peterdresslar/attn-signal-lab
About Attention Signal Lab
This is an experimental project companion for Peter Dresslar's Master's Degree Applied Project, Complex Systems Science, Arizona State University, Spring 2026. Advisor: Professor Bryan Daniels.
Attention Signal Lab is an application toolbench for exploring attention-based language models, especially the geometry of multi-layer hidden states, attention-derived signals, prompt batteries, and intervention response surfaces.
This application is intended to provide an interactive atlas and exploratory interface for current findings. The present version is an internal research preview.
Current Browser
The atlas browser visualizes hidden-state coordinate maps from baseline model runs. Each point is one prompt. Coordinates are derived from PCA over selected hidden-state slices, including last-token states, mean-pooled states, embedding-layer states, final-layer states, and all-layer concatenations.
The current app shell is powered by the single maintained manifest at public/available-data-manifest.json and wraps the existing fast standalone Plotly pages under public/pca-browser/. The next data migration will remove legacy signal-lab intervention fields from public bundles and switch public prompt exports to generated-only battery-6o.
References
- Peter Dresslar. Effective Geometry: Exploring an Inference-Time Layer Intervention on Hybrid Linear Attention Models. Work in progress, 2026.
- Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Lukasz Kaiser, and Illia Polosukhin. "Attention Is All You Need." 2017.
- Nelson Elhage, Neel Nanda, Catherine Olsson, Tom Henighan, Nicholas Joseph, Ben Mann, Amanda Askell, Yuntao Bai, Anna Chen, Tom Conerly, Nova DasSarma, Dawn Drain, Deep Ganguli, Zac Hatfield-Dodds, Danny Hernandez, Andy Jones, Jackson Kernion, Liane Lovitt, Kamal Ndousse, Dario Amodei, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish, and Chris Olah. "A Mathematical Framework for Transformer Circuits." 2021.
- Guillaume Alain and Yoshua Bengio. "Understanding Intermediate Layers Using Linear Classifier Probes." 2016.
- Kevin Meng, David Bau, Alex Andonian, and Yonatan Belinkov. "Locating and Editing Factual Associations in GPT." 2022.
- Kevin Wang, Alexandre Variengien, Arthur Conmy, Buck Shlegeris, and Jacob Steinhardt. "Interpretability in the Wild: A Circuit for Indirect Object Identification in GPT-2 Small." 2022.