Recurrent networks endowed with structural priors explain suboptimal animal behavior

Rapid, systematic updating of movement by accumulated decision evidence

Acting in the natural world requires not only deciding among multiple options but also converting decisions into motor commands. How the dynamics of decision formation influence the fine kinematics of response movement remains, however, poorly understood. Here we investigate how the accumulation of decision evidence shapes the response orienting trajectories in a task where freely-moving rats combine prior expectations and auditory information to select between two possible options. Response trajectories and their motor vigor are initially determined by the prior. Rats movements then incorporate sensory information in less than 100 ms after stimulus onset by accelerating or slowing depending on how much the stimulus supports their initial choice. When the stimulus evidence is in strong contradiction, rats change their mind and reverse their initial trajectory. Human subjects performing an equivalent task display a remarkably similar behavior. We encapsulate these results in a computational model that maps the decision variable onto the movement kinematics at discrete time points, capturing subjects' choices, trajectories and changes of mind. Our results show that motor responses are not ballistic. Instead, they are systematically and rapidly updated, as they smoothly unfold over time, by the parallel dynamics of the underlying decision process.

Emergent perceptual biases from state-space geometry in trained spiking recurrent neural networks

A stimulus held in working memory is perceived as contracted toward the average stimulus. This contraction bias has been extensively studied in psychophysics, but little is known about its origin from neural activity. By training recurrent networks of spiking neurons to discriminate temporal intervals, we explored the causes of this bias and how behavior relates to population firing activity. We found that the trained networks exhibited animal-like behavior. Various geometric features of neural trajectories in state space encoded warped representations of the durations of the first interval modulated by sensory history. Formulating a normative model, we showed that these representations conveyed a Bayesian estimate of the interval durations, thus relating activity and behavior. Importantly, our findings demonstrate that Bayesian computations already occur during the sensory phase of the first stimulus and persist throughout its maintenance in working memory, until the time of stimulus comparison.

Flexible multitask computation in recurrent networks utilizes shared dynamical motifs

Flexible computation is a hallmark of intelligent behavior. However, little is known about how neural networks contextually reconfigure for different computations. In the present work, we identified an algorithmic neural substrate for modular computation through the study of multitasking artificial recurrent neural networks. Dynamical systems analyses revealed learned computational strategies mirroring the modular subtask structure of the training task set. Dynamical motifs, which are recurring patterns of neural activity that implement specific computations through dynamics, such as attractors, decision boundaries and rotations, were reused across tasks. For example, tasks requiring memory of a continuous circular variable repurposed the same ring attractor. We showed that dynamical motifs were implemented by clusters of units when the unit activation function was restricted to be positive. Cluster lesions caused modular performance deficits. Motifs were reconfigured for fast transfer learning after an initial phase of learning. This work establishes dynamical motifs as a fundamental unit of compositional computation, intermediate between neuron and network. As whole-brain studies simultaneously record activity from multiple specialized systems, the dynamical motif framework will guide questions about specialization and generalization.

Synergistic information supports modality integration and flexible learning in neural networks solving multiple tasks

Striking progress has been made in understanding cognition by analyzing how the brain is engaged in different modes of information processing. For instance, so-called synergistic information (information encoded by a set of neurons but not by any subset) plays a key role in areas of the human brain linked with complex cognition. However, two questions remain unanswered: (a) how and why a cognitive system can become highly synergistic; and (b) how informational states map onto artificial neural networks in various learning modes. Here we employ an information-decomposition framework to investigate neural networks performing cognitive tasks. Our results show that synergy increases as networks learn multiple diverse tasks, and that in tasks requiring integration of multiple sources, performance critically relies on synergistic neurons. Overall, our results suggest that synergy is used to combine information from multiple modalities-and more generally for flexible and efficient learning. These findings reveal new ways of investigating how and why learning systems employ specific information-processing strategies, and support the principle that the capacity for general-purpose learning critically relies on the system's information dynamics.

