such a quantity in terms of durations of bouts of sampling at a stimulus before pauses, and the impact of alternative stimuli on those bout durations and subsequent actions following a pause, can contain a richness of behavior that is not captured in a single palatability measure. Here we carry out multiple analyses of these dynamics, with a particular focus on assessing how the hedonic value of one taste stimulus impacts the behavior of a rat sampling a second taste stimulus during a preference test. We find evidence for an explicit competitive interaction between bout durations, such that the more palatable a stimulus the longer the bout durations when the rat samples the stimulus and the shorter the bout durations at the alternative. Such competition is reproduced in a model of a neural circuit that could underlie the continuous decision of when to end a sampling bout. We find that the competitive impact on bout durations is relatively short-lived whereas a competitive impact on the choice of which stimulus to approach following a pause persists. Such a discrepancy in the timescales for the decay of the impact of the alternative stimulus suggests different neural processes are involved in the choice of which stimulus to approach versus the choice of how long to sample from it. Since these two choices together combine to determine net consumption and therefore the inferred palatability or preference of a gustatory stimulus, our results suggest that palatability is not a unitary quantity but the result of at least two distinct, context-dependent neural processes.
Food or taste preference tests are analogous to naturalistic decisions in which the animal selects which stimuli to sample and for how long to sample them. The data acquired in such tests, the relative amounts of the alternative stimuli that are sampled and consumed, indicate the preference for each. While such preferences are typically recorded as a single quantity, an analysis of the ongoing sampling dynamics producing the preference can reveal otherwise hidden aspects of the decision-making process that depend on its underlying neural circuit mechanisms. Here, we perform a dynamic analysis of two factors that give rise to preferences in a two-alternative task, namely the distribution of durations of sampling bouts of each stimulus and the likelihood of returning to the same stimulus or switching to the alternative-that is, the transition probability-following each bout. The results of our analysis support a specific computational model of decision making whereby an exponential distribution of bout durations has a mean that is positively correlated with the palatability of that stimulus but also negatively correlated with the palatability of the alternative. This impact of the alternative stimulus on the distribution of bout durations decays over a timescale of tens of seconds, even though the memory of the alternative stimulus lasts far longer-long enough to impact the transition probabilities upon ending bouts. Together, our findings support a state transition model for bout durations and suggest a separate memory mechanism for stimulus selection.
¶ These authors contributed equally to this work. AbstractWe demonstrate the ability of a randomly connected attractor network with dynamic synapses to discriminate between similar sequences containing multiple stimuli and suggest such networks provide a general basis for neural computations in the brain. The network is based on units representing assemblies of pools of neurons, with preferentially strong recurrent excitatory connections within each unit. Such excitatory feedback to a unit can generate bistability, though in many networks only under conditions of net excitatory input from other units. Weak interactions between units leads to a multiplicity of attractor states, within which information can persist beyond stimulus offset. When a new stimulus arrives, the prior state of the network impacts the encoding of the incoming information, with short-term synaptic depression ensuring an itinerancy between sets of active units. We assess the ability of such a network to encode the identity of sequences of stimuli, so as to provide a template for sequence recall, or decisions based on accumulation of evidence. Across a range of parameters, such networks produce the primacy (better final encoding of the earliest stimuli) and recency (better final encoding of the latest stimuli) observed in human recall data and can retain the information needed to make a binary choice based on total number of presentations of a specific stimulus. Similarities and differences in the final states of the network produced by different sequences lead to predictions of specific errors that could arise when an animal or human subject generalizes from training data, when the training data comprises a subset of the entire stimulus repertoire. We suggest that such networks can provide the robust general purpose computational engines needed for us to solve many cognitive tasks.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
customersupport@researchsolutions.com
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.
Copyright © 2025 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.