Compositional generalization through abstract representations in human and artificial neural networks

Ito, Takuya; Klinger, Tim; Schultz, Douglas H.; Murray, John D.; Cole, Jennifer; Rigotti, Mattia

doi:10.48550/arxiv.2209.07431

Cited by 7 publications

(10 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…By contrast, we observed that neural manifolds representing space were highly aligned across contexts in most brain regions. This resembles the “neural structure alignment” that has recently been reported to accompany decision tasks in both humans and monkeys, whereby contexts sharing common structure are represented with parallel neural geometries, potentially because this allows a decoder trained in one context to be generalised to the other [46–51].…”

Section: Discussionmentioning

confidence: 99%

Goal-seeking compresses neural codes for space in the human hippocampus and orbitofrontal cortex

Muhle-Karbe

Sheahan

Pezzulo

et al. 2023

Preprint

View full text Add to dashboard Cite

Humans can navigate flexibly to meet their goals. Here, we asked how the neural representation of allocentric space is distorted by goal-directed behaviour. Participants navigated an agent to two successive goal locations in a grid world environment comprising four interlinked rooms, with a contextual cue indicating the conditional dependence of one goal location on another. Examining the neural geometry by which room and context were encoded in fMRI signals, we found that map-like representations of the environment emerged in both hippocampus and neocortex. Cognitive maps in hippocampus and orbitofrontal cortices were compressed so that locations cued as goals were coded together in neural state space, and these distortions predicted successful learning. This effect was captured by a computational model in which current and prospective locations are jointly encoded in a place code, providing a theory of how goals warp the neural representation of space in macroscopic neural signals.

show abstract

Section: Discussionmentioning

confidence: 99%

Goal-seeking compresses neural codes for space in the human hippocampus and orbitofrontal cortex

Muhle-Karbe

Sheahan

Pezzulo

et al. 2023

Preprint

View full text Add to dashboard Cite

show abstract

“…How can a neural or biological network efficiently encode multiple variables simultaneously 14,28 ? One solution is to encode variables in an abstract format so they can be reused in novel situations to facilitate generalization and compositionality 21,[29][30][31][32][33] . Here, we show that in the human brain, such a disentangled representation emerged as a function of learning to perform inference in our task.…”

Section: Discussionmentioning

confidence: 99%

Abstract representations emerge in human hippocampal neurons during inference behavior

Courellis,

Mixha,

Cardenas

et al. 2023

Preprint

View full text Add to dashboard Cite

Humans have the remarkable cognitive capacity to rapidly adapt to changing environments. Central to this capacity is the ability to form high-level, abstract representations that take advantage of regularities in the world to support generalization1,2. However, how these representations are encoded in populations of neurons, how they emerge through learning, and how they relate to behavior remains unknown3. Here we characterized the representational geometry of populations of neurons (single-units) recorded in the hippocampus, amygdala, medial frontal cortex, and ventral temporal cortex of neurosurgical patients who are performing an inferential reasoning task. We find that only the neural representations formed in the hippocampus simultaneously encode multiple task variables in an abstract, or disentangled, format. This representational geometry is uniquely observed after patients learn to perform inference, and consisted of disentangled directly observable and discovered latent task variables. Interestingly, learning to perform inference by trial and error or through verbal instructions led to the formation of hippocampal representations with similar geometric properties. The observed relation between representational format and inference behavior suggests that abstract/disentangled representational geometries are important for complex cognition.

show abstract

“…the ability to conceptualize prior experience in terms of components that can be re-configured in a novel situation [167,168]. More broadly, compositional generalization has long been understood to be a crucial component for human-like learning and generalization [163], in large part due to the diversity and breadth of its applications: compositional generalization often involves the composition of many rules, relations, or attributes [163,[169][170][171]. The comparative simplicity of TI enabled us to identify how minimally structured learning systems can implement the inductive biases needed for this task.…”

Section: Discussionmentioning

confidence: 99%

A mathematical theory of relational generalization in transitive inference

Lippl,

Kay,

Jensen

et al. 2023

Preprint

View full text Add to dashboard Cite

Humans and animals routinely infer relations between different items or events and generalize these relations to novel combinations of items ("compositional generalization"). This allows them to respond appropriately to radically novel circumstances and is fundamental to advanced cognition. However, how learning systems (including the brain) can implement the necessary inductive biases has been unclear. Here we investigated transitive inference (TI), a classic relational task paradigm in which subjects must learn a relation (A>B and B>C) and generalize it to new combinations of items (A>C). Through mathematical analysis, we found that a broad range of biologically relevant learning models (e.g. gradient flow or ridge regression) perform TI successfully and recapitulate signature behavioral patterns long observed in living subjects. First, we found that models with item-wise additive representations automatically encode transitive relations. Second, for more general representations, a single scalar "conjunctivity factor" determines model behavior on TI and, further, the principle of norm minimization (a standard statistical inductive bias) enables models with fixed, partly conjunctive representations to generalize transitively. Finally, neural networks in the "rich regime," which enables representation learning and often leads to better generalization, deviate in task behavior from living subjects and can make generalization errors. Our findings show systematically how minimal statistical learning principles can explain the rich behaviors empirically observed in TI in living subjects, uncover the mechanistic basis of transitive generalization in standard learning models, and lay out a formally tractable approach to understanding the neural basis of relational generalization.

show abstract

Compositional generalization through abstract representations in human and artificial neural networks

Cited by 7 publications

References 30 publications

Goal-seeking compresses neural codes for space in the human hippocampus and orbitofrontal cortex

Goal-seeking compresses neural codes for space in the human hippocampus and orbitofrontal cortex

Abstract representations emerge in human hippocampal neurons during inference behavior

A mathematical theory of relational generalization in transitive inference

Contact Info

Product

Resources

About