Gül Sena Altıntaş — accepted AI-conference papers

2026

TokSuite: Measuring the Impact of Tokenizer Choice on Language Model Behavior

Gül Sena Altıntaş, Malikeh Ehghaghi, Brian Lester, Fengyuan Liu, Wanru Zhao, Marco Ciccone +1

ICML 2026oral

Tokenizers provide the fundamental basis through which text is represented and processed by language models (LMs). Despite the importance of tokenization, its role in LM performance and behavior is poorly understood due to the challenge of measuring the impact of tokenization in isolation. To addres…

Cited by 0 Source ↗Scholar ↗

2025

The Butterfly Effect: Neural Network Training Trajectories Are Highly Sensitive to Initial Conditions

Gül Sena Altıntaş, Devin Kwok, Colin Raffel, David Rolnick

ICML 2025poster

Neural network training is inherently sensitive to initialization and the randomness induced by stochastic gradient descent. However, it is unclear to what extent such effects lead to meaningfully different networks, either in terms of the models' weights or the underlying functions that were learne…

Cited by 0 Source ↗PDF ↗Scholar ↗