Outcomes record · Publications

Publications

Papers and posters, written up as the work matures. Negative results included.

2026PendingPaper

Subspace dictionaries across depth: comparing SASA and sparse autoencoders in Gemma 3 1B

Submitted to a NeurIPS 2026 workshop
Daniel Helo Puccini, et al.

Does a model encode why it is uncertain, not just how much? Subspace dictionaries vs. SAEs across all 26 layers of Gemma 3 1B. Both carry uncertainty signal; the split by cause is a dataset confound.

Read the write-up →
2026-04PresentedPoster

From commands to invitations: the pragmatic softening of child-directed speech, 1960s–2000s

UURAF 2026 · Poster
Daniel Helo Puccini; MSU Language Acquisition Lab

2.47 million parent tokens from CHILDES, read as an accidental historical corpus. Commands fall 41%, softeners rise 76%, and the structural features that should not move do not.

See the poster and write-up →
2024-07PublishedPaper

Impact of network topology on Byzantine resilience in decentralized federated learning

arXiv:2407.05141
Siddhartha Bhattacharya, Daniel Helo Puccini, Josh Siegel

Byzantine-robust aggregators, tested on sparse peer-to-peer graphs instead of the fully connected ones they assume. They break.

Read the write-up →
2024-04AwardPoster

Stanza Extravaganza: improving Latin American Spanish morphological classifiers

UURAF 2024 · Best in Linguistics
Daniel Helo Puccini, Kiara Gonzalez; advised by Cristina Schmitt

Stanford's Stanza mis-tags Spanish verbs that carry clitic pronouns. A clitic-aware preprocessing pass fixes the 18.6% it missed, on child-acquisition corpora from Paraguay and Argentina.

See the poster and write-up →