Anna Onstad-Hargrave

Anna Onstad-Hargrave currently holds the position of Talent Operations Specialist and is pursuing her PhD at Stanford University, where she is co-advised by Stefano Ermon and Chris Re. Her research primarily focuses on the efficient training of machine learning systems and the integration of long-range context into these systems. Anna’s work demonstrates a belief in multiple approaches to achieving high-performing language models, highlighting that sparsity can be both hardware-friendly and maintain quality. She anticipates a future shift in the field towards more efficient inference processes. Anna is also interested in exploring when and why attention mechanisms are necessary and investigating potential alternatives. Her academic influences include references to notable works like ‘A Kernel Theory of Modern Data Augmentation’ and ‘FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness’, as well as collaborations with peers such as Dan Fu, Albert Gu, and Phil Wang. Notably, she has also interacted with industry figures like Young-Jun Ko from Inflection and referenced benchmarks like MLPerf in her studies. Anna is set to join Princeton University as an assistant professor next year, continuing her impactful work in machine learning and systems.