Graphical Design of Interpretable Architectures

Source

• Source name: arXiv
• Source kind: paper
• Authors / organization: Pietro Barbiero
• Venue: arXiv: cs.LG
• Published: 2026-08-19
• Collected: 2026-08-20
• URL: http://arxiv.org/abs/2608.18936v1
• PDF: https://arxiv.org/pdf/2608.18936v1
• DOI: UNKNOWN
• Registry group: academic_apis_and_preprints
• Evidence role: canonical_academic_index
• Language: en
• Country / region: UNKNOWN

Plain-Language Brief

The text discusses a new graphical notation for designing interpretable AI architectures, which addresses the limitations of existing representations. The proposed notation provides a global view of an architecture and maps directly onto PyTorch einsum code.

What the Document Says

• The text introduces a new graphical notation for designing interpretable AI architectures.
• Existing representations fall short in either not providing a global view or not describing actual tensor manipulations.
• The proposed notation gives a global view of an architecture and maps one to one onto PyTorch einsum code.
• The notation is used to describe several interpretable architectures by construction, including concept bottlenecks, sparse probes, prototype networks, neural additive models, mixtures of linear models, and Steerling-8B.

Key Concepts Mentioned

• graphical notation
• interpretable AI architectures
• tensor manipulations
• PyTorch einsum code

Claims Made by the Document

• Existing representations fall short in providing a global view.
• Probabilistic graphical models and flowcharts do not describe actual tensor manipulations.
• The proposed notation provides a global view of an architecture.

Evidence Used

• The text states that existing representations fall short in one of two ways.

Methods / Approach

• Introduction of a new graphical notation for designing interpretable AI architectures

Limitations Stated by the Authors

UNKNOWN

Limitations Noted by the Agent

Agent interpretation — not from the source.

It is not stated whether the proposed notation has any limitations according to the author.

Open Questions

• What are the specific benefits of using this new graphical notation over existing ones?

Notes for Future Q&A

May include interpretation. Labeled where it does.

The text provides insights into the advantages of the new graphical notation but does not directly compare it with existing ones.

Raw Abstract / Excerpt

Designing, implementing, and comparing interpretable architectures requires a formal language to represent them. The most common representations fall short in one of two ways. Symbolic equations give no global view of an architecture at a glance. Probabilistic graphical models and flowcharts do not describe actual tensor manipulations, thus hiding key insights and limiting reproducibility. To close this gap, we introduce a graphical notation for designing interpretable AI architectures, adapted from Penrose tensor notation. This graphical notation gives a global view of an architecture and maps one to one onto PyTorch einsum code. We first use this notation to describe architectures that are interpretable by construction, including concept bottlenecks, sparse probes, prototype networks, neural additive models, and mixtures of linear models. We then diagram the key architectural components of Steerling-8B, a frontier interpretable language model. The diagram yields global insights into the architecture (e.g., showing that Steerling is a residual model), a geometric interpretation of each individual operation, and a direct translation into 33 lines of PyTorch code.