finding

active

finding:diverse-computer-vision-models-trained-on-visual-recognition-tasks-converge-to-remarkably-similar-internal-feature-representations-regardless-of-architecture-training-procedure-or-implementation-details-closely-matching-the-organization-of-animal-visual-cortex

Diverse computer vision models trained on visual recognition tasks converge to remarkably similar internal feature representations regardless of architecture, training procedure, or implementation details, closely matching the organization of animal visual cortex

Empirical evidence for the universality hypothesis cited as supporting the possibility of convergent consciousness-like solutions

Source paper

extracted_from

cimcWhitepaper

Neighborhood — ranked by edge-count

Hypotheses (1)

hypothesis

Different learning systems facing similar computational problems will converge to similar consciousness-like solutions, including potentially biological and artificial systems
associated_withsupports
Extension of the Universality Hypothesis to consciousness: if consciousness solves a well-defined computational problem, different systems will discover it independently

Concepts (1)

concept

Universality Hypothesis
supports
The hypothesis that analogous features and circuits reliably form across different neural network models and tasks

Related by similarity (8)

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

Olah et al. (2020) found that automatically trained computer vision models, regardless of architecture and training procedure, all arrive at similar functional structures organizing similar features into similar compositional hierarchies, closely resembling the primate visual cortex.finding0.876
Empirical finding supporting the Universality Hypothesis; extended by the paper to consciousness
Neural networks, trained with different objectives on different data and modalities, are converging to a shared statistical model of reality in their representation spaces.quote0.816
The paper's central thesis statement, presented prominently after the abstract
Training identical architectures on the same data with different objective functions should produce systematically different internal evaluative representations, detectable through interpretability tools, even when final task performance is matchedhypothesis0.808
Second falsifiable prediction linking objective function structure to valence profile
Different neural network models trained on different objectives and modalities are converging to a shared statistical model of reality in their representation spaceshypothesis0.808
The central hypothesis of the paper; the platonic representation hypothesis itself
Performance-optimized hierarchical models predict neural responses in higher visual cortex (Yamins et al., 2014)concept0.802
Demonstrated CNN representations predict neurons in visual cortex; background motivation for neural-network-brain correspondence.
There is a growing similarity in how datapoints are represented in different neural network models, spanning different architectures, training objectives, and data modalitiesclaim0.799
Primary empirical claim of the paper
How do representations differ or converge between architectures, tasks, and modalities?question0.796
Broader research question MAS is positioned to address, citing multiple recent works.
Different models cannot converge to the same representation if they have access to fundamentally different information; convergence is capped by mutual information between input signalsclaim0.795
Key limitation of the PRH for non-bijective observations