Generalization in Sensorimotor Networks Configured with Natural Language Instructions

Riveland, Reidar; Pouget, Alexandre

doi:10.1101/2022.02.22.481293

Cited by 7 publications

(6 citation statements)

References 75 publications

(139 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The study of cognitive tasks through simulations in artificial networks has led to substantial advances in understanding neural computation in the past decade [9][10][11][12][13][14][15][16][17][18] . However, researchers typically trained artificial neural networks to perform single tasks in isolation, with few exceptions [19][20][21][22] , somewhat limiting the insights into biological neural circuits that perform many tasks. One exception to this trend is Yang et al 2019 20 , in which the authors trained a single network to perform twenty related tasks and thereby identified clustered representations in state space that supported task compositionality.…”

Section: Introductionmentioning

confidence: 99%

Flexible multitask computation in recurrent networks utilizes shared dynamical motifs

Driscoll

Shenoy

Sussillo

2022

Preprint

View full text Add to dashboard Cite

Flexible computation is a hallmark of intelligent behavior. Yet, little is known about how neural networks contextually reconfigure for different computations. Humans are able to perform a new task without extensive training, presumably through the composition of elementary processes that were previously learned. Cognitive scientists have long hypothesized the possibility of a compositional neural code, where complex neural computations are made up of constituent components; however, the neural substrate underlying this structure remains elusive in biological and artificial neural networks. Here we identified an algorithmic neural substrate for compositional computation through the study of multitasking artificial recurrent neural networks. Dynamical systems analyses of networks revealed learned computational strategies that mirrored the modular subtask structure of the task-set used for training. Dynamical motifs such as attractors, decision boundaries and rotations were reused across different task computations. For example, tasks that required memory of a continuous circular variable repurposed the same ring attractor. We show that dynamical motifs are implemented by clusters of units and are reused across different contexts, allowing for flexibility and generalization of previously learned computation. Lesioning these clusters resulted in modular effects on network performance: a lesion that destroyed one dynamical motif only minimally perturbed the structure of other dynamical motifs. Finally, modular dynamical motifs could be reconfigured for fast transfer learning. After slow initial learning of dynamical motifs, a subsequent faster stage of learning reconfigured motifs to perform novel tasks. This work contributes to a more fundamental understanding of compositional computation underlying flexible general intelligence in neural systems. We present a conceptual framework that establishes dynamical motifs as a fundamental unit of computation, intermediate between the neuron and the network. As more whole brain imaging studies record neural activity from multiple specialized systems simultaneously, the framework of dynamical motifs will guide questions about specialization and generalization across brain regions.

show abstract

Section: Introductionmentioning

confidence: 99%

Flexible multitask computation in recurrent networks utilizes shared dynamical motifs

Driscoll

Shenoy

Sussillo

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…Our work instantiates these concepts by learning by using gradient descent for latent updates. Similar use of gradient descent to update latent variables was proposed as model for schizophrenia by Yamashita and Tani (2012) and more recently to retrieve language instructions for cognitive tasks (Riveland and Pouget, 2022). Our work uses these methods as a component in a continually learning agent to learn disentangled representations.…”

Section: Related Workmentioning

confidence: 97%

Thalamus: a brain-inspired algorithm for biologically-plausible continual learning and disentangled representations

Hummos¹

2022

Preprint

View full text Add to dashboard Cite

Animals thrive in a constantly changing environment and leverage the temporal structure to learn well-factorized causal representations. In contrast, traditional neural networks suffer from forgetting in changing environments and many methods have been proposed to limit forgetting with different trade-offs. Inspired by the brain thalamocortical circuit, we introduce a simple algorithm that uses optimization at inference time to generate internal representations of temporal context and to infer current context dynamically, allowing the agent to parse the stream of temporal experience into discrete events and organize learning about them. We show that a network trained on a series of tasks using traditional weight updates can infer tasks dynamically using gradient descent steps in the latent task embedding space (latent updates). We then alternate between the weight updates and the latent updates to arrive at Thalamus, a task-agnostic algorithm capable of discovering disentangled representations in a stream of unlabeled tasks using simple gradient descent. On a continual learning benchmark, it achieves competitive end average accuracy and demonstrates knowledge transfer. After learning a subset of tasks it can generalize to unseen tasks as they become reachable within the well-factorized latent space, through one-shot latent updates. The algorithm meets many of the desiderata of an ideal continually learning agent in open-ended environments, and its simplicity suggests fundamental computations in circuits with abundant feedback control loops such as the thalamocortical circuits in the brain.Preprint. Under review.

show abstract

“…Several recent studies in neuroscience have applied analytic tools to identify the neural basis of rapid generalization in biological neural networks. Such studies employed various measures -crosscondition generalization [2,40,8,4], state-space projections of task-related compositional codes [51,43,25], and Parallelism Score [2] -to quantify the generalizability and abstraction of representations. Prior work in neuroscience has primarily evaluated compositionality in limited context settings (e.g., up to 10 contexts), or without manipulating different types of features (e.g., higher-order vs. sensory/motor features).…”

Section: Related Workmentioning

confidence: 99%

Compositional generalization through abstract representations in human and artificial neural networks

Ito¹,

Klinger²,

Schultz³

et al. 2022

Preprint

View full text Add to dashboard Cite

Humans have a remarkable ability to rapidly generalize to new tasks that is difficult to reproduce in artificial learning systems. Compositionality has been proposed as a key mechanism supporting generalization in humans, but evidence of its neural implementation and impact on behavior is still scarce. Here we study the computational properties associated with compositional generalization in both humans and artificial neural networks (ANNs) on a highly compositional task. First, we identified behavioral signatures of compositional generalization in humans, along with their neural correlates using whole-cortex functional magnetic resonance imaging (fMRI) data. Next, we designed pretraining paradigms aided by a procedure we term primitives pretraining to endow compositional task elements into ANNs. We found that ANNs with this prior knowledge had greater correspondence with human behavior and neural compositional signatures. Importantly, primitives pretraining induced abstract internal representations, excellent zero-shot generalization, and sample-efficient learning. Moreover, it gave rise to a hierarchy of abstract representations that matched human fMRI data, where sensory rule abstractions emerged in early sensory areas, and motor rule abstractions emerged in later motor areas. Our findings give empirical support to the role of compositional generalization in human behavior, implicate abstract representations as its neural implementation, and illustrate that these representations can be embedded into ANNs by designing simple and efficient pretraining procedures.

show abstract

Generalization in Sensorimotor Networks Configured with Natural Language Instructions

Cited by 7 publications

References 75 publications

Flexible multitask computation in recurrent networks utilizes shared dynamical motifs

Flexible multitask computation in recurrent networks utilizes shared dynamical motifs

Thalamus: a brain-inspired algorithm for biologically-plausible continual learning and disentangled representations

Compositional generalization through abstract representations in human and artificial neural networks

Contact Info

Product

Resources

About