Performance errors during rodent learning reflect a dynamic choice strategy

Humans, even as infants, use cognitive strategies, such as exploration and hypothesis testing, to learn about causal interactions in the environment. In animal learning studies, however, it is challenging to disentangle higher-order behavioral strategies from errors arising from imperfect task knowledge or inherent biases. Here, we trained head-fixed mice on a wheel-based auditory two-choice task and exploited the intra- and inter-animal variability to understand the drivers of errors during learning. During learning, performance errors are dominated by a choice bias, which, despite appearing maladaptive, reflects a dynamic strategy. Early in learning, mice develop an internal model of the task contingencies such that violating their expectation of reward on correct trials (by using short blocks of non-rewarded "probe" trials) leads to an abrupt shift in strategy. During the probe block, mice behave more accurately with less bias, thereby using their learned stimulus-action knowledge to test whether the outcome contingencies have changed. Despite having this knowledge, mice continued to exhibit a strong choice bias during reinforced trials. This choice bias operates on a timescale of tens to hundreds of trials with a dynamic structure, shifting between left, right, and unbiased epochs. Biased epochs also coincided with faster motor kinematics. Although bias decreased across learning, expert mice continued to exhibit short bouts of biased choices interspersed with longer bouts of unbiased choices and higher performance. These findings collectively suggest that during learning, rodents actively probe their environment in a structured manner to refine their decision-making and maintain long-term flexibility.

Learning to learn: Single session acquisition of new rules by freely moving mice

Learning from examples and adapting to new circumstances are fundamental attributes of human cognition. However, it is unclear what conditions allow for fast and successful learning, especially in nonhuman subjects. To determine how rapidly freely moving mice can learn a new discrimination criterion (DC), we design a two-alternative forced-choice visual discrimination paradigm in which the DCs governing the task can change between sessions. We find that experienced animals can learn a new DC after being exposed to only five training and three testing trials. The propensity for single session learning improves over time and is accurately predicted based on animal experience and criterion difficulty. After establishing the procedural learning of a paradigm, mice continuously improve their performance in new circumstances. Thus, mice learn to learn.

Rapid, systematic updating of movement by accumulated decision evidence

Acting in the natural world requires not only deciding among multiple options but also converting decisions into motor commands. How the dynamics of decision formation influence the fine kinematics of response movement remains, however, poorly understood. Here we investigate how the accumulation of decision evidence shapes the response orienting trajectories in a task where freely-moving rats combine prior expectations and auditory information to select between two possible options. Response trajectories and their motor vigor are initially determined by the prior. Rats movements then incorporate sensory information as early as 60 ms after stimulus onset by accelerating or slowing depending on how much the stimulus supports their initial choice. When the stimulus evidence is in strong contradiction, rats change their mind and reverse their initial trajectory. Human subjects performing an equivalent task display a remarkably similar behavior. We encapsulate these results in a computational model that, by mapping the decision variable onto the movement kinematics at discrete time points, captures subjects' choices, trajectories and changes of mind. Our results show that motor responses are not ballistic. Instead, they are systematically and rapidly updated, as they smoothly unfold over time, by the parallel dynamics of the underlying decision process.

Trial-history biases in evidence accumulation can give rise to apparent lapses in decision-making

Trial history biases and lapses are two of the most common suboptimalities observed during perceptual decision-making. These suboptimalities are routinely assumed to arise from distinct processes. However, previous work has suggested that they covary in their prevalence and that their proposed neural substrates overlap. Here we demonstrate that during decision-making, history biases and apparent lapses can both arise from a common cognitive process that is optimal under mistaken beliefs that the world is changing i.e. nonstationary. This corresponds to an accumulation-to-bound model with history-dependent updates to the initial state of the accumulator. We test our model's predictions about the relative prevalence of history biases and lapses, and show that they are robustly borne out in two distinct decision-making datasets of male rats, including data from a novel reaction time task. Our model improves the ability to precisely predict decision-making dynamics within and across trials, by positing a process through which agents can generate quasi-stochastic choices.

Neural networks: Explaining animal behavior with prior knowledge of the world

Animal behavior is both facilitated and constrained by innate knowledge and previous experience of the world. A new study, exploiting the power of recurrent neural networks, has revealed the existence of such structural priors and their impact on animal behavior.