Jacob Mitchell Springer

Education

Publications

^* In submission.

1. J. Springer, M. Advani, L. Aichberger, A. Bradley, E. Malach, O. Saremi, S. Williamson, P. Nakkiran, E. Littwin, A. Raghunathan. Annotations mitigate post-training mode collapse. ICML, 2026.
2. I. Watts, C. Li, S. Goyal, J. Springer, A. Raghunathan. Sharpness-aware pretraining mitigates catastrophic forgetting. ICML, 2026. Oral, ICBINB Workshop @ ICLR 2026
3. L. Feng, G. R. Ghosal, J. Springer, Z. Zhong, A. Raghunathan. Mix early, forget less: data mixing during pretraining builds resistance to forgetting. In submission, 2025^*.
4. A. Kulkarni, J. Springer, A. Subramonian, S. Swayamdipta. Disentangling geometry, performance, and training in language models. ICML, 2026. Spotlight
5. J. Springer, S. Goyal, K. Wen, T. Kumar, X. Yue, S. Malladi, G. Neubig, A. Raghunathan. Overtrained language models are harder to fine-tune. ICML, 2025. Outstanding Paper, SCOPE Workshop @ ICLR 2025; Entropic Paper Award, ICBINB Workshop @ ICLR 2025
6. J. Springer, S. Kotha, D. Fried, G. Neubig, A. Raghunathan. Repetition improves language model embeddings. ICLR, 2025.
7. T. Kim, J. Springer, A. Raghunathan, M. Sap. Mitigating bias in RAG: controlling the embedder. ACL Findings, 2025.
8. J. Springer, V. Adlakha, S. Reddy, A. Raghunathan, M. Mosbach. Understanding the influence of synthetic data for text embedders. ACL Findings, 2025.
9. J. Springer, V. Nagarajan, A. Raghunathan. Sharpness-aware minimization enhances feature quality via balanced learning. ICLR, 2024.
10. S. Kotha, J. Springer, A. Raghunathan. Understanding catastrophic forgetting in language models via implicit inference. ICLR, 2024.
11. H. T. Jones, J. Springer, G. T. Kenyon, J. Moore. If you've trained one you've trained them all: inter-architecture similarity increases with robustness. UAI, 2022. Oral presentation
12. J. Springer, M. Mitchell, G. T. Kenyon. A little robustness goes a long way: leveraging robust features for targeted transfer attacks. NeurIPS, 2021.
13. J. Springer, B. M. Reinstadler, U.-M. O'Reilly. STRATA: simple, gradient-free attacks for models of code. Workshop on Adversarial Learning Methods @ KDD, 2021.
14. J. Springer, G. T. Kenyon. It's hard for neural networks to learn the Game of Life. IJCNN, 2021.
15. J. Springer, M. Mitchell, G. T. Kenyon. Adversarial perturbations are not so weird: entanglement of robust and non-robust features in neural network classifiers. Preprint, 2021.
16. D. A. Wang, C. M. S. Strauss, J. Springer, A. Thresher, H. Pritchard, G. T. Kenyon. Sparse MP4. IEEE SSIAI, 2020.
17. J. Springer, C. S. Strauss, A. M. Thresher, E. Kim, G. T. Kenyon. Classifiers based on deep sparse coding architectures are robust to deep learning transferable examples. Preprint, 2018.
18. J. Springer, W. Feng. Teaching with angr: a symbolic execution curriculum and CTF. USENIX Workshop on Advances in Security Education, 2018.

Research Experience

Invited Talks

• Don't (Just) Minimize Your Pretraining Loss. ML Foundations Seminar, Apr 2026.
• Overtrained Language Models Are Harder to Fine-Tune. Waymo, Aug 2025.
• Echo Embeddings & Overtrained Language Models Are Harder to Fine-Tune. Summer of Data Seminar, DatologyAI, Jun 2025.
• Overtrained Language Models Are Harder to Fine-Tune. Translate Reading Group, Google, Jun 2025.
• Overtrained Language Models Are Harder to Fine-Tune. FLAME Center Seminar, Carnegie Mellon University, Apr 2025.
• Repetition Improves Language Model Embeddings. Foundation and Language Model (FLAME) Seminar, Carnegie Mellon University, Mar 2024.
• What Can Adversarial Examples Tell Us About Similarities Between Neural Networks? Pacific Northwest Seminar on Topology, Algebra, and Geometry in Data Science, University of Washington, Feb 2023.

Awards