Article IV · Section 3

Research

Every active project has a target venue and a submission date.

ProjectTarget venueDeadlineStatus
Subspace dictionaries across depth: SASA vs. sparse autoencoders in Gemma 3 1B
Alignment & interpretability · Helo Puccini et al. · 26-layer sweep, 3 seeds
NeurIPS 2026 · Workshop2026Pending
Horus: navigating and steering SAE feature space
Alignment & interpretability · Gemma-2-2B · Neuronpedia
Open-source release2026Building
Impact of network topology on Byzantine resilience in decentralized federated learning
ML for science · Bhattacharya, Helo Puccini et al.
arXiv:2407.051412024-07Published
Spanish morphological tagging with data augmentation (Stanza)
Language models · +30% accuracy, Latin American Spanish · UURAF best in Linguistics
MSU Language Acquisition Lab2024Award
From commands to invitations: pragmatic softening of child-directed speech
Language models · Helo Puccini · 2.47M parent tokens, CHILDES · MSU Language Acquisition Lab
UURAF 20262026-04Presented

Placements and awards

Where Members go

Internships, REUs, national labs, industry research.

2026-05
Software engineering intern, Microsoft
Seattle · Global Intern Hackathon, 1st place Best Hack (Healthy Futures)
2025-01
Software development engineer co-op, Amazon Robotics
Boston · Spring co-op
2024-04
UURAF, 2× best research (Linguistics; Education)
Michigan State University · Award
Article IV · Section 8

Research standards

No external submission

without internal review by at least one Member not on the project.

Authorship follows contribution.

Disputes go to the Director.

Code and data

for any submitted work are kept in an Institute-controlled repository.

Responsible disclosure.

Hazardous capability findings go to the faculty advisor before publication.